1 2011-01-25 00:06:33 <Diablo-D3> [1/24/11 8:05:58 PM] ERROR: Invalid block found on ATI RV770 (#1), possible driver or hardware issue
  2 2011-01-25 00:06:35 <Diablo-D3> an improvement!
  3 2011-01-25 00:07:08 <tcatm> still trying to understand the kernel code?
  4 2011-01-25 00:07:35 <Diablo-D3> not understand as much as make work
  5 2011-01-25 00:08:03 <Diablo-D3> fW* is obviously an optimization
  6 2011-01-25 00:08:39 <tcatm> not really
  7 2011-01-25 00:08:56 <tcatm> he just moved code outside the loop
  8 2011-01-25 00:09:12 <Diablo-D3> yes, which IS an optimization
  9 2011-01-25 00:09:17 <Diablo-D3> since you couldnt do this without the loop
 10 2011-01-25 00:09:52 <Diablo-D3> http://pastebin.com/LFkwNNhw
 11 2011-01-25 00:09:56 <Diablo-D3> but obviously thats not right.
 12 2011-01-25 00:10:25 <tcatm> why don't you use the kernel as is?
 13 2011-01-25 00:10:38 <Diablo-D3> because its wrong
 14 2011-01-25 00:10:48 <tcatm> huh? working fine here
 15 2011-01-25 00:11:06 <Diablo-D3> it doesnt match the args m0's and mine use
 16 2011-01-25 00:11:09 <Diablo-D3> so its wrong
 17 2011-01-25 00:11:15 <tcatm> fix your code
 18 2011-01-25 00:11:39 <tcatm> or even fix the args, doesn't really matter
 19 2011-01-25 00:11:47 <Diablo-D3> fix my code? that'd involve fixing bitcoin and all other miners too
 20 2011-01-25 00:11:55 <luke-jr> lol
 21 2011-01-25 00:12:41 <tcatm> just change ctx->* to your args and it should work
 22 2011-01-25 00:12:55 <Diablo-D3> look at the pastebin
 23 2011-01-25 00:13:22 <tcatm> great
 24 2011-01-25 00:13:29 <m0mchil> tcatm, thank you for the patch BTW
 25 2011-01-25 00:13:42 <Diablo-D3> tcatm: but that doesnt look right
 26 2011-01-25 00:13:42 <tcatm> m0mchil: the fake-push thing?
 27 2011-01-25 00:13:48 <m0mchil> yup
 28 2011-01-25 00:13:52 <tcatm> does it work reliable?
 29 2011-01-25 00:14:21 <andrew12> "does it work reliable"
 30 2011-01-25 00:14:29 <m0mchil> hmhmh, didn't tested it much
 31 2011-01-25 00:14:35 <Diablo-D3> reliably damnit
 32 2011-01-25 00:15:22 <tcatm> it doesn't reconnect to the node when the connection dies so it might end up only sending getworks when it found a solution (= worse than before)
 33 2011-01-25 00:16:36 <m0mchil> am I getting it right that it waits for a new block to be broadcast to request new work from bitcoind / pool?
 34 2011-01-25 00:16:49 <tcatm> yep, block or tx
 35 2011-01-25 00:17:18 <andrew12> hm
 36 2011-01-25 00:17:49 <m0mchil> perhaps block or 60 seconds would be better, not sure
 37 2011-01-25 00:17:52 <andrew12> what was the usd/btc rate in october 2010?
 38 2011-01-25 00:17:58 <Diablo-D3> hey m0mchil
 39 2011-01-25 00:18:04 <Diablo-D3> whats the first three ints of the second block of the header?
 40 2011-01-25 00:18:06 <tcatm> so if you fix the reconnect thing it should reduce getwork/s a lot
 41 2011-01-25 00:18:28 <tcatm> Diablo-D3: merke, ntime, nnonce?
 42 2011-01-25 00:18:37 <tcatm> merkle*
 43 2011-01-25 00:18:45 <Diablo-D3> its not merkle, ntime, and nbits?
 44 2011-01-25 00:18:51 <tcatm> ah yes
 45 2011-01-25 00:18:55 <tcatm> i missed nbits
 46 2011-01-25 00:18:58 <m0mchil> Diablo-D3, yes, end of merkle, time, difficulty
 47 2011-01-25 00:19:06 <Diablo-D3> damnit, so Im doing this right
 48 2011-01-25 00:19:17 <tcatm> Diablo-D3: can you paste your complete kernel?
 49 2011-01-25 00:19:30 <Diablo-D3> you've seen his, right?
 50 2011-01-25 00:19:34 <tcatm> yep
 51 2011-01-25 00:19:35 <m0mchil> tcatm: I am reluctant to distribute python bitcoin node with miner... for now
 52 2011-01-25 00:19:48 <echelon> can someone add iy6ni3wkqazp4ytu.onion and bitcoinbudtoeks7.onion to the fallback nodes?
 53 2011-01-25 00:20:30 <tcatm> echelon: which fallback nodes? those hardcoded in the source?
 54 2011-01-25 00:20:47 <echelon> on the wiki.. http://www.bitcoin.org/wiki/doku.php?id=fallback_nodes
 55 2011-01-25 00:20:49 <Diablo-D3> http://pastebin.com/iJByrL4e
 56 2011-01-25 00:20:58 <Diablo-D3> tcatm: the rest is the same, I commented out the loop to make debugging easier
 57 2011-01-25 00:21:02 <m0mchil> tcatm, also, 5870 would need new hash space every ~12 seconds
 58 2011-01-25 00:21:02 <tcatm> echelon: there's a new wiki at en.bitcoin.it
 59 2011-01-25 00:21:11 <echelon> oh
 60 2011-01-25 00:21:54 <echelon> why is it on a different domain
 61 2011-01-25 00:22:10 <andrew12> because its a different server
 62 2011-01-25 00:22:42 <tcatm> m0mchil: well you could detect if it's close to nonce overrun and issue a getwork
 63 2011-01-25 00:23:04 <tcatm> or just keep askrate at 5s and additionally getwork on new blocks/tx
 64 2011-01-25 00:23:17 <tcatm> or even better: switch to push
 65 2011-01-25 00:23:46 <m0mchil> yes, and there are many slower GPUs... but yes, push it is
 66 2011-01-25 00:23:51 <tcatm> Diablo-D3: what did you do to W3/tnonce?
 67 2011-01-25 00:24:08 <m0mchil> I now deeply regret my initial getwork...
 68 2011-01-25 00:24:32 <tcatm> getwork was just logical as it was very easy to add to the RPC interface
 69 2011-01-25 00:24:44 <tcatm> push requires a multithreaded TCP server
 70 2011-01-25 00:25:14 <tcatm> did you start any work on a push interface to bitcoind?
 71 2011-01-25 00:26:04 <m0mchil> no
 72 2011-01-25 00:26:24 <molecular> didn't slush start work on implementing push on his pool using a proxy?
 73 2011-01-25 00:26:37 <Diablo-D3> tcatm: turned loop off
 74 2011-01-25 00:26:47 <tcatm> I'm developing a binary push-based interface (actually porting my push based approach)
 75 2011-01-25 00:27:41 <m0mchil> I am not sure if push should get in official bitcoind... in custom setups it's ok
 76 2011-01-25 00:28:04 <luke-jr> s/official/original
 77 2011-01-25 00:28:24 <tcatm> That's why I made it as a module that can be patched in.
 78 2011-01-25 00:28:30 <m0mchil> great
 79 2011-01-25 00:28:49 <Diablo-D3> tcatm: so Im not sure where it went wrong
 80 2011-01-25 00:29:16 <tcatm> https://github.com/tcatm/bitcoin/tree/FMI early dev state (just a skeleton)
 81 2011-01-25 00:30:03 <m0mchil> and... on a side note... we need more exchanges... moar!
 82 2011-01-25 00:30:16 <m0mchil> my nightmare is mtgox just dissapears
 83 2011-01-25 00:30:16 <tcatm> Diablo-D3: you know, that the code only searches for H==0?
 84 2011-01-25 00:30:24 <tcatm> m0mchil: bitcoin central is opensource
 85 2011-01-25 00:30:27 <Diablo-D3> tcatm: yeah, so does mine
 86 2011-01-25 00:30:42 <tcatm> so what's wrong?
 87 2011-01-25 00:30:54 <Diablo-D3> tcatm: its returning bullshit results
 88 2011-01-25 00:31:01 <tcatm> wait, you're searching for H==0 without the loop?
 89 2011-01-25 00:31:39 <Diablo-D3> yes, the if(H + shit) line just does output[0] = tnonce; like it does in my kernel
 90 2011-01-25 00:32:01 <m0mchil> bitcoin central is nice... but not convenient as mtgox. And mtgox can be much more convenient tha now
 91 2011-01-25 00:32:32 <tcatm> Diablo-D3: what's output[0]? A global variable for all kernels?
 92 2011-01-25 00:33:30 <tcatm> Where's your code?
 93 2011-01-25 00:34:20 <Diablo-D3> https://github.com/Diablo-D3/DiabloMiner
 94 2011-01-25 00:37:21 <tcatm> strange code
 95 2011-01-25 00:37:30 <Diablo-D3> strange in what way?
 96 2011-01-25 00:37:34 <Diablo-D3> its almost identical to m0's
 97 2011-01-25 00:38:21 <tcatm> around line 448 if(buffer.getInt(0)...
 98 2011-01-25 00:38:38 <tcatm> what's that block supposed to do?
 99 2011-01-25 00:39:01 <Diablo-D3> its a bytebuffer, it gets an int.
100 2011-01-25 00:39:18 <tcatm> the whole if-block?
101 2011-01-25 00:40:07 <Diablo-D3> that line isnt near 448
102 2011-01-25 00:40:55 <Diablo-D3> oh wait yes it is, 444
103 2011-01-25 00:41:01 <Diablo-D3> if(buffer.getInt(0) > 0) {
104 2011-01-25 00:41:03 <Diablo-D3> if the returned int is above 0
105 2011-01-25 00:41:06 <tcatm> oh, github messes up line numbers again, sorry
106 2011-01-25 00:41:09 <tcatm> yes, but what's the logic?
107 2011-01-25 00:41:29 <tcatm> is it: if kernel returns nonce != 0 calculate the hash again, this time including G, and check result, then reset buffer to 0?
108 2011-01-25 00:41:38 <Diablo-D3> yes
109 2011-01-25 00:41:47 <Diablo-D3> the miner fully calculates the hash to catch gpu bullshit
110 2011-01-25 00:43:41 <tcatm> What about every miner thread (in GPU) return to output[myid] instead of output[0] and also only return a flag (0 or !0) instead of nonce, then loop (in java) through that array re-calculating every nonce that has the flag set?
111 2011-01-25 00:44:06 <Diablo-D3> why would it?
112 2011-01-25 00:44:21 <Diablo-D3> the only way that'd make sense if I was likely to get more than one valid nonce per session
113 2011-01-25 00:44:36 <Diablo-D3> s/session/kernel/
114 2011-01-25 00:45:04 <tcatm> That's a lot more likely than solving the block at current difficulties
115 2011-01-25 00:45:13 <echelon> does "#" work as a comment in bitcoin.conf?
116 2011-01-25 00:45:19 <Diablo-D3> its not as likely as you think
117 2011-01-25 00:45:28 <Diablo-D3> when I last tested that, it had zero change on my rate of H==0
118 2011-01-25 00:45:47 <Diablo-D3> not only that, my rate of H==0 matches the estimated output.
119 2011-01-25 00:45:52 <tcatm> My miner find H==0 quite often, even multiple times per kernel run
120 2011-01-25 00:46:18 <Diablo-D3> anyhow, this doesnt solve my problem
121 2011-01-25 00:47:38 <Diablo-D3> tcatm: btw
122 2011-01-25 00:47:41 <Diablo-D3> setting a flag makes no sense
123 2011-01-25 00:47:46 <Diablo-D3> != 0 IS the flag
124 2011-01-25 00:47:56 <Diablo-D3> so it'd only change a single output to a few thousand outputs.
125 2011-01-25 00:48:04 <Diablo-D3> and art's kernel still suffers from the same problem
126 2011-01-25 00:48:18 <Diablo-D3> a loop of 1024 nonces could still, theoretically, have multiple valid nonces
127 2011-01-25 00:48:26 <tcatm> yep
128 2011-01-25 00:48:50 <tcatm> that's why he calculates all those 1024 hashes again to find any hash<=target
129 2011-01-25 00:49:18 <Diablo-D3> heh
130 2011-01-25 00:49:37 <tcatm> doing 1024 or even 4096 hashes in CPU costs almost no time, you can do that while waiting for getwork response
131 2011-01-25 00:49:43 <Diablo-D3> he only gave the kernel, not the miner source
132 2011-01-25 00:50:09 <tcatm> he described his miner pretty well over the last months
133 2011-01-25 00:50:17 <Diablo-D3> not as well as you think
134 2011-01-25 00:50:20 <Diablo-D3> anyhow, this doesnt solve my problem
135 2011-01-25 00:51:13 <tcatm> hm hm
136 2011-01-25 00:51:35 <molecular> I might be talking crap: but what about the 3 rounds at the beginning and at the end. aren't these included in art's loop?
137 2011-01-25 00:51:48 <Diablo-D3> molecular: dont think so
138 2011-01-25 00:51:52 <echelon> what's the timeout for each connection on tor?
139 2011-01-25 00:51:54 <Diablo-D3> at least I know the last three arent
140 2011-01-25 00:51:54 <echelon> err
141 2011-01-25 00:51:56 <echelon> on bitcoin
142 2011-01-25 00:52:05 <comboy> Diablo-D3: thanks for your miner, great work, very stable and my cpu stays cool
143 2011-01-25 00:53:02 <Diablo-D3> molecular: notice it stops at W12 and then does if(H + shit)... thats the hallmark of one that doesnt do the last 3
144 2011-01-25 00:53:08 <Diablo-D3> comboy: welcome
145 2011-01-25 00:53:32 <molecular> Diablo-D3, ok
146 2011-01-25 00:53:36 <tcatm> Diablo-D3: http://pastebin.com/Bghjj3Lr this code calculates the ctx (here called blk)
147 2011-01-25 00:53:54 <Diablo-D3> tcatm: hah
148 2011-01-25 00:54:13 <Diablo-D3> tcatm: its identical to my code
149 2011-01-25 00:54:23 <comboy> anybody know if and can I find somewhere history of difficulty over time?
150 2011-01-25 00:54:27 <tcatm> Diablo-D3: great
151 2011-01-25 00:54:58 <Diablo-D3> tcatm: but he misses out on the obvious win
152 2011-01-25 00:55:04 <tcatm> comboy: http://nullvoid.org/bitcoin/difficultiez.php
153 2011-01-25 00:55:07 <Diablo-D3> you only need 6 of the 8 parts for cty
154 2011-01-25 00:55:25 <tcatm> saving what? 2 mov's?
155 2011-01-25 00:55:33 <Diablo-D3> because A and E dont change with only 2 rounds
156 2011-01-25 00:55:34 <comboy> sweeet, thanks! :)
157 2011-01-25 00:55:35 <Diablo-D3> er 3 rounds
158 2011-01-25 00:55:56 <tcatm> also that precalc_hash is from my miner
159 2011-01-25 00:56:16 <Diablo-D3> both me and m0 do it this way
160 2011-01-25 00:56:39 <tcatm> sure, the compiler optimizes that pretty good
161 2011-01-25 00:57:12 <tcatm> + 64 more bits as a constant don't impact performance at all as the DMA will probably transfer a larger chunk of memory anyway
162 2011-01-25 00:57:27 <Diablo-D3> took it out anyhow
163 2011-01-25 01:01:14 <Diablo-D3> hrrrrrrrm
164 2011-01-25 01:01:16 <Diablo-D3> tcatm: hey you
165 2011-01-25 01:01:22 <tcatm> ?
166 2011-01-25 01:01:25 <Diablo-D3> show me the source for R
167 2011-01-25 01:01:29 <tcatm> r?
168 2011-01-25 01:01:32 <Diablo-D3> the macro
169 2011-01-25 01:01:44 <tcatm> Just a SHA round
170 2011-01-25 01:01:53 <Diablo-D3> yeah, but does it match mine/
171 2011-01-25 01:02:14 <tcatm> maybe?
172 2011-01-25 01:02:29 <tcatm> http://pastebin.com/LEsKW5cB
173 2011-01-25 01:03:10 <Diablo-D3> rut row
174 2011-01-25 01:03:23 <Diablo-D3> those rotates seem to be the opposite direction of mine
175 2011-01-25 01:03:43 <tcatm> #define rotate(x,y) ((x<<y) | (x>>(sizeof(x)*8-y)))
176 2011-01-25 01:04:07 <Diablo-D3> static int rot(int x, int y) {
177 2011-01-25 01:04:08 <Diablo-D3> return (x >>> y) | (x << (32 - y));
178 2011-01-25 01:04:09 <Diablo-D3> faaawk
179 2011-01-25 01:04:22 <Diablo-D3> no wonder this shits broke
180 2011-01-25 01:05:26 <Diablo-D3> meh nm
181 2011-01-25 01:05:32 <Diablo-D3> it should end up producing the same answer
182 2011-01-25 01:06:17 <Diablo-D3> 32 - 6 = 26, 32 - 11 = 21, 32 - 25 = 7
183 2011-01-25 01:06:19 <tcatm> yep
184 2011-01-25 01:06:28 <Diablo-D3> so its not that
185 2011-01-25 01:06:29 <Diablo-D3> damnit
186 2011-01-25 01:06:29 <tcatm> I'm just rotating right
187 2011-01-25 01:06:59 <tcatm> what's x >>> y (triple >)?
188 2011-01-25 01:07:13 <Diablo-D3> C >>
189 2011-01-25 01:07:21 <tcatm> what's >> in java?
190 2011-01-25 01:07:30 <Diablo-D3> same thing, but it respects the sign
191 2011-01-25 01:07:43 <tcatm> strange language
192 2011-01-25 01:08:04 <Diablo-D3> well, Java has no unsigned variables
193 2011-01-25 01:08:09 <dirtyfilthy> yeh it sucks
194 2011-01-25 01:08:15 <Diablo-D3> C >> on signed will do the same thing as Java >>
195 2011-01-25 01:08:20 <Diablo-D3> it wont shift the sign bit
196 2011-01-25 01:08:29 <Diablo-D3> so you need >>> to do full shifts
197 2011-01-25 01:08:47 <Diablo-D3> its just one of those things
198 2011-01-25 01:09:19 <tcatm> can you paste your whole kernel.cl?
199 2011-01-25 01:10:04 <Diablo-D3> http://pastebin.com/CG1CbRGu
200 2011-01-25 01:11:01 <tcatm> hm hm
201 2011-01-25 01:11:06 <tcatm> this is funny
202 2011-01-25 01:11:47 <Diablo-D3> cant figure out why it fails?
203 2011-01-25 01:11:54 <Diablo-D3> it should be identical if his kernel works in the first place
204 2011-01-25 01:12:15 <tcatm> did you verify his kernel?
205 2011-01-25 01:13:08 <lfm> is it some byteswapping thing? he sed he doesnt but getwork does
206 2011-01-25 01:13:24 <tcatm> shouldn't matter in this case
207 2011-01-25 01:14:01 <luke-jr> should 500 W PSU be enough for two HD 5770?
208 2011-01-25 01:18:22 <Diablo-D3> doesnt matter, I byteswap to make mine work
209 2011-01-25 01:18:30 <Diablo-D3> tcatm: havent verified it
210 2011-01-25 01:18:33 <Diablo-D3> but I assume it works
211 2011-01-25 01:18:42 <lfm> luke ati recomends 600
212 2011-01-25 01:20:18 <luke-jr> lfm: for one, or for two?
213 2011-01-25 01:20:28 <lfm> for ywo
214 2011-01-25 01:20:33 <lfm> two
215 2011-01-25 01:20:49 <luke-jr> hmm
216 2011-01-25 01:20:52 <luke-jr> think I could resell one?
217 2011-01-25 01:23:06 <Keefe> luke-jr: how much 12V watts can your psu do?
218 2011-01-25 01:23:15 <luke-jr> ?
219 2011-01-25 01:23:20 <luke-jr> dunno?
220 2011-01-25 01:23:30 <Keefe> tell me the model # and i'll look it up
221 2011-01-25 01:23:36 <luke-jr> http://www.newegg.com/Product/Product.aspx?Item=N82E16817371007
222 2011-01-25 01:24:29 <luke-jr> auction ends in 5 min XD
223 2011-01-25 01:24:37 <Keefe> 408W 12V
224 2011-01-25 01:24:41 <Keefe> auction for what?
225 2011-01-25 01:24:48 <luke-jr> $155 for two 5770
226 2011-01-25 01:25:13 <Keefe> if you have a low-end system, 408W 12V should be plenty for two 5770's
227 2011-01-25 01:25:44 <luke-jr> I think it uses about 150 W right now
228 2011-01-25 01:25:53 <Keefe> my mining boxes have sempron 140 and 1gb ram and one drive, and use about 50W without graphics
229 2011-01-25 01:26:00 <luke-jr> (measured at AC level and calculated)
230 2011-01-25 01:26:08 <Keefe> still should be ok
231 2011-01-25 01:26:21 <Keefe> 5770 uses ~110W iirc
232 2011-01-25 01:26:23 <luke-jr> $155 sound good deal for that?
233 2011-01-25 01:26:26 <Keefe> yep
234 2011-01-25 01:26:38 <Keefe> but i bet the bidding goes much higher :)
235 2011-01-25 01:26:51 <luke-jr> well, what should I max?
236 2011-01-25 01:27:01 <Keefe> $200
237 2011-01-25 01:27:05 <luke-jr> looks like one 5770 goes $146 retail
238 2011-01-25 01:27:15 <Keefe> i'm selling mine for $120
239 2011-01-25 01:27:28 <Keefe> if you are really patient, you could probably get one for $100 on ebay
240 2011-01-25 01:27:30 <luke-jr> $195.01 + $5 shipping then? :P
241 2011-01-25 01:27:33 <luke-jr> hmm
242 2011-01-25 01:27:36 <gribble> (bc,calc <an alias, 1 argument>) -- Alias for "echo The average time to generate a block at $1 Khps, given current difficulty of [bc,diff], is [time elapsed [math calc 1/((2**224-1)/[bc,diff]*$1*1000/2**256)]]".
243 2011-01-25 01:27:36 <luke-jr> ;;bc,calc
244 2011-01-25 01:27:38 <lfm> i got one 5770 new for 105 on sale
245 2011-01-25 01:27:53 <luke-jr> but one for $100 is > 2 for $190
246 2011-01-25 01:27:56 <luke-jr> ;;bc,stats
247 2011-01-25 01:27:58 <gribble> Current Blocks: 104441 | Current Difficulty: 18437.64439217 | Next Difficulty At Block: 104831 | Next Difficulty In: 390 blocks | Next Difficulty In About: 2 days, 5 hours, 50 minutes, and 30 seconds | Next Difficulty Estimate: 21920.36910764
248 2011-01-25 01:28:44 <luke-jr> I bit 165.01
249 2011-01-25 01:29:12 <luke-jr> outbid :p
250 2011-01-25 01:30:57 <Keefe> auction over?
251 2011-01-25 01:31:34 <luke-jr> yeah
252 2011-01-25 01:31:40 <luke-jr> it went up past $200 total
253 2011-01-25 01:32:01 <Keefe> link?
254 2011-01-25 01:32:25 <Keefe> nvm, found it
255 2011-01-25 01:33:38 <Keefe> i can sell you mine for $130 incl shipping and can ship it tomorrow
256 2011-01-25 01:34:31 <Keefe> not a great price though. but i'm patient
257 2011-01-25 01:34:56 <Keefe> fwiw, it OCs to 960 at stock V just fine
258 2011-01-25 01:35:16 <Keefe> gigabyte super overclock edition
259 2011-01-25 01:36:26 <Keefe> i ran a 5970 and 5770 together on a psu with 450W 12V for a month before upgrading
260 2011-01-25 01:36:43 <Keefe> but my base system only pulls 50W
261 2011-01-25 01:36:45 <Diablo-D3> damnit
262 2011-01-25 01:36:49 <Diablo-D3> where the fuck is art when you need him,
263 2011-01-25 01:37:02 <Kiba> hey
264 2011-01-25 01:37:09 <Kiba> you want to draw art?
265 2011-01-25 01:37:22 <Diablo-D3> no, I want art to figure out why his kernel doesnt work
266 2011-01-25 01:37:30 <Keefe> Diablo-D3: can't you figure out that opencl stuff on your own?
267 2011-01-25 01:37:43 <luke-jr> Keefe: I'm not in a rush
268 2011-01-25 01:37:51 <luke-jr> I passed by a 5770 for $100 earlier
269 2011-01-25 01:38:00 <luke-jr> I figure 5770 is only worth $80 at most :p
270 2011-01-25 01:38:08 <Keefe> ok :)
271 2011-01-25 01:38:50 <tcatm> re
272 2011-01-25 01:38:57 <tcatm> Diablo-D3: does it work now?
273 2011-01-25 01:39:20 <Diablo-D3> tcatm: no?
274 2011-01-25 01:39:25 <Keefe> luke-jr: but keep in mind that if it takes you a week more to find the deal you want, you might as well have paid $20 and got it right away
275 2011-01-25 01:40:14 <Keefe> more like $28/wk actually
276 2011-01-25 01:40:30 <luke-jr> for a 5770?
277 2011-01-25 01:40:33 <Keefe> yep
278 2011-01-25 01:40:37 <tcatm> Diablo-D3: you made a mistake while copy&pasting the code
279 2011-01-25 01:40:50 <luke-jr> ;;bc,calc 156000
280 2011-01-25 01:40:51 <gribble> The average time to generate a block at 156000 Khps, given current difficulty of 18437.64439217 , is 5 days, 21 hours, and 22 seconds
281 2011-01-25 01:41:03 <gribble> The average time to generate a block at 177000 Khps, given current difficulty of 18437.64439217 , is 5 days, 4 hours, 16 minutes, and 35 seconds
282 2011-01-25 01:41:03 <Keefe> ;;bc,calc 177000
283 2011-01-25 01:41:10 <luke-jr> $17/wk :p
284 2011-01-25 01:41:47 <luke-jr> ;;bc,calc 240000
285 2011-01-25 01:41:48 <gribble> The average time to generate a block at 240000 Khps, given current difficulty of 18437.64439217 , is 3 days, 19 hours, 39 minutes, and 14 seconds
286 2011-01-25 01:42:26 <luke-jr> $181 max bid on single 5850 sound good?
287 2011-01-25 01:42:57 <Diablo-D3> tcatm: er, I did?
288 2011-01-25 01:43:21 <luke-jr> actually, for $184 I could BuyItNow
289 2011-01-25 01:43:33 <Keefe> luke-jr: $28/wk at $0.42/btc and 177 mhps (960mhz OC)
290 2011-01-25 01:43:37 <tcatm> Diablo-D3: yep
291 2011-01-25 01:43:49 <Diablo-D3> which line?
292 2011-01-25 01:43:51 <luke-jr> Keefe: I don't OC
293 2011-01-25 01:44:12 <luke-jr> and this is all assuming I manage to get a GPU miner working
294 2011-01-25 01:44:22 <tcatm> Diablo-D3: fW0 and the following few lines
295 2011-01-25 01:44:24 <Keefe> so $25/wk then
296 2011-01-25 01:44:27 <luke-jr> which isn't looking too good from my experience with the Mac
297 2011-01-25 01:44:45 <luke-jr> although nevermind& I just remembered, the Mac OS ver is too old for OpenCL XD
298 2011-01-25 01:45:58 <Keefe> $180 for 5850 sounds fair
299 2011-01-25 01:46:54 <luke-jr> but $184 too much? :P
300 2011-01-25 01:47:15 <Keefe> why do you ask me
301 2011-01-25 01:47:18 <hacim> whats the channel with live block stats?
302 2011-01-25 01:47:20 <Keefe> i prefer 5970's
303 2011-01-25 01:47:26 <luke-jr> hacim: #bitcoin-monitor
304 2011-01-25 01:47:31 <luke-jr> Keefe: sure, but that's $600
305 2011-01-25 01:47:37 <hacim> ah thanks
306 2011-01-25 01:47:44 <Keefe> i'll sell you a new 5970 for $460
307 2011-01-25 01:48:14 <Keefe> $478 incl sh
308 2011-01-25 01:48:29 <luke-jr> 5850 still a better deal :P
309 2011-01-25 01:48:33 <Keefe> (within usa)
310 2011-01-25 01:48:49 <hacim> if you spend a lot of time looking, you can get a 5970 for cheaper than that
311 2011-01-25 01:48:53 <tcatm> Diablo-D3: also there's another difference in your code and art's
312 2011-01-25 01:48:54 <molecular> tcatm you mean it should read
313 2011-01-25 01:49:01 <Keefe> time is money ;)
314 2011-01-25 01:49:03 <molecular> const uint fW0 = block0 + (rotr(block1, 7) ^ rotr(block1, 18) ^ (block1 >> 3));
315 2011-01-25 01:49:06 <molecular> const uint fW1 = block1 + (rotr(block2, 7) ^ rotr(block2, 18) ^ (block2 >> 3)) + 0x01100000;
316 2011-01-25 01:49:12 <luke-jr> 5850 = 1.33 MH/s/$ vs 5970 = 1.12 MH/s/$
317 2011-01-25 01:49:32 <Diablo-D3> argh
318 2011-01-25 01:49:56 <tcatm> better smack yourself
319 2011-01-25 01:49:59 <luke-jr> Keefe: $405 total or no sale! :P
320 2011-01-25 01:50:03 <Keefe> no deal
321 2011-01-25 01:50:04 <Kiba> Diablo-D3: you want to pay me to draw some art for ya?
322 2011-01-25 01:50:23 <noagendamarket> so who bought all the 5970's in the Us ?
323 2011-01-25 01:50:25 <noagendamarket> lol
324 2011-01-25 01:50:30 <Diablo-D3> noagendamarket: art did
325 2011-01-25 01:50:36 <noagendamarket> heh
326 2011-01-25 01:50:37 <Keefe> i bought 9
327 2011-01-25 01:50:39 <Diablo-D3> tcatm: whats the other difference?
328 2011-01-25 01:50:41 <luke-jr> Keefe: in reality, AMD says I'd need to buy a new PSU to get it anyway
329 2011-01-25 01:50:44 <Diablo-D3> art bought... what, 26?
330 2011-01-25 01:50:56 <noagendamarket> I can only find them in canada
331 2011-01-25 01:51:06 <noagendamarket> http://www.a-power.com/product-14264
332 2011-01-25 01:51:25 <noagendamarket> yeah art bought that many
333 2011-01-25 01:51:38 <Keefe> for a 5970, ya 408W 12V isn't enough when you're starting at 150W
334 2011-01-25 01:52:01 <noagendamarket> we need to source some for #bitcoind
335 2011-01-25 01:52:51 <tcatm> Diablo-D3: you don't want to know
336 2011-01-25 01:53:21 <Diablo-D3> theres none
337 2011-01-25 01:53:24 <Diablo-D3> kernel works now
338 2011-01-25 01:54:53 <Diablo-D3> okay, what is it then?
339 2011-01-25 01:54:59 <molecular> now what has to be done to put the loop in?
340 2011-01-25 01:55:18 <Diablo-D3> yes obviously the loop is turned off
341 2011-01-25 01:55:24 <molecular> I know
342 2011-01-25 01:55:28 <tcatm> turning the loop off is stupid
343 2011-01-25 01:55:37 <Diablo-D3> tcatm: its for debugging reasons
344 2011-01-25 01:56:32 <tcatm> so turn it on
345 2011-01-25 01:56:45 <molecular> any change to the java-code necessary when loop is on?
346 2011-01-25 01:57:05 <Diablo-D3> tcatm: what is art feeding the kernel for the nonce base?
347 2011-01-25 01:57:15 <Diablo-D3> is he skipping 1024 at a time in the host code?
348 2011-01-25 01:58:57 <tcatm> what would you say? RTFC?
349 2011-01-25 01:59:34 <Diablo-D3> RTSL
350 2011-01-25 01:59:46 <Diablo-D3> but neither your code nor arts is open source
351 2011-01-25 02:00:19 <tcatm> err.. the .cl contains all information?
352 2011-01-25 02:01:16 <Diablo-D3> tcatm: yes, but I'd still like to know, for sure, what its being fed
353 2011-01-25 02:02:07 <tcatm> well if you're lazy you could do tnonce = (ctx->nonce + myid)<<10
354 2011-01-25 02:02:50 <tcatm> then loop for(ctx->nonce = 0; ctx->nonce < ~0>>10; ctx->nonce++)
355 2011-01-25 02:03:07 <Diablo-D3> well, if I was REALLY lazy
356 2011-01-25 02:03:18 <Diablo-D3> ... yeah, it'd be that
357 2011-01-25 02:04:05 <Diablo-D3> (base + myid) << 10 seems to be right for my use case
358 2011-01-25 02:04:20 <tcatm> it avoids nasty byteswaps :P
359 2011-01-25 02:04:59 <tcatm> while you're at it, why not move f* to java and add args to the kernel?
360 2011-01-25 02:05:44 <Diablo-D3> tcatm: I might at some point
361 2011-01-25 02:07:29 <tcatm> also (E & F) | (G & (E & F)) might be optimized a little
362 2011-01-25 02:07:45 <Diablo-D3> probably
363 2011-01-25 02:08:05 <Diablo-D3> /AMD doesn't do BFI_INT via bitselect yet; when it does change (g^(e&(f^g))) to (bitselect(e, g, f))
364 2011-01-25 02:08:13 <tcatm> saving you 1..2 instructions per round
365 2011-01-25 02:08:15 <Diablo-D3> thats a note from my kernel
366 2011-01-25 02:08:25 <Diablo-D3> the REAL optimization is using the hardware for it
367 2011-01-25 02:08:32 <tcatm> even without bitselect you can cache half of the calculation
368 2011-01-25 02:08:46 <Diablo-D3> yeah, but jesus, fucking AMD
369 2011-01-25 02:09:57 <Diablo-D3> I'll have to limit the nonce to 2**22, btw
370 2011-01-25 02:10:19 <tcatm> why?
371 2011-01-25 02:10:26 <tcatm> wait, that's what I told you?
372 2011-01-25 02:10:31 <Diablo-D3> const uint tnonce = (base + myid) << 10;
373 2011-01-25 02:10:34 <Diablo-D3> thats why
374 2011-01-25 02:10:52 <Diablo-D3> since Im consuming nonces 2**10 times faster, the max isnt 2**32 anymore
375 2011-01-25 02:11:01 <tcatm> yep
376 2011-01-25 02:11:15 <tcatm> guess what's ~0>>10+1 is?
377 2011-01-25 02:11:32 <Diablo-D3> 0?
378 2011-01-25 02:12:00 <Diablo-D3> oh wait you said ~
379 2011-01-25 02:12:13 <tcatm> yep
380 2011-01-25 02:12:14 <Diablo-D3> its, what, 2**21?
381 2011-01-25 02:13:19 <tcatm> 2**22
382 2011-01-25 02:15:17 <Diablo-D3> heh
383 2011-01-25 02:15:18 <molecular> I have it running with a loop-count of 16 here. of course the hashmeter is wrong. but when I multiply with 16 it gives me 528mhash/s (compared to 535mhash/s before).
384 2011-01-25 02:15:39 <molecular> damn
385 2011-01-25 02:15:53 <Diablo-D3> what hardware?
386 2011-01-25 02:16:00 <tcatm> 5970
387 2011-01-25 02:16:01 <molecular> 1 5970 stock mhz
388 2011-01-25 02:16:08 <luke-jr> ;;bc,calc 53000
389 2011-01-25 02:16:09 <gribble> The average time to generate a block at 53000 Khps, given current difficulty of 18437.64439217 , is 2 weeks, 3 days, 7 hours, 2 minutes, and 13 seconds
390 2011-01-25 02:16:26 <Diablo-D3> lets see what it does on my 4850
391 2011-01-25 02:16:38 <tcatm> My 5970 do 556 Mhash/s stock
392 2011-01-25 02:16:49 <molecular> I think I saw some higher value before, when the loop was disabled.. will have to verify that, though
393 2011-01-25 02:17:03 <molecular> tcatm using which miner?
394 2011-01-25 02:17:10 <tcatm> m0's
395 2011-01-25 02:17:24 <Diablo-D3> yeah but you got some weird shit going on
396 2011-01-25 02:17:32 <molecular> weird, couldn't get that kind of performance with m0s
397 2011-01-25 02:17:38 <tcatm> Diablo-D3: not really
398 2011-01-25 02:17:49 <tcatm> you could easily port it to your miner
399 2011-01-25 02:19:11 <tcatm> Actually I'm currently using m0's kernel. unmodified
400 2011-01-25 02:19:20 <Diablo-D3> dont you have that mismatched stream sdk set?
401 2011-01-25 02:19:28 <tcatm> yep
402 2011-01-25 02:19:33 <tcatm> 2.3 compiler backported to 2.1
403 2011-01-25 02:20:02 <molecular> that could well make a difference
404 2011-01-25 02:20:28 <tcatm> where backported = cp clc and llc and some libs to 2.1 directory :P
405 2011-01-25 02:21:40 <molecular> what did you not copy back? I mean: what's wrong with using sdk 2.3 fully?
406 2011-01-25 02:21:58 <tcatm> the opencl lib is broken
407 2011-01-25 02:22:14 <Diablo-D3> molecular: basically, the actual compiler is in the driver not steam sdk
408 2011-01-25 02:22:44 <Diablo-D3> molecular: also, it may not matter with this kernel
409 2011-01-25 02:22:44 <tcatm> btw, you can patch llc to produce bfi instead of bytealign
410 2011-01-25 02:24:32 <Diablo-D3> yeah probably
411 2011-01-25 02:24:33 <molecular> just hung my system using 256 loops
412 2011-01-25 02:24:38 <Diablo-D3> molecular: not hung
413 2011-01-25 02:24:44 <Diablo-D3> loop is just going to take a long while
414 2011-01-25 02:24:50 <molecular> I know
415 2011-01-25 02:25:04 <molecular> it's been running for a while, finding blocks and reporting hashrate
416 2011-01-25 02:25:12 <molecular> then I hit ctrl-c and it froze
417 2011-01-25 02:25:21 <Diablo-D3> yeah it'll take awhile to come back
418 2011-01-25 02:25:22 <molecular> ping worked, no ssh, htough
419 2011-01-25 02:25:31 <molecular> ok, maybe I wasn't patient enough
420 2011-01-25 02:25:32 <Diablo-D3> it WILL come back
421 2011-01-25 02:25:43 <Diablo-D3> I once waited an hour for mine to come back after I had kernel time runaway
422 2011-01-25 02:26:22 <molecular> the hashmeter gets screwed badly when a block is found (show too high)
423 2011-01-25 02:26:51 <tcatm> solution: don't find blocks :P
424 2011-01-25 02:26:59 <luke-jr> LOL
425 2011-01-25 02:27:13 <Diablo-D3> thats not true at all
426 2011-01-25 02:27:19 <Diablo-D3> its largely variable anyhow
427 2011-01-25 02:27:26 <tcatm> variable?
428 2011-01-25 02:27:31 <Diablo-D3> it goes up and down
429 2011-01-25 02:27:39 <tcatm> hashrate is pretty much constant
430 2011-01-25 02:27:52 <Diablo-D3> not when you have other shit running on the machine
431 2011-01-25 02:28:13 <luke-jr> ok, I bid on a 5850
432 2011-01-25 02:29:18 <tcatm> Diablo-D3: can you paste your working kernel.cl?
433 2011-01-25 02:29:29 <molecular> should it be "const uint tnonce = (base + myid) << 10;" or "const uint tnonce = base + myid << 8;" ?
434 2011-01-25 02:29:37 <molecular> s/8/10
435 2011-01-25 02:29:46 <Diablo-D3> first one
436 2011-01-25 02:29:51 <tcatm> molecular: when you don't do 1024 nonces at one, yep
437 2011-01-25 02:31:33 <molecular> btw: Diablo, you could change "debug("Block found, but rejected by Bitcoin, on " + deviceName);" to an info. might save some people trouble.
438 2011-01-25 02:32:15 <Diablo-D3> molecular: nope, because its meaningless.
439 2011-01-25 02:32:20 <molecular> is it?
440 2011-01-25 02:32:34 <Diablo-D3> if you're on a real bitcoind, it just means the chain changed while you wernt looking
441 2011-01-25 02:33:28 <molecular> hm, I see. actual erroneous calculations will be caught by "( H == 0 )" farther out
442 2011-01-25 02:33:44 <Diablo-D3> yeah, it'll emit an error if the calc is actually wrong
443 2011-01-25 02:34:09 <molecular> ok, I retreat my suggestion then ;)
444 2011-01-25 02:34:33 <molecular> It's late here, I'll be going to bed...
445 2011-01-25 02:35:08 <Diablo-D3> also, hrrrrm
446 2011-01-25 02:35:15 <Diablo-D3> Im getting 2996 khash/sec
447 2011-01-25 02:35:23 <molecular> how big the loop?
448 2011-01-25 02:35:27 <Diablo-D3> 1024
449 2011-01-25 02:35:37 <Diablo-D3> and I changed my mhash meter to * 1024.
450 2011-01-25 02:35:39 <molecular> if you multiply that, that would be a lot
451 2011-01-25 02:35:42 <molecular> ah, ok
452 2011-01-25 02:36:02 <molecular> doesn't look so good. are we overlooking something?
453 2011-01-25 02:36:43 <molecular> I got 300*256 = 76800 before, but the 300 are extremely rough, the hash-meter jittered like hell
454 2011-01-25 02:36:58 <Diablo-D3> because kernel execution is too long
455 2011-01-25 02:37:10 <Diablo-D3> it wasnt meant to handle extremely long kernel executions
456 2011-01-25 02:37:12 <molecular> yeah
457 2011-01-25 02:37:16 <Diablo-D3> so the minimum size is too big
458 2011-01-25 02:47:17 <tcatm> Diablo-D3: just ported art's changes to m0's miner: 561 Mhash/s on a stock 5970
459 2011-01-25 02:47:29 <Diablo-D3> heh nice
460 2011-01-25 02:47:37 <Diablo-D3> I wonder if this just shits on loops
461 2011-01-25 02:48:24 <tcatm> i.e. 350 Mhash/s on a 5870 @ 900 MHz
462 2011-01-25 02:49:07 <tcatm> let's try without vectors
463 2011-01-25 02:49:22 <tcatm> 336 Mhash/s
464 2011-01-25 02:49:27 <tcatm> (5870 @ 900)
465 2011-01-25 02:49:30 <Diablo-D3> okay whats
466 2011-01-25 02:49:37 <Diablo-D3> 189052229 / 1024
467 2011-01-25 02:49:54 <tcatm> 185 Mhash/s?
468 2011-01-25 02:50:00 <Diablo-D3> I DONT THINK SO TIM
469 2011-01-25 02:50:12 <Diablo-D3> thats over double the fucking hash rate
470 2011-01-25 02:50:31 <tcatm> divide by 2?
471 2011-01-25 02:51:01 <hacim> that seems slightly better than m0n's miner
472 2011-01-25 02:51:24 <hacim> oh, wait, tcatm you ported art's changes to m0's?
473 2011-01-25 02:51:30 <tcatm> hacim: yep
474 2011-01-25 02:51:40 <hacim> tcatm: nice! got a link to the changes?
475 2011-01-25 02:51:45 <Diablo-D3> something isnt fucking right here
476 2011-01-25 02:51:58 <tcatm> hacim: some pastebin somewhere, don't have the link anymore
477 2011-01-25 02:52:07 <hacim> tcatm: no, I mean to your modified m0
478 2011-01-25 02:52:10 <hacim> :)
479 2011-01-25 02:52:21 <hacim> tcatm: what were you getting before with your 5870 @ 900 Mhz?
480 2011-01-25 02:52:22 <tcatm> I'm not done.
481 2011-01-25 02:52:30 <hacim> ah cool
482 2011-01-25 02:52:32 <tcatm> ~345
483 2011-01-25 02:52:51 <hacim> yeah, thats what I'm getting... more around 325
484 2011-01-25 02:53:03 <tcatm> using vectors (-v flag)?
485 2011-01-25 02:53:11 <hacim> yeah, -f 5 -d 0 -w 256 -v
486 2011-01-25 02:53:23 <tcatm> try -f 1 -v -w 128
487 2011-01-25 02:53:25 <Diablo-D3> I dont use vectors
488 2011-01-25 02:53:33 <hacim> 1300Mhz memory
489 2011-01-25 02:53:34 <Diablo-D3> and btw, 64 seems most optimum on such hardware
490 2011-01-25 02:53:57 <Diablo-D3> something is seriously fucked
491 2011-01-25 02:53:58 <tcatm> Diablo-D3: depends on the kernel
492 2011-01-25 02:54:07 <Diablo-D3> its saying
493 2011-01-25 02:54:20 <tcatm> 335
494 2011-01-25 02:54:38 <tcatm> so -w 128 is optimum
495 2011-01-25 02:54:45 <Diablo-D3> 513852
496 2011-01-25 02:54:53 <hacim> with -f 1 -v -w 128 i'm getting 333
497 2011-01-25 02:55:31 <hacim> 321 with -w 64
498 2011-01-25 02:56:20 <hacim> does seem like -w 128 I get more
499 2011-01-25 02:56:40 <hacim> but I can't quite make 345
500 2011-01-25 02:56:56 <tcatm> I'm using a hacked SDK 2.1
501 2011-01-25 02:57:05 <tcatm> my compiler is more efficient
502 2011-01-25 02:57:06 <Diablo-D3> okay this is wrong
503 2011-01-25 02:57:09 <Diablo-D3> and Ive broken something
504 2011-01-25 02:57:11 <hacim> tcatm: did you tweak your memory mhz?
505 2011-01-25 02:57:17 <tcatm> hacim: no
506 2011-01-25 02:57:23 <hacim> tcatm: I have mine at 1300
507 2011-01-25 02:57:25 <tcatm> the code doesn't even touch the memory
508 2011-01-25 02:57:31 <Diablo-D3> 509183/1024 is 496
509 2011-01-25 02:57:40 <Diablo-D3> 496 khash.
510 2011-01-25 02:57:44 <hacim> tcatm: interesting, what does your --od-getclocks show for your memory mhz?
511 2011-01-25 02:57:58 <tcatm> 1200
512 2011-01-25 02:58:07 <hacim> i wonder if I scaled mine back to 1200 I might get more
513 2011-01-25 02:58:12 <hacim> seems unlikely
514 2011-01-25 02:58:22 <tcatm> as I said. no memory access in kernel
515 2011-01-25 02:59:01 <hacim> i wonder why you are able to get 10k more than I am
516 2011-01-25 02:59:15 <tcatm> hacked SDK ;)
517 2011-01-25 02:59:19 <hacim> what
518 2011-01-25 02:59:22 <hacim> are you using 2.1?
519 2011-01-25 02:59:31 <tcatm> 2.1 with compiler from 2.2
520 2011-01-25 02:59:37 <hacim> whoa nelly
521 2011-01-25 02:59:48 <Diablo-D3> okay now its saying 259 mhash
522 2011-01-25 02:59:48 <hacim> i'm using 2.3
523 2011-01-25 02:59:58 <tcatm> so performance from 2.2 but works with multigpu
524 2011-01-25 03:00:28 <hacim> you mean performance from 2.1, but works with multigpu?
525 2011-01-25 03:01:18 <tcatm> compiler from 2.2, libopencl from 2.1
526 2011-01-25 03:01:30 <hacim> (my understanding is that stream sdk 2.1 is much more performant)
527 2011-01-25 03:01:50 <tcatm> no, 2.2 is faster
528 2011-01-25 03:03:14 <Diablo-D3> something is fucked here
529 2011-01-25 03:03:23 <tcatm> you can't code :P
530 2011-01-25 03:03:26 <Diablo-D3> its saying Im doing absurd levels of mhash
531 2011-01-25 03:05:06 <Diablo-D3> I think I broke my local work size management code
532 2011-01-25 03:05:43 <tcatm> I wonder if this works... :)
533 2011-01-25 03:06:02 <Diablo-D3> yeah it "works" its listing 55 mhash/sec
534 2011-01-25 03:06:11 <Diablo-D3> wait there we go, back up to 74
535 2011-01-25 03:06:35 <tcatm> what should you get on what card?
536 2011-01-25 03:06:44 <Diablo-D3> about that.
537 2011-01-25 03:07:01 <Diablo-D3> the highest consistent score Ive seen is 74.6
538 2011-01-25 03:08:38 <Diablo-D3> http://pastebin.com/WFh5VHGD
539 2011-01-25 03:08:41 <Diablo-D3> thats what mine is now
540 2011-01-25 03:11:39 <hacim> tcatm: are you just setting your LD_LIBRARY_PATH to the 2.1 lib dir for the openCL?
541 2011-01-25 03:12:03 <tcatm> hacim: mixed files from both SDKs
542 2011-01-25 03:12:21 <tcatm> LogicError: clSetKernelArg failed: invalid arg size
543 2011-01-25 03:12:23 <tcatm> ??? :)
544 2011-01-25 03:12:46 <Diablo-D3> tcatm: heh
545 2011-01-25 03:13:48 <Diablo-D3> tcatm: wheres that happening?
546 2011-01-25 03:14:04 <hacim> hm, i dont notice any appreciable difference using 2.2
547 2011-01-25 03:14:48 <tcatm> Diablo-D3: when settings args :P
548 2011-01-25 03:14:55 <Diablo-D3> tcatm: yeah but check the headers
549 2011-01-25 03:14:56 <Diablo-D3> identical
550 2011-01-25 03:15:00 <tcatm> yep
551 2011-01-25 03:15:11 <Diablo-D3> so obviously thats wrong
552 2011-01-25 03:15:31 <hacim> whoa, 924 khash/s is way worse
553 2011-01-25 03:15:44 <tcatm> three cards?
554 2011-01-25 03:16:03 <Diablo-D3> tcatm: not only that, its not saying that on mine
555 2011-01-25 03:16:16 <Diablo-D3> oh durrrrr
556 2011-01-25 03:16:26 <hacim> wow, using the 2.2 libs really hosed the hash rate
557 2011-01-25 03:16:34 <Diablo-D3> output[0] = it ^ tnonce;
558 2011-01-25 03:16:36 <Diablo-D3> hehhh
559 2011-01-25 03:16:44 <Diablo-D3> hacim: yes it has if you dont use the hack
560 2011-01-25 03:16:47 <Diablo-D3> its consuming all your cpu time
561 2011-01-25 03:17:37 <hacim> whats this 'hack'?
562 2011-01-25 03:17:48 <Diablo-D3> what miner are you on?
563 2011-01-25 03:18:20 <hacim> Diablo-D3: i've been trying m0
564 2011-01-25 03:18:32 <Diablo-D3> his doesnt use it
565 2011-01-25 03:18:46 <Diablo-D3> tcatm: now that I fixed the obvious oversight
566 2011-01-25 03:19:02 <Diablo-D3> I get 74.8 on his without loop, 74.6 on mine without loop
567 2011-01-25 03:19:15 <Diablo-D3> actually closer to 74.9
568 2011-01-25 03:19:34 <Diablo-D3> so about half a percent
569 2011-01-25 03:19:35 <Diablo-D3> lol
570 2011-01-25 03:22:35 <hacim> tcatm: alright, I give, how did you mix those two?
571 2011-01-25 03:23:10 <tcatm> hacim: don't remember what files I copied
572 2011-01-25 03:23:12 <Diablo-D3> lol okay
573 2011-01-25 03:23:16 <Diablo-D3> now that shits working
574 2011-01-25 03:23:18 <Diablo-D3> with the loop on
575 2011-01-25 03:23:22 <Diablo-D3> Im getting about
576 2011-01-25 03:23:33 <Diablo-D3> 48 mhash
577 2011-01-25 03:23:37 <hacim> hehe
578 2011-01-25 03:23:47 <tcatm> that's not bad
579 2011-01-25 03:23:54 <tcatm> meanwhile I'm at 352 @ 900
580 2011-01-25 03:24:04 <Diablo-D3> 48? down from 75?
581 2011-01-25 03:24:21 <hacim> damn
582 2011-01-25 03:24:36 <hacim> you've gone up 10k more, so now are 20k ahead of me
583 2011-01-25 03:24:54 <mrb_> picture of the future dual-GPU HD 6990: http://www.fudzilla.com/graphics/item/21635-amds-antilles-card-pixellized
584 2011-01-25 03:25:05 <Diablo-D3> mrb_: OH GOD I CAME
585 2011-01-25 03:25:07 <Diablo-D3> someone get me a towel
586 2011-01-25 03:25:49 <mrb_> two 8-pin connectors. wonder if officially it means it will be a 375W card... and what are the repercussions for AMD breaking the PCIe specs wrt. max wattage
587 2011-01-25 03:25:54 <mrb_> Diablo-D3: eww
588 2011-01-25 03:26:00 <tcatm> woah, finally a good cooler design
589 2011-01-25 03:26:11 <hacim> is that stephen colbert?
590 2011-01-25 03:26:23 <Diablo-D3> no
591 2011-01-25 03:26:27 <Diablo-D3> hes not that ugly
592 2011-01-25 03:26:34 <mrb_> the card seems shorter than a 5970
593 2011-01-25 03:27:04 <hacim> tcatm: did you find this hydra lib/compiler mix somewhere? i'm wondering how you figured this out
594 2011-01-25 03:27:32 <tcatm> hacim: no, did it myself
595 2011-01-25 03:27:34 <Diablo-D3> I wonder if I drop the loop size it'll get better
596 2011-01-25 03:28:47 <hacim> damn
597 2011-01-25 03:29:12 <Diablo-D3> lets try 16
598 2011-01-25 03:30:37 <Diablo-D3> 72.7
599 2011-01-25 03:30:59 <Diablo-D3> lets try 8
600 2011-01-25 03:33:18 <Diablo-D3> 70.5
601 2011-01-25 03:33:22 <Diablo-D3> lets try 4
602 2011-01-25 03:34:25 <echelon> when specifying your own nodes using addnode, does bitcoin still attempt to connect to the hardcoded nodes?
603 2011-01-25 03:34:54 <Diablo-D3> 67
604 2011-01-25 03:35:18 <Diablo-D3> lets try 32
605 2011-01-25 03:36:28 <Diablo-D3> 70.6
606 2011-01-25 03:36:45 <tcatm> echelon: at least it'll connect to other nodes
607 2011-01-25 03:36:49 <Diablo-D3> lets try 2
608 2011-01-25 03:36:51 <Diablo-D3> lets try 24
609 2011-01-25 03:36:59 <echelon> huh
610 2011-01-25 03:38:33 <Diablo-D3> 73.1
611 2011-01-25 03:38:57 <Diablo-D3> lets try 20
612 2011-01-25 03:39:59 <jgarzik> echelon: hardcoded nodes are just for seeding the address list.  that is separate from the connection logic.
613 2011-01-25 03:40:32 <jgarzik> echelon: -connect specifies that you should -only- connect to specified nodes, and none other.  otherwise, -addnode, IRC seeds and hardcoded seeds all go into the address poll, from which connections are made.
614 2011-01-25 03:40:47 <echelon> i used -addnode and -noirc
615 2011-01-25 03:41:01 <tcatm> 566 Mhash/s on a stock 5970
616 2011-01-25 03:41:05 <Diablo-D3> about 73.3
617 2011-01-25 03:41:17 <jgarzik> echelon: in that case, it will attempt connections to hardcoded nodes If And Only If there are no other addresses in your database.
618 2011-01-25 03:41:31 <echelon> ohh kk
619 2011-01-25 03:46:13 <echelon> wait, you would use -connect instead of -addnode?
620 2011-01-25 03:59:07 <tcatm> Error:E010:Irreducible ControlFlow Detected
621 2011-01-25 03:59:09 <tcatm> fsck :)
622 2011-01-25 04:00:55 <hacim> ok, so its been 120 blocks since the block I found, and I still don't have 50btc
623 2011-01-25 04:02:46 <hacim> am I doing something wrong?
624 2011-01-25 04:03:52 <tcatm> Diablo-D3: btw, the loop decreases performance for me, too
625 2011-01-25 04:06:12 <Diablo-D3> what are you on for hardware?
626 2011-01-25 04:06:24 <tcatm> 5870/5970
627 2011-01-25 04:06:59 <Diablo-D3> art says it sfaster for him
628 2011-01-25 04:07:14 <tcatm> Using my custom miner it is faster, too
629 2011-01-25 04:07:19 <tcatm> strange
630 2011-01-25 04:07:31 <tcatm> still 352 Mhash/s on a 5870 isn't bad :)
631 2011-01-25 04:08:25 <Diablo-D3> lets switch to 2.3
632 2011-01-25 04:08:31 <Diablo-D3> 67 without loop
633 2011-01-25 04:10:09 <Diablo-D3> actually. 68.5
634 2011-01-25 04:11:11 <Diablo-D3> lets tru 20 loops
635 2011-01-25 04:11:51 <hacim> ah damn, i miscalculated 120
636 2011-01-25 04:12:18 <Diablo-D3> 68.4 with 20
637 2011-01-25 04:12:24 <Diablo-D3> sorry, 68.5
638 2011-01-25 04:12:26 <tcatm> total gain: 70 Mhash/s on my cluster
639 2011-01-25 04:12:42 <Diablo-D3> lets try 1024
640 2011-01-25 04:14:25 <Diablo-D3> tt
641 2011-01-25 04:14:28 <Diablo-D3> 52
642 2011-01-25 04:16:35 <Diablo-D3> ooh its reached 51
643 2011-01-25 04:16:36 <Diablo-D3> er 61
644 2011-01-25 04:17:08 <Diablo-D3> lets try with sync objects
645 2011-01-25 04:18:55 <Diablo-D3> about the same
646 2011-01-25 04:19:02 <Diablo-D3> so loop just isnt worth it for me
647 2011-01-25 04:19:48 <kron> NOOOOOOOOO
648 2011-01-25 04:19:56 <kron> I LOST ALL MY BITCOINS GAMBLING
649 2011-01-25 04:19:58 <kron> FML
650 2011-01-25 04:24:15 <echelon> lol
651 2011-01-25 04:24:26 <echelon> how much did you have
652 2011-01-25 04:25:31 <tcatm> hacim: http://pastebin.com/g8QJ1R3r
653 2011-01-25 04:32:43 <hacim> tcatm: hm, i get various hunk failures
654 2011-01-25 04:32:50 <hacim> i wonder if I am using a different version than you are
655 2011-01-25 04:34:01 <tcatm> probably
656 2011-01-25 04:34:23 <tcatm> patch it manually :)
657 2011-01-25 04:35:15 <hacim> i'd prefer to make sure I have the latest
658 2011-01-25 04:37:42 <hacim> seems I have the latest git HEAD
659 2011-01-25 04:37:58 <tcatm> maybe i didn't have the latest
660 2011-01-25 04:45:40 <kron> i had 300
661 2011-01-25 04:47:39 <kron> if anyone wants to send me some cause im sad =[
662 2011-01-25 04:47:40 <kron> 1NYTPaUfig9wuXd3ToTDrUsn7Yh9wREYqR
663 2011-01-25 04:47:43 <mrb_> kron: want to recover them? pick a number between 1 and 10.
664 2011-01-25 04:47:57 <kron> 7!
665 2011-01-25 04:48:06 <mrb_> incorrect. I was thinking of 5.
666 2011-01-25 04:48:10 <mrb_> you owe me 300 BTC.
667 2011-01-25 04:48:16 <kron> -.-
668 2011-01-25 04:48:18 <mrb_> :)
669 2011-01-25 04:48:34 <mrb_> where were you gambling?
670 2011-01-25 04:48:40 <nanotube> kron: well... let that be a lesson to you for the future. :P
671 2011-01-25 04:48:43 <kron> the double or trouble
672 2011-01-25 04:49:05 <kron> they must be raking in the bitcoins
673 2011-01-25 04:49:20 <nanotube> as is any gambling establishment. the house edge.
674 2011-01-25 04:50:06 <kron> right
675 2011-01-25 04:50:11 <kron> shoulda quite while i was ahead
676 2011-01-25 04:50:14 <kron> of course
677 2011-01-25 04:50:51 <kron> i was thinking the same thing lol
678 2011-01-25 04:50:52 <OneFixt> Is there any problem with getting rid of the rounding in AmountFromValue? Will all nodes accept transactions of .01xxxxxx?
679 2011-01-25 04:51:01 <nanotube> OneFixt: yes
680 2011-01-25 04:51:03 <nanotube> they will
681 2011-01-25 04:51:43 <OneFixt> nanotube: thanks
682 2011-01-25 04:52:52 <kron> ;_;
683 2011-01-25 04:53:08 <mrb_> "Facebook confirms plans to make Credits the mandatory "universal currency - http://venturebeat.com/2011/01/24/facebook-credits-mandatory/
684 2011-01-25 04:53:26 <kron> if facebook adopted bitcoins
685 2011-01-25 04:53:29 <kron> it would be over
686 2011-01-25 04:54:50 <tcatm> ??lr
687 2011-01-25 04:54:55 <tcatm> dd
688 2011-01-25 04:55:05 <lfm> can you convert fb credits back to local money?
689 2011-01-25 04:55:38 <nanotube> mrb_: wow, 30% fee on buying fb credits. that's /bold/
690 2011-01-25 04:58:44 <hacim> tcatm: hm, i'm getting some crazy compilation errors
691 2011-01-25 04:59:02 <tcatm> hacim: you probably need to fix it for plain sdk 2.1
692 2011-01-25 04:59:49 <hacim> tcatm: http://pastebin.com/xSHHirKQ
693 2011-01-25 05:00:26 <tcatm> that's strange
694 2011-01-25 05:00:58 <tcatm> try this kernel http://pastebin.com/RF5R9d10
695 2011-01-25 05:01:13 <kron> quick question, how long will it take me to generate coins with a quad core AMD and a nvidia 8800gtx
696 2011-01-25 05:01:22 <kron> should i join a pool?
697 2011-01-25 05:01:27 <hacim> oddly it works with 2.2, but not 2.1 or 2.3
698 2011-01-25 05:01:38 <tcatm> oh wait
699 2011-01-25 05:01:47 <tcatm> change line 20 to #define rotr(x, y) amd_bitalign((u)x, (u(x, (u)y)
700 2011-01-25 05:01:49 <tcatm> er...
701 2011-01-25 05:02:01 <tcatm> change line 20 to #define rotr(x, y) amd_bitalign((u)x, (u)x, (u)y)
702 2011-01-25 05:04:06 <hacim> hm, that doesn't seem to help
703 2011-01-25 05:04:32 <hacim> oh wait, you said try that kernel
704 2011-01-25 05:04:40 <tcatm> yes and change that line
705 2011-01-25 05:06:35 <Keefe> ;;bc,calc 30000
706 2011-01-25 05:06:36 <gribble> The average time to generate a block at 30000 Khps, given current difficulty of 18437.64439217 , is 4 weeks, 2 days, 13 hours, 13 minutes, and 55 seconds
707 2011-01-25 05:06:40 <Keefe> kron: ^
708 2011-01-25 05:07:07 <nanotube> kron: in other words... yes you should. :)
709 2011-01-25 05:07:51 <Keefe> a pool will give you about 50 btc over that time, but you'll get a little each day instead of all at once
710 2011-01-25 05:08:09 <nanotube> ;;bc,stats
711 2011-01-25 05:08:10 <gribble> Current Blocks: 104460 | Current Difficulty: 18437.64439217 | Next Difficulty At Block: 104831 | Next Difficulty In: 371 blocks | Next Difficulty In About: 2 days, 3 hours, 31 minutes, and 40 seconds | Next Difficulty Estimate: 21827.79160716
712 2011-01-25 05:08:10 <hacim> tcatm: interesting that seems to work, although it blurts this out before it starts hashing: http://pastebin.com/jxR5dbgh
713 2011-01-25 05:08:28 <tcatm> that's okay...
714 2011-01-25 05:08:43 <tcatm> how much did you get before that patch?
715 2011-01-25 05:08:49 <kron> nice i have to get that set up
716 2011-01-25 05:09:53 <hacim> tcatm: 332663, now I'm getting 336162
717 2011-01-25 05:10:03 <hacim> not the increase you were seeing
718 2011-01-25 05:10:20 <tcatm> at least it's an improvement
719 2011-01-25 05:11:07 <hacim> why is it blurting that out before it hashes?
720 2011-01-25 05:11:43 <nanotube> it's just a warning, shouldn't affect anything.
721 2011-01-25 05:11:49 <tcatm> I didn't care to do some calculations "the correct way"
722 2011-01-25 05:12:26 <hacim> you dont worry that doing them the less than correct way could impact potential results?
723 2011-01-25 05:13:05 <hacim> probably not an impact to speed, but potentially to accuracy or validity?
724 2011-01-25 05:13:23 <tcatm> I verified it and it's working correct.
725 2011-01-25 05:20:24 <hacim> tcatm: in case you want to update to the latest git, this is the diff: http://pastebin.com/v98gWWLc
726 2011-01-25 05:21:31 <tcatm> thanks
727 2011-01-25 05:22:24 <hacim> now if I could only ramp up to 350
728 2011-01-25 05:59:18 <Diablo-D3> hmm
729 2011-01-25 05:59:25 <Diablo-D3> wonder if I should push new version of my miner
730 2011-01-25 06:07:05 <joe_1> YESSSSSSSSSS diablo released a new miner if i can get it first i can take over the world
731 2011-01-25 06:08:32 <Diablo-D3> ?
732 2011-01-25 06:08:33 <tcatm> too late, already released a patch for m0's
733 2011-01-25 06:08:46 <Diablo-D3> tcatm: yeah but its not as good as mine
734 2011-01-25 06:09:11 <tcatm> What does yours do better?
735 2011-01-25 06:09:26 <Diablo-D3> I think its a little late to be asking that.
736 2011-01-25 06:09:28 <tcatm> (hint: with testing only H==0 in kernel I don't see any improvement)
737 2011-01-25 06:10:47 <tcatm> Can you make a binary that doesn't define rotate using bitalign?
738 2011-01-25 06:11:02 <tcatm> Then I could run your miner and see how fast it is.
739 2011-01-25 06:11:12 <Diablo-D3> erm, mine doesnt if you dont have it
740 2011-01-25 06:11:32 <tcatm> Problem is: I have bitalign, but I don't have amd_bitalign()
741 2011-01-25 06:11:49 <tcatm> i.e. my compiler will translate rotate() into bitalign() on its own
742 2011-01-25 06:12:02 <Diablo-D3> see above.
743 2011-01-25 06:12:14 <tcatm> ?
744 2011-01-25 06:12:35 <Diablo-D3> my miner already works on targets that dont have bitalign
745 2011-01-25 06:12:42 <Diablo-D3> otherwise I could never run it myself
746 2011-01-25 06:13:14 <tcatm> it says #if BITALIGN... right on top of your kernel!
747 2011-01-25 06:13:32 <Diablo-D3> yes, and look very closely at that code!
748 2011-01-25 06:13:52 <tcatm> it defines rot() using bitalign when that flag is set?
749 2011-01-25 06:14:02 <Diablo-D3> yes
750 2011-01-25 06:14:09 <tcatm> whthat's wrong
751 2011-01-25 06:14:37 <tcatm> or... your java code shouldn't define that flag when running on my miner
752 2011-01-25 06:14:52 <Diablo-D3> how so? if you dont have amd_media_ops, then my miner never defines BITALIGN
753 2011-01-25 06:15:11 <Diablo-D3> so it does rotate()
754 2011-01-25 06:16:02 <tcatm> is it possible to change your kernel when using your binaries?
755 2011-01-25 06:16:14 <Diablo-D3> yes, just edit it
756 2011-01-25 06:16:32 <Diablo-D3> but I dont get why you're having a problem
757 2011-01-25 06:16:57 <Diablo-D3> if you dont have cl_amd_media_ops, then it never uses amd_bitalign
758 2011-01-25 06:17:35 <jgarzik> http://www.bitcoin.org/smf/index.php?topic=820.msg40022#msg40022
759 2011-01-25 06:18:15 <Diablo-D3> heh
760 2011-01-25 06:18:31 <tcatm> I should probably rewrite 4way. I can think of many improvements after writing a few opencl miner
761 2011-01-25 06:19:08 <Pegasus-Rider> BTW, 4-way is for 64-bit machines only, right?
762 2011-01-25 06:19:17 <tcatm> yep
763 2011-01-25 06:19:48 <jgarzik> well
764 2011-01-25 06:19:58 <Diablo-D3> ArtForz: I now have your kernel in my miner
765 2011-01-25 06:20:39 <ArtForz> neat
766 2011-01-25 06:21:23 <tcatm> I ported them to m0's. 567 MHash/s on stock 5970
767 2011-01-25 06:21:28 <Diablo-D3> I get about a half of a percent increase on my 4850
768 2011-01-25 06:21:36 <jgarzik> 4way works just fine on 32-bit
769 2011-01-25 06:21:57 <ArtForz> well, I never bothered testing how it performs on 4xxx
770 2011-01-25 06:22:07 <ArtForz> 4xxx is hopeless anyways
771 2011-01-25 06:22:28 <lfm> but its ionly faster on machines that also support 64 bit even if you are only running in 32 bit mode
772 2011-01-25 06:23:19 <tcatm> Diablo-D3: how do I tell your miner which GPU to use?
773 2011-01-25 06:23:31 <jgarzik> true
774 2011-01-25 06:23:37 <ArtForz> 567, not too shabby
775 2011-01-25 06:23:37 <lfm> -d
776 2011-01-25 06:24:11 <Diablo-D3> tcatm: cant yet.
777 2011-01-25 06:24:22 <Diablo-D3> it uses all
778 2011-01-25 06:25:09 <tcatm> that's stupid
779 2011-01-25 06:25:17 <Diablo-D3> not really
780 2011-01-25 06:25:26 <Diablo-D3> most people would never want to run a miner just on one
781 2011-01-25 06:26:33 <tcatm> your miner does 308 Mhash/s on a 5870 @ 900
782 2011-01-25 06:27:11 <ArtForz> thats... bad
783 2011-01-25 06:27:14 <tcatm> Mkay, it just jumped to 332
784 2011-01-25 06:27:28 <ArtForz> should be closer to 350
785 2011-01-25 06:27:28 <CyanDynamo> http://arstechnica.com/web/news/2011/01/philanthroper-hopes-to-boost-charity-of-the-day-1-at-a-time.ars
786 2011-01-25 06:27:36 <tcatm> never seen such a jumpy hashmeter on my GPUs before
787 2011-01-25 06:28:06 <Diablo-D3> its only a 30 second average
788 2011-01-25 06:28:38 <tcatm> ArtForz: m0's with your kernel does ~352
789 2011-01-25 06:28:52 <Diablo-D3> tcatm: also, give it time
790 2011-01-25 06:28:53 <tcatm> Diablo-D3: make a binary of your new kernel
791 2011-01-25 06:28:58 <Diablo-D3> you can wait.
792 2011-01-25 06:39:06 <ArtForz> how the F do you optimize for 5xxx on a 4xxx ?
793 2011-01-25 06:39:27 <Diablo-D3> ArtForz: not specifically for 5xxx, just in general
794 2011-01-25 06:39:49 <ArtForz> good luck with that
795 2011-01-25 06:48:15 <Diablo-D3> its now running about 76.2
796 2011-01-25 06:59:04 <Diablo-D3> it now does about 76.6
797 2011-01-25 07:02:26 <tcatm> You should really get a 5870 or 5970
798 2011-01-25 07:10:04 <Diablo-D3> I dont think its getting any faster than this for the time being
799 2011-01-25 07:15:02 <Diablo-D3> so lets say
800 2011-01-25 07:15:32 <Diablo-D3> a 3% increase in speed on my 4850
801 2011-01-25 07:15:42 <Diablo-D3> that seems about right
802 2011-01-25 07:17:22 <Diablo-D3> ArtForz: isnt that what you got on your shit vs m0's?
803 2011-01-25 07:18:07 <ArtForz> somewhere around there
804 2011-01-25 07:18:23 <Diablo-D3> heh, interesting, your kernel did very well on 4xxx then
805 2011-01-25 07:19:23 <ArtForz> well, converting W from array to variables seems to help the compiler quite a bit
806 2011-01-25 07:21:14 <CIA-98> DiabloMiner: Patrick McFarland master * rdc5b12e / (3 files in 3 dirs): Switch to ArtForz's kernel, 3% speed increase - http://bit.ly/exOmWo
807 2011-01-25 07:23:19 <Diablo-D3> tcatm: there
808 2011-01-25 07:23:22 <Diablo-D3> new binary uploaded
809 2011-01-25 07:38:41 <tcatm> Diablo-D3: did only the kernel change?
810 2011-01-25 07:39:33 <sipa> seems a bit faster on my 4870 too
811 2011-01-25 07:40:34 <Diablo-D3> tcatm: and the stuff to support it, yes
812 2011-01-25 07:40:42 <sipa> aha, you enable the looping trick only on 5xxx
813 2011-01-25 07:40:53 <Diablo-D3> sipa: it wont work on 4xxx
814 2011-01-25 07:41:01 <sipa> ok
815 2011-01-25 07:41:12 <Diablo-D3> for small loops, it doesnt win, for large loops, it shits itself
816 2011-01-25 07:41:25 <Diablo-D3> and I suspect nvidia cant win at all here
817 2011-01-25 07:45:25 <tcatm> Diablo-D3: 342 Mhash/s on 5870 @ 900MHz
818 2011-01-25 07:46:58 <Diablo-D3> what was the target?
819 2011-01-25 07:47:09 <tcatm> diff 1
820 2011-01-25 07:47:14 <Diablo-D3> no I meant
821 2011-01-25 07:47:14 <tcatm> using the pool
822 2011-01-25 07:47:17 <Diablo-D3> what was I supposed to hit
823 2011-01-25 07:47:23 <tcatm> don't know?
824 2011-01-25 07:47:38 <Diablo-D3> well, the wiki ways 314 mhash
825 2011-01-25 07:47:40 <sipa> over 9000!
826 2011-01-25 07:48:04 <Diablo-D3> so thats a 9% increase
827 2011-01-25 07:48:05 <tcatm> for 850 MHz IIRC
828 2011-01-25 07:48:06 <Diablo-D3> and hard to believe
829 2011-01-25 07:48:15 <Diablo-D3> ahh yes 850
830 2011-01-25 07:48:24 <tcatm> so 323 Mhash/s @ 850 MHz
831 2011-01-25 07:48:35 <Diablo-D3> so a 3% increase
832 2011-01-25 07:48:43 <tcatm> still 9 Mhash/s worse than m0's
833 2011-01-25 07:49:13 <sipa> than m0's own kernel, or than m0 with Art's kernel?
834 2011-01-25 07:49:24 <tcatm> m0's with art's
835 2011-01-25 07:49:38 <tcatm> though it's pretty much the same kernel as Diablo-D3 uses
836 2011-01-25 07:49:54 <Diablo-D3> did you move the stuff out of the kernel like I did?
837 2011-01-25 07:49:58 <tcatm> sure
838 2011-01-25 07:50:02 <Diablo-D3> heh
839 2011-01-25 07:50:31 <Diablo-D3> you're running mine with -w 64 -f 1?
840 2011-01-25 07:50:33 <tcatm> What's that "Attempt x found" stuff?
841 2011-01-25 07:50:39 <Diablo-D3> H == 0.
842 2011-01-25 07:50:41 <tcatm> -w 128 -f 1
843 2011-01-25 07:50:53 <Diablo-D3> -w 64 should be faster
844 2011-01-25 07:51:06 <tcatm> for m0's 128 is faster
845 2011-01-25 07:52:16 <tcatm> no change
846 2011-01-25 07:52:27 <Diablo-D3> heh, interesting
847 2011-01-25 07:52:40 <tcatm> you should smooth the hashmeter
848 2011-01-25 07:52:57 <Diablo-D3> it used to be 100% smooth
849 2011-01-25 07:53:00 <Diablo-D3> but people bitched about it
850 2011-01-25 07:53:09 <tcatm> it's jumping +/- 70 mhash/s
851 2011-01-25 07:53:22 <Diablo-D3> with -f 1? Im not surprised
852 2011-01-25 07:53:33 <tcatm> what's that -f 1 anyway?
853 2011-01-25 07:53:36 <Diablo-D3> my miner tends to destroy the system clock due to interrupt spam
854 2011-01-25 07:53:42 <Diablo-D3> it runs one kernel per second.
855 2011-01-25 07:54:00 <tcatm> can I tell it to run one kernel after the other?
856 2011-01-25 07:54:08 <Diablo-D3> huh?
857 2011-01-25 07:54:27 <tcatm> just enqueue the next kernel after the previous finished and you read the buffer
858 2011-01-25 07:54:41 <Diablo-D3> erm, it already does that?
859 2011-01-25 07:54:49 <tcatm> so it should output a constant hashrate
860 2011-01-25 07:54:51 <Diablo-D3> and its already running 3 concurrently?
861 2011-01-25 07:54:59 <tcatm> on 3 GPUs, yep
862 2011-01-25 07:55:05 <Diablo-D3> on _1_ GPU.
863 2011-01-25 07:55:12 <Diablo-D3> it forces the driver queue to be stuffed.
864 2011-01-25 07:56:24 <tcatm> strange
865 2011-01-25 07:56:51 <tcatm> when using the hardware clock as timing reference the hashrate is smooth
866 2011-01-25 07:57:05 <Diablo-D3> yeah, because it fucks over time
867 2011-01-25 07:57:05 <tcatm> 345
868 2011-01-25 07:57:28 <Diablo-D3> happens on art's shit too
869 2011-01-25 07:58:45 <cosurgi> hiya in the morning.
870 2011-01-25 07:58:58 <Diablo-D3> so thats more like a 4% increase
871 2011-01-25 07:59:02 <cosurgi> Diablo-D3: it just 'git pull'ed your code. Does it work?
872 2011-01-25 07:59:08 <cosurgi> s/it/I.
873 2011-01-25 07:59:16 <Diablo-D3> cosurgi: one could assume
874 2011-01-25 07:59:16 <tcatm> cosurgi: yep
875 2011-01-25 07:59:33 <cosurgi> ok, let me give it a try :)
876 2011-01-25 08:00:08 <Diablo-D3> tcatm: dont you have a 5970?
877 2011-01-25 08:00:16 <tcatm> I have a few, yes.
878 2011-01-25 08:00:23 <tcatm> I usually do testing on the 5870, though.
879 2011-01-25 08:01:14 <Diablo-D3> well, this would get 552 on a 5970 according to your numbers
880 2011-01-25 08:01:20 <Diablo-D3> which is almost as much as art's 556
881 2011-01-25 08:01:21 <tcatm> let's try
882 2011-01-25 08:04:11 <cosurgi> Diablo-D3: radeon 5830, before: 213 Mh, now: 178 Mh (but it seems to go up)
883 2011-01-25 08:04:45 <Diablo-D3> wait for it to settle after awhile
884 2011-01-25 08:04:52 <cosurgi> ok. 184 now.
885 2011-01-25 08:04:55 <Diablo-D3> also, who knows how this'll act on a 5830
886 2011-01-25 08:05:02 <Diablo-D3> thats a rather weird chip
887 2011-01-25 08:05:06 <tcatm> ~560 on 5970
888 2011-01-25 08:05:15 <Diablo-D3> tcatm: niiiiiiice
889 2011-01-25 08:05:31 <sipa> maybe the looping should be disabled separately>
890 2011-01-25 08:05:36 <cosurgi> 202 Mh ....
891 2011-01-25 08:05:58 <tcatm> sometimes it even shows a total of 2000000 khash/s for 2 cards ;)
892 2011-01-25 08:06:15 <Diablo-D3> yeah its probably sticking somewhere
893 2011-01-25 08:06:31 <cosurgi> and back down to 186 Mh.
894 2011-01-25 08:06:35 <Diablo-D3> I should throw in a check kernel work size check
895 2011-01-25 08:06:47 <tcatm> does that cause problems?
896 2011-01-25 08:06:54 <cosurgi> ok. I'm rolling back to previous version, and check on dual 5850
897 2011-01-25 08:06:54 <Diablo-D3> it'll run with 0
898 2011-01-25 08:06:57 <Diablo-D3> which its nuts
899 2011-01-25 08:07:03 <ArtForz> I gave up on dynamically adjusting globalworksize
900 2011-01-25 08:11:29 <cosurgi> dual 5850, before: 518 Mh, now: 452 Mh. Seems that it stopped going up.
901 2011-01-25 08:11:44 <Diablo-D3> cosurgi: try -w 64 or -w 128
902 2011-01-25 08:11:51 <cosurgi> I am doing -w 64
903 2011-01-25 08:12:01 <cosurgi> whoa, now it's 397 Mh
904 2011-01-25 08:12:04 <tcatm> Diablo-D3: your miner is nice for multi gpu use, though. much easier to track than one process/gpu
905 2011-01-25 08:12:07 <cosurgi> I'll try -w 128
906 2011-01-25 08:12:29 <Diablo-D3> cosurgi: what sdk is this? 2.1?
907 2011-01-25 08:12:38 <cosurgi> yes. I didn't update
908 2011-01-25 08:12:47 <cosurgi> should I?
909 2011-01-25 08:12:48 <sipa> tcatm: do you enable the vectorization in m0/art on 5970?
910 2011-01-25 08:13:04 <Diablo-D3> sipa: it doesnt make it faster.
911 2011-01-25 08:13:32 <tcatm> sipa: yep
912 2011-01-25 08:13:45 <tcatm> depends on kernel
913 2011-01-25 08:14:09 <cosurgi> 07:01 < hacim> oddly it works with 2.2, but not 2.1 or 2.3
914 2011-01-25 08:14:23 <Diablo-D3> cosurgi: the real problem is, your systems could be fucked
915 2011-01-25 08:14:24 <tcatm> It's all about using as little registers as possible while doing as much as work in a thread as possible
916 2011-01-25 08:14:29 <Diablo-D3> cosurgi: you're lower than you should be
917 2011-01-25 08:14:30 <cosurgi> hacim: the new Diablo + ArtForz kernel works best with 2.2 SDK ?
918 2011-01-25 08:14:53 <sipa> tcatm: could you give a number?
919 2011-01-25 08:15:04 <Diablo-D3> cosurgi: dual 5850 should be about 482 or more
920 2011-01-25 08:15:21 <cosurgi> 441 Mh now, -w 128. With your previous miner it was 518 Mh.\n191924
921 2011-01-25 08:15:39 <tcatm> sipa: 352 with vectors, 335 without (5870 @ 900)
922 2011-01-25 08:15:46 <cosurgi> I didn't set any -f, what's the default?
923 2011-01-25 08:15:51 <Diablo-D3> 60
924 2011-01-25 08:15:54 <sipa> tcatm: both with art's kernel?
925 2011-01-25 08:16:03 <tcatm> Diablo-D3: your miner now shows close to 601 Mhash/s for stock 5970
926 2011-01-25 08:16:14 <tcatm> sipa: yep
927 2011-01-25 08:16:29 <Diablo-D3> ArtForz: whats your high score now?
928 2011-01-25 08:16:30 <sipa> looks like Diablo-D3 should implement the vector trick too :)
929 2011-01-25 08:16:39 <Diablo-D3> sipa: erm, I implemented it first
930 2011-01-25 08:16:43 <sipa> oh
931 2011-01-25 08:16:45 <Diablo-D3> I got rid of it awhile ago
932 2011-01-25 08:16:48 <sipa> i'm sorry
933 2011-01-25 08:17:07 <sipa> i'll reformulate: looks like Diablo-D3 should re-implement the vector trick :)
934 2011-01-25 08:17:09 <Diablo-D3> the only platform it was legitimately faster on was 4xxx, and AMD fixed that
935 2011-01-25 08:17:24 <Diablo-D3> the compiler currently autovectorizes correctly.
936 2011-01-25 08:18:04 <ArtForz> 601Mh/s on stock 5970 using OCL? not likely.
937 2011-01-25 08:18:08 <sipa> apparently not, if i can believe tcatm's numbers
938 2011-01-25 08:18:25 <tcatm> ArtForz: yep, his hashmeter is broken
939 2011-01-25 08:18:30 <Diablo-D3> it is not.
940 2011-01-25 08:18:33 <cosurgi> heh, it's easier to believe that it's slower :)
941 2011-01-25 08:18:44 <cosurgi> which is my case ;-)
942 2011-01-25 08:19:01 <sipa> i think on cosurgi's hardware maybe the looping slows things down
943 2011-01-25 08:19:13 <cosurgi> I tried -w 64 -f 60, and -w 64 0f 5.
944 2011-01-25 08:19:16 <cosurgi> omg, wait.
945 2011-01-25 08:19:23 <Diablo-D3> sipa: unlikely
946 2011-01-25 08:19:32 <cosurgi> -w 64 -f 5 gives 534 Mh
947 2011-01-25 08:19:55 <cosurgi> buuu, it went down to 539 Mh.
948 2011-01-25 08:20:08 <tcatm> it shows 1988477 khash/s for 2x 5970 @ 810 MHz
949 2011-01-25 08:20:15 <cosurgi> but still higher than original 518 Mh
950 2011-01-25 08:20:23 <sipa> 534 -> 539, nice slowdown!
951 2011-01-25 08:20:30 <Diablo-D3> I love slowdowns like that
952 2011-01-25 08:20:52 <cosurgi> oh, fuck, it went up, and I can't read
953 2011-01-25 08:20:53 <slush1> Diablo-D3: is the binary also updated to artforz kernel?
954 2011-01-25 08:20:58 <Diablo-D3> 1988 mhash? wtf
955 2011-01-25 08:21:03 <Diablo-D3> now thats just impossible
956 2011-01-25 08:21:09 <Diablo-D3> slush1: yes
957 2011-01-25 08:21:17 <slush1> great, i'll try it too
958 2011-01-25 08:21:40 <ArtForz> I'll have to count operations, but iirc with 100% VLIW usage and optimizing out all constant calcs, ~670Mh/s is max on stock 5970
959 2011-01-25 08:22:01 <ArtForz> *570
960 2011-01-25 08:22:21 <cosurgi> ok, it seems to be steady at 530 Mh now.
961 2011-01-25 08:22:33 <ArtForz> theoretical max with BFI_INT is ~620
962 2011-01-25 08:22:46 <cosurgi> um, now it's really jumping around, 540 now, 539, 591 ???? 550  528, wtf?
963 2011-01-25 08:23:08 <Diablo-D3> cosurgi: at low -f values, it bends time
964 2011-01-25 08:23:13 <cosurgi> ok.
965 2011-01-25 08:23:25 <tcatm> Diablo-D3: what about using the hardware clock as reference?
966 2011-01-25 08:23:39 <Diablo-D3> tcatm: it uses whatever your kernel gives it
967 2011-01-25 08:23:50 <tcatm> which is probably cpu time
968 2011-01-25 08:24:18 <cosurgi> so the problem might be with hasmeter. It could give wrong value.... ? How wrong?
969 2011-01-25 08:24:21 <tcatm> or you could use those new HPET
970 2011-01-25 08:24:36 <Diablo-D3> new?
971 2011-01-25 08:24:40 <Diablo-D3> they havent been new for like 10 years