Old BitMinter thread (use the other one)

[DrHaribo]

Yes, you really need manual vector handling on AMD. Their auto vectorization isn't very good.

I made some changes to vectors. Appreciate it if you could try again. BFI_INT on and off are different kernels, so might yield different results.

[1714842308]

1714842308

[OCedHrt]

Getting better performance than phatk on my 4850. Up to 80Mhash/s vs 77Mhash/s.

Would be nice to have an average mint speed because my card runs at 110C so the driver down clocks every 5 seconds.

[DrHaribo]

Cool, OCedHrt, good to hear you are getting good performance. Adding "average speed" to my TODO list - good idea!

And... First block created!

[OCedHrt]

What's the hashrate of your pool so far?

Btw, Catalyst 11.7 with SDK 2.5 gave me huge increases to DiabloMiner and Phatk but only about 1% on your miner.

DiabloMiner 82MHash/s
Phoenix Phatk 80Mhash/s
BitMiner 81Mhash/s

And downclocked to 500 Mhz because my GPU runs at 110C
DiabloMiner 59Mhash/s
Phoenix Phatk 56Mhash/s
BitMiner 57Mhash/s

The kernel for phatk compiled differently under SDK 2.5 and I assume the same for DiabloMiner. Are you including a precompiled kernel with BitMiner or is it compiled at run time?

I also have manual vectors and BFI disabled on your miner. BFI isn't supported on a 4850 so enabling it won't do anything, but weirdly, manual vectors run slower on BitMinter but vectors run faster on Phatk/DiabloMiner as long as worksize is <= 128. Actually, on SDK 2.5 DiabloMiner runs faster with worksize = 128 while on SDK 2.4 it was worksize = 64. Phatk is still worksize = 64. And for BitMinter worksize = 64 is optimal as well.

Edit: And..I can't connect now.

[DrHaribo]

The hashrate is still quite low. And I still haven't programmed in an estimated hashrate for the pool - but it's coming soon!

Sorry about connection issues. While I was working on the server I lost all connectivity with it. I got things running again as soon as I was able to reconnect.

As for hashing speed of BitMinter: I made a couple optimizations the other day. I can't remember if I put those up on bitminter.com friday or saturday. Then I tested it on radeon 6990 and geforce gtx 580. Both GPUs ran faster on BitMinter than DiabloMiner and Phoenix w/phatk. Although with DiabloMiner the difference was barely 1%. That was on Catalyst 11.6 though. I'll have to try on Catalyst 11.7.

The OpenCL kernel is compiled at runtime so different drivers can definitely make a difference there.

The vectors thing seems a little random. For me manual vectors help a lot on radeon 6990, but makes geforce gtx580 run much slower. The GTX580 usually runs faster with vectors with other miners, though. Must be a weirdness with my kernel.

[DrHaribo]

Pool backend updated: Faster block change detection. Better cheat protection. Some changes to scale better.

Seems to be running fine.

Regarding source code: The miner, pool backend and web application are all closed source. At least for now. If you prefer open source or a command-line miner, then DiabloMiner seems to me like a good alternative. But all miners should work with the pool.

[DrHaribo]

Oops, we have two threads. Better use the other one: https://bitcointalk.org/index.php?topic=27062.0


To answer your question: Yep, more stats are coming.