Pretaktovanie.sk Folding@Home (najmä GPU) tím

[vindy]

pre porovnanie:
OC Q6600@3,4GHz - Project 2653 = ~2500-3000 PPD
OC 9600gt - Project 5796 = 3416 PPD
OC 8800GTX - Project 5766 = 5865 PPD
OC 9800GX2 - Project 5768 = ~11500 PPD

[amd_sk]

kks sak ma predbehne aj 9600GT

[cool_point]

V Linuxe mi staci napisat ./fah6 -smp a spusti sa mi SMP CPU klient
bez problemov, jednu ulohu za 1920 p. spocita na Q9400 za 7 hod.
Skusal som 6.23 beta console na Win 7, aj 6.24, cez Properties som pridal -smp switch,
ale nechce mi stiahnut ulohy, stale dookola sa pripaja, ale nic nestiahne.
Funguje vam ten CPU klient vo Win 7 64-bit s tym parametrom -smp ?

Čo je ten parameter SMP? Lebo ja mám Win7 x64 a ide mi normalne CPU klient.

[dexterav]

parameter -smp zariadi že viac jadrové cpu bude používať všetky jadrá
inak ide len 1
čo sa týka ATI ich výkon vo FAH má veľmi daleko k optimálnemu, bohužiaľ

[cool_point]

Prečo mi potom v správcovi úloh ukazuje na dvoj jadre 100%? Obidve jadrá pracujú na 100%.

[dexterav]

potom nemáš čo riešiť...
mne bez toho parametra ide vždy len 1 CPU...
dokonca aj readme je napísané že bez toho parametru pôjde iba 1 cpu ...

[Panoramix0903]

GPU3: Next generation GPU core, based on OpenMM. We have been making major advances in GPU simulation, with the key advances going into OpenMM, our open library for molecular simulation. OpenMM started with our GPU2 code as a base, but has really flourished since then. Thus, we have rewritten our GPU core to use OpenMM and we have been testing that recently as well. It is designed to be completely backward compatible, but should make simulations much more stable on the GPU as well as add new science features. A key next step for OpenMM is OpenCL support, which should allow much more efficient use of new ATI GPUs and beyond.

All the RV870 (Radeon HD 5850/5870) folders rejoice We will now be able to access and run threads in protected memory (the advantage NV GPUs have had over ATi GPUs for far too long). This is a given due to the usage of OpenCL and OpenMM. I would guesstimate more than TWICE the performance of a GTX295 with a single RV870 (and that's being conservative). Of course I'm basing this on both of them using the GPU3 client. A GTX295 has the computational power of ~1.2TFLOPs (don't be fooled by the missing MUL numbers nVIDIA quote of 1.8TFLOPs). When you compare this computational ability with that of RV870 (2.72TFLOPs) you get the idea. This isn't a game, this is pure OpenCL code, TMUs, MEM bandwidth and ROPs play no role in determining performance. The determining factor, performance wise, is in the hands of the developer as well as the computational architecture itself and if Vijay Pande's words are any indication... he's bent on tweaking it in order to pull as much performance as possible out of AMD cards.

Using OpenCL should also boost performance for all Radeon HD 4xxx cards as well.

[Panoramix0903]

Ahojte

V robote som namontoval do kompu fungl novú Sapphire Radeon HD4650 - 512 MB DDR2
Ale keď pustím Folding Home GPU2 klienta, tak mi nič nepočíta a hlási chybu s nestabilitou.

[12:58:02] *------------------------------*
[12:58:02] Folding@Home GPU Core - Beta
[12:58:02] Version 1.24 (Mon Feb 9 11:00:12 PST 2009)
[12:58:02]
[12:58:02] Compiler : Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 14.00.50727.762 for 80x86
[12:58:02] Build host: amoeba
[12:58:02] Board Type: AMD
[12:58:02] Core :
[12:58:02] Preparing to commence simulation
[12:58:02] - Looking at optimizations...
[12:58:02] - Created dyn
[12:58:02] - Files status OK
[12:58:02] - Expanded 96567 -> 489152 (decompressed 506.5 percent)
[12:58:02] Called DecompressByteArray: compressed_data_size=96567 data_size=489152, decompressed_data_size=489152 diff=0
[12:58:02] - Digital signature verified
[12:58:02]
[12:58:02] Project: 5739 (Run 3, Clone 106, Gen 396)
[12:58:02]
[12:58:03] Assembly optimizations on if available.
[12:58:03] Entering M.D.
[12:58:09] Tpr hash work/wudata_05.tpr: 1735828925 693484286 2690237198 2189565565 3848647806
[12:58:09] Working on Protein
[12:58:10] Client config found, loading data.
[12:58:10] Starting GUI Server
[12:58:19] mdrun_gpu returned
[12:58:19] Nonzero force sum on GPU
[12:58:19]
[12:58:19] Folding@home Core Shutdown: UNSTABLE_MACHINE
[12:58:22] CoreStatus = 7A (122)
[12:58:22] Sending work to server
[12:58:22] Project: 5739 (Run 3, Clone 106, Gen 396)
[12:58:22] - Error: Could not get length of results file work/wuresults_05.dat
[12:58:22] - Error: Could not read unit 05 file. Removing from queue.
[12:58:22] EUE limit exceeded. Pausing 24 hours.

Už som aj viackrát preinštaloval klienta, zmazal všetky config dáta týkajúce sa Folding Home GPU, a nič.
Ako som povedal, karta je fungl nová, nepretaktovaná. Windows XP nezamrza, obraz sa nerozpadáva, skrátka som z toho jeleň a budem vďačný za každú radu, okrem - vymeň kartu za nVidia...

[dexterav]

často to je OC a rovnako často nikto nevie čím to je
mne nexce zobrať druhú GK
s prakticky rovnakým hlásením

[adun]

Este treba skontrolovat aj stabilitu CPU, lebo pri ATI kartach to zerie aj 40 % z CPU.

mne to zere cca 50-60%

[Panoramix0903]

Zatiaľ som to vyriešil tak, že som tú HD4650 dal do kelu, kúpil HD4770 a tam mi folding ide v pohode.
Len by som ešte mal jednu otázku, ohľadom teploty karty. Tu som zosnímal teploty pri maximálnej záťaži karty.
Sú OK, alebo moc vysoké?

[valistar]

ked fakt ide fan iba na 30% tak su moc nizke
ale nie, velmi pekne teploty, do 90 stupnov si v pohode

[Panoramix0903]

Vieš, mám relatívne dobre vetraný case LianLi. Vpredu 2x 12cm Noctua 800 RPM, vzadu 1x 12cm Noctua 1200 RPM, v zdroji 14cm Enermax ventilátor, tak sa snažím mať nejaký ťah vo vnútri, aby mi tam nestál teplý vzduch.

[Panoramix0903]

2962 PPD mi dáva HD4770

[vindy]

FAH na Q9400, na 3.2 GHz, tak mas 6000 ppd z CPU.

to asi tazko, ak bude realne dosahovat 2500PPD tak moze byt rad, ale nejaka rezia ide na system, programy a GPU FOLDING, takze PPD aj pod 2000 by som sa necudoval...