Arrivato a pagina 9 quel thread è stato lockato, per forza, tutti i Folder con le ATI sono incazzati neri...
Le ATI 2xxx e le più lente 3xxx non riescono manco a finire la WU prima della deadline.
Il client è lento perchè è retrocompatibile con le 2xxx e 3xxx quindi va veloce come loro più o meno
Le NVIDIA invece da subito hanno guadagnato il primo posto
Le 4xxx sono fuori da anni ma non hanno fatto niente
E idem ora con le 5xxx
Mi viene da piangere.
It is well documented here on the forum that HD2xxx and some HD3xxx hardware cannot turn in certain proteins in time, which basically throws away 4 days of work (time and electricity) every time that happens.
There have also been various statements made by mhouston and others (I really don't have to link to them do I) that the work on HD4xxx and now HD5xxx optimizations has not begun.
So we are in a peculiar situation where the client is supporting hardware that the server rejects result from (deadline constraints), meanwhile that support is making the top end hardware fold slower (than its potential).
If you feel any of these statements are inaccurate then I would happily be proven wrong (please supply link to further reading), as you can tell from my post count I do not come here often, I don't read every forum post, I only pick up bits of information here and there.
Now that I am about to go out and invest in HD58x0 hardware and I am already reading that it will perform at roughly the same speed as what I have already (despite being twice as powerful in all other applications), as a *hardware enthusiast* that does leave me sorely disappointed.
With all due respect to the time mHouston (and everyone else) have put into this project and how far they have brought it, when we are constantly teased about 'near nVidia speed on large proteins' .. '50% better performance on large proteins' and so on, and we do not actually get to fold any of said large proteins, together with the fact that it has been a long time since the HD4xxx series launch and we have yet to see any client optimizations for it (being told it could run faster given the right code), it does leave me a sad panda.
Catch 22: HD 4000 and 5000 can't fold fast because code is compatible with 3000 and 2000 which couldn't finish on time .
People ask Panda are PG ready for HD 5000, and Panda have never done anything for 4000 (near 2 years), joke.
Two years ago nVidia has taken a lead, and ATi still has stability issues on all systems across 32 and 64 bits with nobody knows for sure environment settings and experimental dlls. Volunteers looking for OpenCL as a remedy but they don't know it is only for 4000 and 5000 series cards (even I'm not sure about 4xxx), and brook+ is eventually dropped.
I understand You lads, You have a hope. I had it as well. Two weeks after 4870 show day I bought it. I thought it would be a speed demon. It made for me less then 4000PPD and my old 3870 other 2000PPD... Full of harms 6k, drivers, scripts, variables...
Two weeks ago I sold them, got 300e (lucky), bought 3 x GTS 250 for 95e each. Now I have 18000PPD and no myths about large proteins, next generation of GPU client, OpenCL and others.
Just have not more patience to wait next 2 years for a beta client for a mature OpenCL API. Don't like waste my money
Sorry, lads I know You working hard but at this ground I see no results.
Qua invece le dichiarazioni sui progetti per il prossimo futuro.
September 25, 2009
Update on new FAH cores and clients
E questo è più o meno tutto sulle novità 58xx e Folding@Home.