Hi there,

I was wondering if it is planned to support Single Precision ATI cards (cards based on the RV810, RV830 or RV840 chip).

Since all new ATI cards (except the High-End ones) lack Double Precision it would be a good idea to write code supporting Single Precision. IMHO.

Because if you don't, you miss out on almost all the new ATI cards being sold these days - since most people go for the 5770 (because of it's really good price-performance ratio).

Regards,
F.

I think that the sort of calculations that are done here in Milkyway needs double precision cards for the accuracy required. However if code could be written to make use of single precision cards, then I would support it.

---

Don't drink water, that's the stuff that rusts pipes

Some apps were written and decided they weren't as good and weren't used.

---

Doesn't expecting the unexpected make the unexpected the expected?
If it makes sense, DON'T do it.

Some apps were written and decided they weren't as good and weren't used.

Of course it is possible to write SP code that computes the same results as DP code - it's just 2-4 times slower than native DP code. But still a, for example, ATI 5770 with 800 shaders, would be much much faster than every CPU - even with SP.

Since more and more people are using those 5xxx ATI cards (because they are really good value for money, because of Nvidias "drama" with the Fermi, etc.) it would be unwise to not support them.

But overall it depends if the project could use the extra crunching power that would be made available. I know that they are currently close to the edge from their server perspective and so they may not need the extra computing capability it would give them. So it may be 'unwise' to do it and just utilise the DP cards as is.

It's all a matter of conjecture whether it would wise or unwise.

So maybe they should spend a few bucks and upgrade their server?

No point when the GPU crunching power deals with the work they want the crunchers to deal with.

As was pointed out, the use of single precision taken to the accuracy the project requires slows the crunching time for each WU down by over 4x. This means there would be the possibility of some GPUs not meeting the deadline return times as they were so slow.

No! Keep with the recognised double precision GPUs from nVidia and ATI that already can be used. After all an HD3850 is not that expensive now, and the HD4850 are reasonably affordable as well.

---

Go away, I was asleep

You can see the crunching capacity for Milkyway compared with SETI here. Just look at the number of Users and Hosts, then compare the outputs of these small numbers compared with the very much larger numbers with SETI.

---

Go away, I was asleep