"Ask ATI" dev thread

[energyman]

just get a unbroken X and 2d&3d will be fine. Performance is ok - for some strange reasons.

And IT MUST BE DONE BY ROOT BECAUSE SYSTEM FILES ARE CHANGED.

Just because ubuntu is fucked up and ubuntards too stupid to type, does not mean that the rest should suffer. Or must.

[lem79]

What is so horrible with the current driver?

My brief experience with Catalyst (9.4 and 9.5) was disappointing, especially with Compiz. Alpha-blur not working, crashes/various instabilities when other OpenGL apps would run at the same time, no vsync, video tearing. The Compiz issues were "show stoppers" for me, I like my eye candy :P

nVidia's Linux driver is definitely awesome.

[vrodic]

Some time has passed, 5xxx series is here, so I have a couple of quick comments regarding this. The comments are not Linux driver specific, and I hope you don't mind.

1. Let's hope OpenCL support will be released sooner rather than later.

2. I'm looking for a replacement card for my 3850, and I hope ATI will release something with same low idle power usage and TDP, with a 256 bit memory bus. My card has a passive aluminum heat sink, and a 9cm 900 RPM fan mounted on top of it, and it stays cool and silent. John, which 5xxx should that card be? It's definitely not 58xx.

3. I'm hoping that problems with folding@home performance will be fixed:

http://www.hardwaresecrets.com/article/640/7

http://foldingforum.org/viewtopic.php?f=51&t=10442

"The primary architectural difference seems to be that Nvidia can store intermediate results in fast short-term memory (like a cache, but managed by the program rather than the hardware) while ATI cannot. For the ATI implementation it is quicker to repeat the calculations than to store to and retrieve from the GPU main memory."


If John could comment on any of this, that would be great.

[energyman]

5870 goes down to 30W at idle - that is less than your 3850

[vrodic]

5870 goes down to 30W at idle - that is less than your 3850

I don't know how much my card alone spends in idle. I only have the complete system numbers: 82 WATTS on a P35 E8200 system. 160W in peak load (Furmark 1.7.0 running under wine )

But still I don't want peak power usage to be more than my 3850 also. So it should be faster, easier to cool and cheaper to run. I guess maybe I'll need to wait for the "< 40 nm" process.

[energyman]

cards get biiger - so consumption will be never fall - less power used? more room for speed and a handfull of transistors.

Wait for the 5850 and have a look.

[vrodic]

cards get biiger - so consumption will be never fall - less power used? more room for speed and a handfull of transistors.

Wait for the 5850 and have a look.

I'd be very happy with a more power efficient, 40 nm version of a 3850, updated to be better suited for GPGPU work, with even more circuits for power and clock gating.

Consumption in peak loads falls when you shrink the process. When you build hardware that enables more efficient software to be written. Total consumption for solving a given problem also fallw when the software itself gets smarter and more efficient about using the hardware's resources. The whole chain should be strong.

Now I'll just shut up and wait for comments from John.

[bridgman]

I can't really comment on unreleased products, unfortunately.

My understanding is that the current F@H implementation uses essentially the same code for 6xx and 7xx parts, so it does not take advantage of LDS/GDS on the 7xx parts. I imagine that's where the discussion of "calculate twice vs store and re-use" comes from.

[Kano]

HD5 cards are released since yesterday.

[energyman]

I'd be very happy with a more power efficient, 40 nm version of a 3850, updated to be better suited for GPGPU work, with even more circuits for power and clock gating.

Consumption in peak loads falls when you shrink the process. When you build hardware that enables more efficient software to be written. Total consumption for solving a given problem also fallw when the software itself gets smarter and more efficient about using the hardware's resources. The whole chain should be strong.

Now I'll just shut up and wait for comments from John.

and that will never happen. Why should AMD waste money on a dead product? Also a structure shrink involves a lot of work. The masks have to be redone: expensive. But not enough! The archtecture of the chip has to be redone in timing and power critical areas - which is expensive and time consumung.

Why wasting all that money, if there is a nice, new card that does not draw significant more power - for a lot more performance? (aka 5850).