NVIDIA 2010 Driver Year In Review

[kazetsukai]

I've been using nothing but nvidia since the 6000 series and I'm fairly certain you don't get performance improvements 6 months after the release of a new generation. Maybe for SLI, but for single cards setup I've never noticed anything. Which is ok, it just means we get proper support asap, not 2 years after we buy the card.

Except in this case all of Gen6 is very similar architecture, so they all will likely benefit from enhancements to the driver. They're all on DDR3, they all use very similar CUDA cores.

[bug77]

All the drivers used permit "generations" 4-6, and the later drivers permit Gen7 (Fermi). I think any of Gen6 would be valid tests, while anything Gen5 or lower probably isn't being optimized anymore.

I've been using nothing but nvidia since the 6000 series and I'm fairly certain you don't get performance improvements 6 months after the release of a new generation. Maybe for SLI, but for single cards setup I've never noticed anything. Which is ok, it just means we get proper support asap, not 2 years after we buy the card.

[kazetsukai]

Shouldn't that be:

1: Geforce 2 series
2: Geforce 3,4 series (IIRC geforce3 is more close to 4 than to geforce2)
3: Geforce FX 5000 series

?

You're quite correct, oops. Its been a while.

[devius]

1: GeForce 200,300 Series

2: GeForce 400 Series

3: GeForce 5000 Series (There were some PCI Express variants of the 5k series)

Shouldn't that be:

1: Geforce 2 series
2: Geforce 3,4 series (IIRC geforce3 is more close to 4 than to geforce2)
3: Geforce FX 5000 series

?

[kazetsukai]

Time permitting, maybe stick to the drivers that do support Fermi and retest in a future article?

Testing on the 9800GTX was actually about as valid as you're going to get. I would fit NVIDIA architecture into the following "generations":

1: GeForce 200,300 Series

2: GeForce 400 Series

3: GeForce 5000 Series (There were some PCI Express variants of the 5k series)

4: GeForce 6000 Series (SLI was introduced, PCI Express became much more mainstream)

5: GeForce 7000 Series (After this series AGP was more or less phased out)

6: GeForce 8000, 9000, 200 Series

7: Fermi

All the drivers used permit "generations" 4-6, and the later drivers permit Gen7 (Fermi). I think any of Gen6 would be valid tests, while anything Gen5 or lower probably isn't being optimized anymore.

[Havner]

"In this game, which is more demanding than Warsow/OpenArena, there was also not the performance regression found between the 195.36.15 and 196.36.24 driver updates."

But the regression found for earlier tests was found between 195.30 and 196.36.15. 195.30 wasn't tested for ETQW and later tests.

[bug77]

Fermi doesn't work with all of the driver releases tested...

Yes, I forgot about that. How could it, it wasn't even released one year ago.

Time permitting, maybe stick to the drivers that do support Fermi and retest in a future article?

[Michael]

The test is a bit flawed. If there were any performance improvements in nvidia drivers, I don't expect them to apply to the 9 series. Hell, they're probably not optimizing for the 200 series anymore.
In short, you should have used a GTX400-something instead of the 9800GTX.

Fermi doesn't work with all of the driver releases tested...

[bug77]

The test is a bit flawed. If there were any performance improvements in nvidia drivers, I don't expect them to apply to the 9 series. Hell, they're probably not optimizing for the 200 series anymore.
In short, you should have used a GTX400-something instead of the 9800GTX.

[kbios]

Phoronix: NVIDIA 2010 Driver Year In Review

At the end of each year for the past five years we have delivered "year in review" articles looking at the performance of NVIDIA's (and ATI/AMD's) proprietary Linux drivers. Both in terms of new features introduced during the year in their driver updates and benchmarking the driver releases to see how the performance has evolved over twelve months. With 2010 coming to an end, it is time for this year's driver reviews. We are starting this year seeing how the NVIDIA performance has matured in 2010.

http://www.phoronix.com/vr.php?view=15558

I think that the strange VDPAU spike in 269.19.21 is caused by a bug that affected the 260 series sice the beginning and that caused a massive amount of ram to be allocated on VDPAU init. In fact VDPAU was unusuable on relatively low-memory systems (1GB) until 260.19.26. Details are on nvnews.