NVIDIA Slaughters AMD Catalyst On Linux In OpenGL 4.x Micro-Benchmarks

[Decebal]

Agree

Hi,

I am just going to be the ugly kid in the corner with the humpback and the protruding eyes who has not learned to run with the pack, and never will.

The benchmark is a great opportunity! One to learn from how to write better code and to measure existing code so it can be improved further. And it is of use to everyone - for Nvidia, for AMD, for Intel, for the Open Source community, any developer really. And what leads to better code leads to happier end users.

I will now go back to my corner and let the experts name and shame the heck out of the benchmark and the results. I do know why you are really doing it (so you can make sense of it ).

I'll just do the same.

[oleid]

I still have an old notebook with an AMD/ATI card and it suffers clipping, lags etc. no workaround from any version of drivers in years !

Install a distribution that includes recent drivers: mesa+kernel (e.g. Antergos). That should be all you need. Even the tiny GPUs in AMDs APUs work great for me using either the OSS or proprietary driver.

[Passso]

The last time I bought an AMD/ATI card was in 2011, I spent dozens of hours trying to find out why it was buggy and slower than my older Nvidia card of 2008... and finally I sent it back to reseller and buy an NVidia again...
I still have an old notebook with an AMD/ATI card and it suffers clipping, lags etc. no workaround from any version of drivers in years !

Open source drivers or not, those AMD/ATI hardwares are now banished in my home and business as I only use Linux.

[Panix]

I have a strong feeling that these AMD driver problems are only with the newest cards (R7, R9) and the high end HD7xxx series. I got HD7770 myself and it works excellent, but a friend of mine has HD7950 and it keeps on crashing, working slowly even to a point where the desktop isn't much usable.

Michael, you should try testing one of the older mid range Radeon GPUs that are known to work ok on Catalyst, not only the newest/fastest ones that are known to be not implemented properly. Most people don't buy 500? GPUs anyway.

In other words, DON'T buy a newer AMD gpu if you use Linux? Neither Foss, nor Catalyst helps? AMD's support is sub par. Maxwell and Kepler cards have at least binary drivers. Newer or most recent AMD cards have neither, it sounds like.

[DeiF]

I posted this in another thread, but it fits here better:

Graham Sellers (AMD's OpenGL guy) said this in a recent tweet (about the Phoronix article with the benchmark):

Point completely missed, I guess. This was never about vendor vs. vendor, but about technique vs. technique.
My lesson learned... "Don't tell developers how to optimize their applications unless you also beat NV."

Full conversation: https://twitter.com/thatjimblack/sta...15058634014720

[tuubi]

Public call:
Is an experienced AMD developer out there that could write a similar test suite with the aim to utterly destroy NVIDIA?

... we could then ask Michael Larabel to include this new test into PTS and perhaps get a more balanced headline in phoronix.com. (well - it will *not* be more balanced, it will just add a counter-weight)

This has been said before, but the aim of these benchmarks was to highlight codepaths needed for low-overhead opengl. They point out a set of functionality for driver developers to implement and optimise. Game developers have been very vocal about the need for better opengl implementations across the field, and it's helpful if the driver teams focus on what is actually useful. Intel and AMD are already aware of the way opengl is developing, so I'm sure it's only a matter of priorities and resources.

It would be childish, not balanced to include benchmarks intended to expose weaknesses of one specific vendor with the aim to "destroy" them. This particular set is obviously not out to assassinate anyone.

[lowflyer]

... to get this discussion a bit more balanced.

Public call:
Is an experienced AMD developer out there that could write a similar test suite with the aim to utterly destroy NVIDIA?

... we could then ask Michael Larabel to include this new test into PTS and perhaps get a more balanced headline in phoronix.com. (well - it will *not* be more balanced, it will just add a counter-weight)

[Cyber Killer]

I have a strong feeling that these AMD driver problems are only with the newest cards (R7, R9) and the high end HD7xxx series. I got HD7770 myself and it works excellent, but a friend of mine has HD7950 and it keeps on crashing, working slowly even to a point where the desktop isn't much usable.

Michael, you should try testing one of the older mid range Radeon GPUs that are known to work ok on Catalyst, not only the newest/fastest ones that are known to be not implemented properly. Most people don't buy 500? GPUs anyway.

[deppman]

I am tired of appologizing for AMD

You know I felt the same when I upgraded from a GT 630 to an R7 260X starting of the year. However I soon realized that it's got a lot to do with the games as well. I have all of the source games from valve and they work exceptionally well on my AMD card but when I run games from other vendors (Metro LL for example) they seem to be crap. Instead of people being so quick on pointing figures at AMD, should instead try and answer the question, why do games from valve span so perfectly on the entire GPU spectrum while most games from other vendors behave horribly? Valve clearly took their time to port their games and it clearly shows! I can say the same thing about games build using unity.

I am tired of appologizing for AMD. Valve and Unity have the influence to tell AMD to "please fix your @#$! POS driver! Its broken right here" and get results. They also almost certainly get direct support from AMD to help work around their other deficiencies.

Smaller devs probably don't get the same level of support - be it API fixes or hardware for testing or development. Nvidia is famously good for providing support to even small vendors, even if that just means providing a free Shield or 750Ti to test against. Look how Michael has to buy all the AMD cards he tests, whereas all the NVidia provides test hardware, often on the first day shipping. Which one seems most committed to success?

Providing the best driver experience is a complex task that requires commitment on many levels. A good start is to provide consistent, high performance, stable drivers. Another important facet is providing support to the developers that use those drivers. AMD has proven inferior on both fronts. How is this not their fault?

[przemoli]

"This micro benchmark its thus irrelevant!"

This code represent (as official, and anything short of specification release can us get for OpenGL) FAST & RELIABLE alternative to DX12/Mantle.

If its not fast nor reliable, then something is wrong. Thus phoronix.com results represent valid and important data point to the GPU vendors (and game devs, and Valve which will probably start swinging their whips again at GPU vendors - if they haven't already).

Also since now we get that test integrated into PTS and thus openbenchamarking, game devs can get idea what works and what not.