[Mini News] AMD launches Radeon Software 18.7.1 with Up To 28% faster Performance for RX Vega 56 in Earthfall

[Mira Yurizaki]

I'm curious as to why drivers need game-specific fixes despite there being at least two layers of abstraction between it and the application.

 

EDIT: And if it's not obvious, I'm poking at everything, not AMD specifically.

[Shadestones]

On 7/23/2018 at 8:14 AM, LAwLz said:

What person are you talking about?

 

Anyway, there was no award. Not sure where that misconception is coming from.

It was just a report from a consulting firm hired by AMD.

 

What happened was that AMD posted a test of a single program, done by a third party company AMD paid to do the test.

For all we know, they could have contacted multiple firms, and gotten 10 reports saying Nvidia drivers were more stable, but they decided to only post the one that showed AMD on top.

Or maybe they requested specific driver versions (for example AMD requested Nvidia drivers they knew had problems in that particular program, and requested using AMD drivers that they knew didn't have as many problems).

 

Since the test was ordered and published by AMD, it should not trusted.

Hell, even if it was trustworthy, it would still just be a single driver version tested, using a single program.

You can't judge overall driver quality by just doing one test on one driver release.

The fact that it was commissioned by AMD does not immediately discredit the results of the test.

Here is a link to the Radeon article talking about what exactly was being tested

 

An excerpt from the article:
"The team at QA Consultants ran each of the systems through multiple instances of CRASH, a four-hour automated program in Microsoft’s Hardware Lab Kit (“HLK”). CRASH contains a variety of graphical functions across DirectX 9, 10 and 11 including changes in resolution, color settings, screen rotations, color overlays, sleeping and waking up."

 

Microsoft's Documentation: "This automated test verifies system stability over a specified time. If the system does not stop responding or encounter a bug check during this specified time, the test is considered a pass."

 

It is interesting to know that QA Consultants, the company that ministered this test, is NOT partnered with AMD. You can find a list of their partners on their website's homepage.

 

The truth is that they're not really far off from each other. NVidia's results came in at 82% pass rating, and AMD's was 93%.

[LAwLz]

6 hours ago, Shadestones said:

The fact that it was commissioned by AMD does not immediately discredit the results of the test.

It being commissioned and published by AMD does discredit it quite a bit though.

 

6 hours ago, Shadestones said:

Here is a link to the Radeon article talking about what exactly was being tested

I've already read the full report.

 

6 hours ago, Shadestones said:

It is interesting to know that QA Consultants, the company that ministered this test, is NOT partnered with AMD. You can find a list of their partners on their website's homepage.

Doesn't matter.

 

My points were these:

1) AMD could have tired multiple companies to do this type of testing, and then buried all the results where they lost. They might have gotten reports from 10 different companies, where all but 1 showed them being worse as far as driver stability goes. This report was published and the other 9 were put in the garbage. The reports are not public unless AMD publishes them.

 

2) Driver stability can vary from version to version quite dramatically. You can not generalize all drivers based on the result of a single one.

The results might have looked very differently if the test was done today, or a month earlier than it was.

 

3) It was just one program tested, and only a crash was reported as failing the test.

A single program is not representative of all programs. CRASH isn't even representative of common programs, so using that to generalize to apply to all programs is not a good idea to say the least. For all we know, the AMD cards could have been glitching out in several ways, but since it was not a full system crash it counts as a pass. Wrong colors? Mouse cursor locking up? Screen flipped upside down? None of that would matter, because they were only measuring system crashes. The same can be said for the Nvidia results by the way.

 

 

At best, that report shows that if you're going to be running CRASH for 3 days straight, using drivers from specifically mid-May then an AMD card has a slightly lower risk of crashing.

Anyone who thinks this is a valid test for generalizing driver stability as a whole are naive or full of wishful thinking.

 

 

Articles like this one are (OpenGL drivers Hall of Fame/Shame) far, FAR better indicator of driver quality, because they are compiled over a long period of time, are detailed and also includes areas such as developers contacting the companies and how they respond.

Please note that this article is now 5 years old and should not be used as an indicator of general driver quality in 2018 either. It is just an example of what type of article you should look for, rather than one test, using one driver, for one program, published by a source that has a conflict of interests.