Discussion in 'Article Discussion' started by bit-tech, 17 Jul 2018.
Didn't see that coming.
didnt read the rest of the article after seeing that and skipped straight to the comments. 1 way to invalidate their study instantly eh?
I mean, I've looked at the report and I can't easily fault their methodology. The fact it was commissioned by AMD doesn't really mean anything, bar the fact that if it had shown Nvidia in the lead it would have been quietly buried instead of being trumpeted out with a press release. (And if it had been commissioned by Nvidia but shown AMD in the lead, the same would have happened.)
(Yes, I know I say there'll probably be a refutation study from Nvidia - but I'm not saying AMD's study is wrong, just that Nvidia will either tweak its drivers or tweak the workload until it's on top - and if it means Nvidia's drivers become more stable in the process, that's by no means a bad thing.)
They tested the best case scenario rather than anything resembling real world use, so the conclusions they draw don't necessarily transfer out of the lab.
Oh, aye - hence the my comment that Nvidia might just tweak the workload by finding something AMD's not so great at and going "NUH-UH WE'RE BETTERER" - but that's true across the board: real-world reliability might be lower for AMD, but it might also be higher - and the same is true for Nvidia. All the report really says is "AMD is more stable for this particular workload" - but then that's true for every test ever, because you can't test everything.
Indeed, especially since gpu drivers have ballooned massively beyond their original scope and now incorporate everything from hw monitoring, streaming, advertising and so on rather than just being drivers, so the number of potential weakpoints has also increased and that is never going to be good for reliability.
Plus how nicely it plays with the effectively infinite selection of 3rd party software would be nowhere near practical to try to replicate in a lab, so YMMV comes into play fast and hard.
if this leads to a war on STABILITY and not PERFORMANCE... I for one will be so blummin' glad
In everything, I'd rather lose a few % performance to get rock solid stability. Period.
I run systems with both, and both in Crossfire/SLi and I've always been confused about the bad comments about AMDs drivers. I've had far more issues with nVidia drivers than AMD.
Not only that, but when there is a problem with AMD it does get patched out (admittedly it could be done quicker) where I've had issues with my 1080 SLi setup since almost release that has never been resolved.
In a report I conducted in my house I was found to be the biggliest best.
Those conducted by my lass and cat found otherwise but they used different testing methodology.
See, twenty years ago, AMD drivers were the crappiest of crap and nVidia didn't ask your immortal soul as part of the purchase.
Times change, opinions don't, especially when loud fanboys have reason to perpetuate old tales.
Anecdotally (though this is a few thousand deployed anecdotes and I get to sample the ones with issues) with Quadros and Radeon Pros (all the FirePros are long gone) I've had far more grief with the Radeon Pros even with their smaller install base, to the point where we no longer even bother trying to troubleshoot and remedy driver issues and just wipe and reimage the box by default: it'll be far faster, and inevitably end up being what's needed anyway. At least when some ancient NVS driver decides to crap itself an in-place reinstall does the trick (outside of an actual card fault).
AMD, NVidia...... Six of one, half a dozen of the other.....
All better than Intel when it comes to graphics drivers.
I seem to remember back in the day I used Omega drivers, rather than official ATi drivers.
Though admittedly that was more to do with ATi insisting I get mobility GPU drivers from my laptop manufacturer (who had gone bust).
So AMD make a test setting out exactly what to test, what cards to use, what drivers too use, what pc build, what os/software, and when the test is done (i.e. they can wait till they fix all the bugs) - how did they not get a 100% pass?
Where did you see that AMD developed the test? According to the report, QA Consultants came up with the test methodology independently and used a Microsoft test suite:
Which is all well and good, but unless they also do some 'real world' testing it is as vulnerable to "design to test" (or outright malfeasence) as any other standardised test. e.g. an image/video encoder that optimised for PSNR (or SSIM) can look like crap compared to a psy-optimised encoder.
According to what AMD commissioned - AMD knew exactly what was being tested, they are the ones paying the bills, it's their rules. AMD would have known what cards, and what MS test suit was being used in plenty of time, they should have been able to pass that test 100%.
Incidentally what do you think would have happened if the test came back and AMD looked bad, would we be reading about it now? Err, no, they would they have been told to wait a while and then try again with these new drivers, new rev of card with fixes and keep going till that particular test looks good. If worst came to the worst it would have been quietly brushed under the carpet and none of us would have known, perhaps it has been several times?
"There are three kinds of lies: lies, damned lies, and statistics."
Separate names with a comma.