Ieldra
I Promise to RTFM
- Joined
- Mar 28, 2016
- Messages
- 3,539
Follow along with the video below to see how to install our site as a web app on your home screen.
Note: This feature may not be available in some browsers.
A little independent testing shows that at least some of these tests weren't actually run at all, for any of the three cards. Some of the games (Doom/Vulkan @ 4k as a prime example) *none* of the cards perform that poorly in actual use.
Because I have access to two of those cards, with a weaker core system, and both the 1070 and Fury X perform better at 4k maxed than what is shown in the listed article. Substantially better. There is no way they actually ran those tests with those drivers on that test system and got the results they posted. Not without something being *very* wrong with the PC itself.
Edit: this is specifically in response to the Doom query.
So, by your own argument in this specific case, because nothing can be reproduced by settings alone, the linked article you posted (for purposes of discussion on Doom) is completely worthless and non-representative of any type of gameplay. Good to know. So the rest of the verified items that *weren't* Doom. Are those allowed?
That's a straw man, and many here know how dull I find them.
I'll entertain it though.
What I said was that the results cannot be compared to other sources unless they test a very similar run. This is not intended to be performance review of a game, the purpose of collecting this data is to compare the cards tested.
It means precisely fuck all that in your test with a fury x you get X change in average framerate.
If you had all the cards tested and performed the same benchmark run on all of them then we could compare relative performance of the cards in your data set to the one compiled by babeltech.
I don't know what "so the rest of the verified items that weren't doom." means, it reads like you wrote half a sentence then abandoned it.
I'm guessing what you're trying to say that is that since the doom numbers aren't representative of performance of the game as a whole, all the other results suffer from the same limitation thus are all invalidated. No, because the point isn't to represent performance of games, but to compare that of different cards.
No benchmark is representative of the game as a whole unless there's very little variation in the complexity of the different scenes.
Unless you have some actual evidence and justification for why this review contains flawed data your argument is null and void.
Nope. Just needed to see you write that the actual performance numbers of the cards themselves don't matter, just the relation of performance of equal loops or benchmarks performed with each card. Essentially, that it's not the absolute number for each card, it's the differences between the cards when run through the same tests. Am I understanding that correctly?
So, the point of the test is specifically determine if maturation of the drivers has significantly impacted performance of each card, solely. Meaning that because of the scope of the tests, it can't even be inferred that the same levels/loops were used for each card, unless the test itself is said to use a benchmark with a given settings list. Right?
Why are you talking about a settings list?
Man...
I'm Mr Babeltech.
I slot in a Fury X and install a driver from last year, I benchmark doom using a predetermined loop of my choosing. I install latest driver and repeat the test.
I slot in a 980ti, I install last year's driver, run the same loop I used on the Fury. Install latest driver, repeat.
Last year the fury X was 85% of a 980ti
Using latest drivers the fury X is 93% of a 980ti.
Fury X improved X%, 980ti improved Y%.
Conclusion.
??? Because being able to reproduce results is the definition of validating a test. You have to have the settings list to be able to verify that your own observations are in line. Say, if I screw up and don't max may FoV in my settings in DE:MD, then my test results wouldn't be comparable. I used Doom as an example of how things are messed up because, even in a full playthrough, I don't remember a point where a section yielded average framerates that tanked that far. Such incredibly low average framerates *across all three cards* has me wondering if that specific test actually took place. Without the definition of the section/loop, as you pointed out, it seems difficult to believe *any* of those quoted changes between driver revisions because the numbers used to establish the amount of change seem to be suspect. For the tests he performed that *did* use built-in benchmarks, some of those are also pretty significantly different that what can easily be reproduced.
LOL I'd trust WCCF over Babeltech any day. AMD black listed that site back in the 7970 days.
AMD blacklisted this site back in the few months ago days.
WCCF's posts are 99% rumors and speculation. Babeltech runs tests, I honestly see no reason why you assume they falsify their data
No. I have access to the same cards, with a weaker core system ([email protected]) and I get higher results. So unless we're willing to entertain that an 8320 is somehow superior to an i7-4790 in a gaming benchmark (which I don't think anyone, including myself, is), then something is wrong with the numbers being used to establish the degree of change.
As an example, a very easily reproducible example, the SoM numbers are from the benchmark tool included with the game. We know this because the in-game engine is capped at 100 FPS, so having the 980 ti and 1070 reporting a framerate higher than 100 FPS means that the benchmark tool was being used.
Now I suppose it's possible that the FuryX tests were performed with actual gameplay and that the NV tests were just benchmark tool runs, but now we're treading on trying to make excuses for changes in testing methods within the same game, although not necessarily within the same card's suites. At the very least, a note saying that some driver comparisons for a game were done with included benchmark tools for some cards, where others were done with actual gameplay. When it comes to rolling out reported performance changes by driver version for each card, those notes mean something.
Since the benchmark tool was used for the NV cards, I'll make the presumption that it was also used with the AMD card. A stock XFX Fury X using 16.11.5 drivers, paired with the [email protected], using all of the settings completely maxed in SoM's benchmark produces a result of 39.7 FPS @ 4k. At 1440p, it's 73.9 and at 1080p, it's 98.5. The numbers are substantially higher than the reported output above, with a substantially weaker core system behind the card.
I'm not saying they changed the settings between the tests. I'm doubting whether they actually ran the tests as they stated in the first place. At least some of them.
Edit: for typo on CPU in their test system.
So, by your own argument in this specific case, because nothing can be reproduced by settings alone, the linked article you posted (for purposes of discussion on Doom) is completely worthless and non-representative of any type of gameplay. Good to know. So the rest of the verified items that *weren't* Doom. Are those allowed?