GPU Test System Update May 2019

[W1zzard]

See my comments inline

Res Evil 2 - using Devil May Cry instead with same engine
Forza Horizon 4 - Uses MS Store, which I find horrible to use
WWZ - Has no single player, offline mode is with bots only, so can't be reliably tested
The Division2 - Multplayer, always online, patches will ruin all data, and can't control what other people do on-screen, so not repeatable
Anthem - same as division 2
Dirt rally 2 - same engine as F1 2018
New Dawn - didn't see much reason to pick this over regular Far Cry 5
Apex legends - same as division 2

How come a 2016 game and a 2017 game are too old while a 2016 and two 2015 games are not?

Depends on the game, too. Civ 6 was updated with better DX12 support in 2019, Rainbow 6 represents an eSports title, Witcher 3 is kind of like a reference

 

[Vayra86]

Are not all games included in the final average?
VII was an example, like all the line up amd

I agree that based on the titles and the games we see that gpu choose, but this is not the case, at least in Italy.

And in Italy the average tpu is very popular...

Well then they should use TPU's reviews to get smarter about it and start to actually read them, don't they?

You cannot make the least interested readers the norm to write a review on. If you do that, the review becomes effectively worthless. What does that say about a lot of other reviewers... yep. They're not very useful. Another popular statement is that 'you need to test with settings most people use' - another useless consideration for a GPU review that is meant to show what hardware can do - the largest market share for GPUs is midrange so that'd mean every review would test medium/high settings... Its the same as 'testing' a Ferrari by going out for groceries.

Mainstream audience/reader has a habit of too short attention span to look at details. I see a lot of that in this topic (not you) and its good to see and recognize that - but its not good to tailor a review to that group. It is THIS group that needs to be educated, not the other way around.

 

[gridracedriver]

See my comments inline




Depends on the game, too. Civ 6 was updated with better DX12 support in 2019, Rainbow 6 represents an eSports title, Witcher 3 is kind of like a reference

Thank you

I have another question:
the cards used are reference for amd and FE for nvidia?

So it's like we're comparing the amd reference blowers against cards that are already at custom nvidia level?

Well then they should use TPU's reviews to get smarter about it and start to actually read them, don't they?

You cannot make the least interested readers the norm to write a review on. If you do that, the review becomes effectively worthless. What does that say about a lot of other reviewers... yep. They're not very useful. Another popular statement is that 'you need to test with settings most people use' - another useless consideration for a GPU review that is meant to show what hardware can do - the largest market share for GPUs is midrange so that'd mean every review would test medium/high settings... Its the same as 'testing' a Ferrari by going out for groceries.

Mainstream audience/reader has a habit of too short attention span to look at details. I see a lot of that in this topic (not you) and its good to see and recognize that - but its not good to tailor a review to that group. It is THIS group that needs to be educated, not the other way around.

in fact it is the main problem, practically nobody who frequtenta the Italian forums reads your reviews in the single game or in the other tests, but they read directly the final average and all for a denigratory purpose only.

Unfortunately, the Italian mean boy is not very intelligent.

PS. I'm italian

 

[londiste]

So it's like we're comparing the amd reference blowers against cards that are already at custom nvidia level?

Fair point. It is not ideal but I don't think there are good options with this for a reviewer though. Nvidia sends out FE cards and the 5% difference in clock speed does not turn out to be as much as it seems. Also note that this is spec for Boost Clock that is inherently variable based on temperature and power.

Looking at the reviews, actual achieved clock speed is much more a function of the power limit than anything else. Because it made me curious, checked the TPU RTX 2080 reviews for details.
Format: Card model, Boost Clock spec - Average and Median from testing - Performance difference with FE in % (1080p/1440p/2160p respectively):
- NVIDIA GeForce RTX 2080 Founders Edition, 1800MHz - Avg 1897MHz, Median 1905MHz
- Palit GeForce RTX 2080 Gaming Pro OC, 1815MHz - Avg 1920MHz, Median 1935MHz - 1/1/1%
- Gigabyte GeForce RTX 2080 Gaming OC, 1815MHz - Avg 1885MHz, Median 1935MHz - 1/2/2%
- Palit GeForce RTX 2080 Super Jetstream, 1860MHz - Avg 1924MHz, Median 1920MHz - 1/2/2%
- ASUS GeForce RTX 2080 STRIX OC, 1860MHz - Avg 1931MHz, Median 1980MHz - 2/3/3%
- MSI GeForce RTX 2080 Gaming X Trio, 1860MHz - Avg 1969MHz, Median 1965MHz - 2/3/4%

When it comes to the experience of someone actually buying a card, looking at RTX2080 listed in TPU GPU database (which should be pretty comprehensive): https://www.techpowerup.com/gpu-specs/geforce-rtx-2080.c3224 - Out of 147 listed cards only 55 (37%) are at base spec and 20 (14%) more have less than 1800MHz in spec.

I don't think using FE is much of an issue in practice.

 

[W1zzard]

Fair point. It is not ideal but I don't think there are good options with this for a reviewer though. Nvidia sends out FE cards and the 5% difference in clock speed does not turn out to be as much as it seems. Also note that this is spec for Boost Clock that is inherently variable based on temperature and power.

I don't think using FE is much of an issue in practice.

That

Out of 147 listed cards only 55 (37%) are at base spec and 20 (14%) more have less than 1800MHz in spec.

Out of those 20, most will have increased power limits, and the different cooler will affect boost clocks as well. There seems to be no way to really get a "NVIDIA reference RTX" card

 

[rtwjunkie]

How come a 2016 game and a 2017 game are too old while a 2016 and two 2015 games are not?

Well one of the games, (TW3) is still a good test for many GPU’s. Also, as W1zzard already explained earlier in the thread, it is one of the highest visited pages of his reviews.

He is able to see how many hits each page has. Most people don’t go to each and every page. They pick and choose how games that affect them score with the GPU’s.

Finally the problem is that to evaluate a card and therefore the purchase, the user looks at the table of averages and not the single game.

Actually, as I just mentioned, users mostly go to the games pages that they play or that interest them. W1z is able to see how many hits each page has. If the majority was to the performance summary, then that would be all he posted. But it’s not.

AC7 is a classic example of UE4 biased, a stock engine that runs bad on amd.

You sure about that? As far as I know it is not UE4, and was developed with AMD assistance, not Nvidia. Nice try though.

EDIT: I did some research. It uses AnvilNext, a proprietary engine developed by Ubisoft Montreal. No UE4 involved.

 

[Vayra86]

in fact it is the main problem, practically nobody who frequtenta the Italian forums reads your reviews in the single game or in the other tests, but they read directly the final average and all for a denigratory purpose only.

Unfortunately, the Italian mean boy is not very intelligent.

PS. I'm italian

If it makes you feel any better - that behavior is not unique to Italians. Many people and not just even gamers are way too quick about reading information. Its a symptom of our time maybe? Too much information, too little time. But... I think it is still important not to cater to that very shallow standard and I am very happy TPU / W1zzard keeps offering that.

As for your statement about testing Nvidia FE versus AMD reference yes you are absolutely correct and it is far from ideal. But at the same time, isn't this AMD's fault as well? They tend to be very late with the release of custom models the past few years, and reviewers will want to include AMD cards that are just released into their testing. There is also the unique case of Radeon VII that ONLY gets a reference (but open air, luckily) cooler. The same stuff happens with AMD driver updates. TPU gets complaints that its still reviewing on an older driver, but there is no rhyme or reason to how often AMD releases one. With Nvidia, both aspects have a very basic frequency; you just know game ready drivers pop up all the time and that a new branch number means its a bigger update; you also just know that there are non-reference cards at launch.

@W1zzard Wrt AMD reference versus Nvidia 'FE OC' results...from Turing onwards the review should really get a disclaimer for that, and I think its safe to improve AMD results by 5% across the board to get a better picture of relative performance. Not saying you should manipulate the charts, but a bold text above them isn't all that bad. Another option is mentioning the fact that Turing's FE's are higher TDP limited cards at the start of the review, and giving all Turing FE's in each chart the 'FE' moniker behind each GPU modelname.

Its important because a large number of users, especially with the price of cards going up, will resort to absolute crap blowers to soften the financial blow. They could easily end up half or a full tier lower than they had expected - Turing's performance spread per tier is pretty tight as it is. And SUPER will probably make it even tighter.

Another option might be to just stick to a 'blower versus blower' comparison, and use a non-ref Nvidia blower, but that kinda detracts from the WYSIWYG idea of your review and it being 'Nvidia ref stock versus AMD ref stock' at a basic level.

 

[rtwjunkie]

I have another question:
the cards used are reference for amd and FE for nvidia?

All older cards other than the one that is the purpose of the review are reference/FE. There are simply too many variations to adequately keep all of them on hand and test all of them. W1z has a lot more to his life and work than testing 150 video cards in 20 games every time a new one comes out. Reference/FE is the only reasonable solution.

If you want to see how a particular model did, go back and look at the review when it came out. You can get a good idea comparing the two to see how yours would do now.

 

[Tatty_Two]

I wonder why

The attitude and attempted baiting that you have displayed is probably why, so it's probably a good idea for you to take a break from this thread.

 

[gridracedriver]

Well one of the games, (TW3) is still a good test for many GPU’s. Also, as W1zzard already explained earlier in the thread, it is one of the highest visited pages of his reviews.

He is able to see how many hits each page has. Most people don’t go to each and every page. They pick and choose how games that affect them score with the GPU’s.


Actually, as I just mentioned, users mostly go to the games pages that they play or that interest them. W1z is able to see how many hits each page has. If the majority was to the performance summary, then that would be all he posted. But it’s not.


You sure about that? As far as I know it is not UE4, and was developed with AMD assistance, not Nvidia. Nice try though.

EDIT: I did some research. It uses AnvilNext, a proprietary engine developed by Ubisoft Montreal. No UE4 involved.

TW3 will also be a test still heavy, as it was crysis, but I don't understand why to use 3-4 year old games, and not only TPU, when the reason to buy a video card today is to play current games or 1 year old at the most.

Ace Combat 7 directly uses unreal engine 4 SDK, so nothing optimization for amd cards, no AnvilNext.

If it makes you feel any better - that behavior is not unique to Italians. Many people and not just even gamers are way too quick about reading information. Its a symptom of our time maybe? Too much information, too little time. But... I think it is still important not to cater to that very shallow standard and I am very happy TPU / W1zzard keeps offering that.

As for your statement about testing Nvidia FE versus AMD reference yes you are absolutely correct and it is far from ideal. But at the same time, isn't this AMD's fault as well? They tend to be very late with the release of custom models the past few years, and reviewers will want to include AMD cards that are just released into their testing. There is also the unique case of Radeon VII that ONLY gets a reference (but open air, luckily) cooler. The same stuff happens with AMD driver updates. TPU gets complaints that its still reviewing on an older driver, but there is no rhyme or reason to how often AMD releases one. With Nvidia, both aspects have a very basic frequency; you just know game ready drivers pop up all the time and that a new branch number means its a bigger update; you also just know that there are non-reference cards at launch.

@W1zzard Wrt AMD reference versus Nvidia 'FE OC' results...from Turing onwards the review should really get a disclaimer for that, and I think its safe to improve AMD results by 5% across the board to get a better picture of relative performance. Not saying you should manipulate the charts, but a bold text above them isn't all that bad. Another option is mentioning the fact that Turing's FE's are higher TDP limited cards at the start of the review, and giving all Turing FE's in each chart the 'FE' moniker behind each GPU modelname.

Its important because a large number of users, especially with the price of cards going up, will resort to absolute crap blowers to soften the financial blow. They could easily end up half or a full tier lower than they had expected - Turing's performance spread per tier is pretty tight as it is. And SUPER will probably make it even tighter.

Another option might be to just stick to a 'blower versus blower' comparison, and use a non-ref Nvidia blower, but that kinda detracts from the WYSIWYG idea of your review and it being 'Nvidia ref stock versus AMD ref stock' at a basic level.

Thanks for the answers

 

[rtwjunkie]

TW3 will also be a test still heavy, as it was crysis, but I don't understand why to use 3-4 year old games, and not only TPU, when the reason to buy a video card today is to play current games or 1 year old at the most.

And....based on your response you ignored what I said about W1zzard’s primary reason for using it: it is one of the most popular pages in every review he does.

Also, since when is there an arbitrary number of 1 year old games at the most for buying a new video card? What kind of nonsense is that? There are many reasons for buying a new card, and a number of them certainly do involve games older than 1 year.

 

[Vayra86]

TW3 will also be a test still heavy, as it was crysis, but I don't understand why to use 3-4 year old games, and not only TPU, when the reason to buy a video card today is to play current games or 1 year old at the most.

For comparison purposes. If you keep updating the selection of games to what is recent, you can never compare between reviews spaced apart more than a few months. That means there are situations where you couldn't even compare cards within the same generation in a good way for some games. It kinda defeats the point of standardized testing. Its always a pick and choose what games end up in a bench suite for a long time, but there is always going to be a small selection of them, and they're usually the games that cripple systems when they launch. TW3 and Crysis fit the bill perfectly. Popularity does help of course. But that mostly concerns popularity of the actual page with the test on it, not the game itself.

Note also that its impossible to pick the games everyone wants to see all the time, just like its impossible to cover all settings and resolutions. So choices must always be made, and its almost always a choice of evils because you'd preferably show as much as possible. So, the games are picked that offer the most interesting and informative set of numbers.

With this knowledge, maybe it makes it easier to think of good suggestions because that is what the topic's for!

 

[P4-630]

Do you still use the same hardware in 2020? (I googled a 2020 setup but could not find it)

(https://www.techpowerup.com/review/gpu-review-system-update-may-2019/)

What is the ambient/room temp you test the GPU's in? (IIRC you used to mention this in the GPU reviews but left it out recently? Around 22 degrees C?)
I don't see you mentioning any case, open bench?

 

[W1zzard]

The 2020 setup is very similar, check the most recent GPU review. Card installed in case, one side open

 

[P4-630]

The 2020 setup is very similar, check the most recent GPU review. Card installed in case, one side open

Ok I see

What about ambient/room temp?

 

[W1zzard]

What about ambient/room temp?

22-23°C.. nothing worth mentioning, especially considering production variances are between coolers/GPUs/TIM