GPU Test System Update March 2021

[W1zzard]

This is one for the history books. We're changing our graphics card test system from Intel to an AMD Zen 3 Ryzen platform, with PCIe 4 and resizable BAR. In this article, we're sharing our thoughts on the test system configuration, the games we added and removed, and why. We also have more testing with raytracing, and improved the power measurements.

Show full review

 

[QUANTUMPHYSICS]

In every test that matters, the 3090 is the 4K Ray Tracing King of the Hill.

The real shame is that it's so hard to get the 3070, 3080 and 3090 without the scalping/ crypto currency nonsense scalping affecting supplies.

I am very fortunate I was able to get mine at MSRP.

 

[nguyen]

oh boy are you gonna test Ampere GPUs again once they have Resizable Bar support?

 

[Vecix6]

Raytracing graphs are really dificult to read and locate due some cards appears 3 times with diferent options, maybe separate graphs could help to understand the results..

 

[ZeppMan217]

After some googling I found our article, detailing the lack of UEFI support on the Radeon VII. So that was the problem, the card simply didn't have the BIOS update that AMD released after our article. Well, FML, the page with the BIOS update no longer exists on AMD's website.

Really? Someone on the web team made the decision "oh I'll just delete these pages that contain an important fix to get the product working", on a product that's not even two years old? (launched Feb 7 2019, page removed no later than Nov 8 2020).

Lucky I found the updated BIOS in our VGA BIOS collection, the card is working perfectly now.

Comedy gold.

 

[Dyatlov A]

I like this change, but that Ryzen processor doesn’t make whea errors with Infinity Fabric at 2000 MHz? To me the 5800x makes tons of whea already at 3600MHz in MSI B550i.

 

[napata]

Woudn't it have been more efficient to wait until the end of the month when Nvidia has rolled out their vBios for Resizable BAR to retest those games? Seems like a bit of double work.

 

[mechtech]

Awesome stuff..........but

"It allows us to kick out some of the older graphics cards, as well as game tests; making room for newer cards and games."

Noooooooooooooooooooooooooooooooooooooooo lol

W1zz you have to keep the RX480 in there

Borderlands 3 and Witcher 3 are the only games I have so that's good they are covered

Edit 2 - "Infinity Fabric at 2000 MHz, memory clocked at DDR4-4000, in 1:1 sync with the Infinity Fabric clock. Timings are at a standard CL19 config, that's easily found on affordable memory modules—spending extra for super tight timings usually is overkill and not worth it for the added performance."

I understand the need to maximize CPU power as to not limit graphic card potential, although I always found it a bit convoluted for people say running a dual core and expecting to get the fps listed with a new GPU. I guess with this new AMD test bench you can test CPUs gaming performance back to 1st gen Ryzen, or I guess the B550 doesn't support them?

 

[phanbuey]

Woudn't it have been more efficient to wait until the end of the month when Nvidia has rolled out their vBios for Resizable BAR to retest those games? Seems like a bit of double work.

You would want to know the difference tho, so it's kind of genius in a way

 

[ThrashZone]

Hi,
Welcome to 2021 too many bullshit lakes on intel side

 

[W1zzard]

oh boy are you gonna test Ampere GPUs again once they have Resizable Bar support?

Of course

I like this change, but that Ryzen processor doesn’t make whea errors with Infinity Fabric at 2000 MHz? To me the 5800x makes tons of whea already at 3600MHz in MSI B550i.

Got a bunch of them in Event Viewer, but no issues with stability. Not a single crash in two weeks of benchmarking

 

[XL-R8R]

In todays world of crazy pricing and lack of availability, seeing where old GPUs stack up against newer releases would certainly be of benefit - could the newer results not be combined with older ones, on a newly created page ("Page XX: Legacy Cards - How Do They Stack Up?")??? If so, this would be a nice addition.

Minimum (1% would be fine) and average FPS would also be a great addition to the bench results and drag them in-line with most e-publications.



Also, 2000 IF seems a tad bit high for a system thats meant to represent "normality" (and not some outrageously specced rig) when you consider most folks have issues getting true stability above 1850.

Running 2000 on a 1:1 is nice though

 

[ppn]

3050Ti should be the lowest card tested. Im really only interested in 3090/4070 and above.

 

[mb194dc]

Nicely done, that must have been a lot of work!

 

[ca_steve]

The power consumption over time graph is awesome! Also, thanks for the soft cap v-sync power consumption.

 

[hardcore_gamer]

That's good! Excellent!

It would be nice to have a combined relative performance graph for ray tracing.

 

[dj-electric]

I can't imagine the amount of work behind this stuff.

Good s@$t, friend. Godspeed.

 

[cellar door]

In the past 10 years - if I needed a GPU bios, I always got it from TPU. What were you thinking @W1zzard looking for one somewhere else

 

[TheoneandonlyMrK]

Love your work totally, but I agree with Mechtech the Rx580 and 1060 should be on the charts, there are so many still gaming on them that they make the vast majority of cards being used and upgraded from, so it could be nice to show what those still using them are missing.

then again with availability due in julember it might just look like rubbing salt in the wounds, come to think of it reviewing 26 cards does look a bit show offy

 

[BSim500]

I like the way you've done the power consumption tests. Some feedback though - I understand the need to update cards, but the problem with kicking too many older cards out (especially most commonly owned ones like the 1050Ti / 1060) is that people are more likely to upgrade from older cards that are still #1 and #2 in Steam survey than they are throw away last year's card, and those owners now have no way of comparing old vs new to get an idea of % improvement. Same reason why the i5-2500K / i7-2600K were included in CPU benchmarks long after Haswell came along.

 

[W1zzard]

In the past 10 years - if I needed a GPU bios, I always got it from TPU. What were you thinking @W1zzard looking for one somewhere else

Indeed, what was I thinking

could the newer results not be combined with older ones

Different test system, games, drivers, won't be apples-to-apples. Our GPU database does have a bar chart that is close enough to what you're looking for.

Minimum (1% would be fine) and average FPS would also be a great addition to the bench results and drag them in-line with most e-publications.

We have the frametime analysis page, which goes into much more detail?

1050Ti

Hasn't been in our charts since May 2019. First time I hear about anyone asking for 1050 Ti. Does any other reviewer include more cards than I do?

 

[Nihilus]

3050Ti should be the lowest card tested. Im really only interested in 3090/4070 and above.

Cool story bro.

Agreed with many of those above on adding some of the more popular older cards.

 

[BSim500]

Hasn't been in our charts since May 2019. First time I hear about anyone asking for 1050 Ti. Does any other reviewer include more cards than I do?

Techspot and computerbase.de still include the 1060 in the 3060 review. With the highest number of cards listed being Guru3D. Likewise the 1050Ti was one of those mentioned here to be making a comeback in production due to shortages. Given that people tend to upgrade in cycles rather than generations, sometimes relevance can be more important than "strict new-ness" cutoffs.

 

[Kaotik]

Does any other reviewer include more cards than I do?

Yeah, crazy russians over at GameGPU got few more, around 30 I think it was. And they test them with over 30 different CPUs, too, each.

 

[xkm1948]

I REALLY wish you can present RT results in a cleaner way. Those bar graphs are so damn busy I have no idea where to focus on without a laser pointer and my imaginative W1zzard's voice narration.