G3D scores explanation

[Qub]

I was looking at GPU "value chart" and noticed that

1st ranked GTX 1650 Super has a "G3D" score of ~9600

while

9th ranked RX 5500 XT has a score of ~8200

This is not a small difference.
In reality (games), 5500 XT beats 1650 S by a small margin.
I would even be content to call them equal, even tho they are not, but 9600 VS 8200?

So which GPU is misplaced, 1st (important for nVidia sales) or 9th (still slightly overpriced AMD GPU) on this list?

More importantly, how to interpret this discrepancy and what to think about the rest of rankings of cards I know nothing about?

Thanks a bunch o/

[David (PassMark)]

1650 Super is currently cheaper. So obviously that helps with the value.
5500 XT is faster with some games, but slower with others (e.g. World of Tanks roughly lines up with out results for these cards).

Our benchmark is never going to exactly match the performance of all games on all systems at all resolutions. One number can't do that.

Each game is different. I could probably cheery pick a few game benchmarks to get the same result as our benchmark. Or you could equally find a different bunch of games that gave a different results. You need to look at a broad range of results when selecting a graphics card.

[Qub]

While I fully understand your explanation, I still think that the gap between G3D scores of 8200 (5500 XT) and 9600 (1650 S) is simply too huge and distorts final values too much, giving nVidia too much credit.

As an argument, I provide a hefty set of AnandTech benchmark comparisons, to avoid sounding like someone petty, bickering over 2 or 3 percent.

https://www.anandtech.com/bench/product/2577?vs=2572

Anyway, I am following Passmark benchmarks for years. Your charts are exactly what every other similar site should have but doesn't. Thats why its the only commercial site for which I have adblock (uBlock Origin) off.

Thank you <3

[David (PassMark)]

We've got some early results from PerformanceTest V10.

5500 XT 3DMark 9,457
1650 Super 3DMark 10,590

So they are slightly closer in these preliminary results.

Even using the results you picked, the 1650 Super is faster on many measurements.
Assassins Creed Odyssey
Grand Theft Auto V
Compute: LuxMark 3.1 - Hotel
Synthetic: TessMark - Image Set 4
Synthetic: Beyond3D Suite
Shadow of the Tomb Raider - 99th PCTL
Strange Brigade - 99th PCTL

There might also be other factors at play. For example,

A) 1650 Super owners be slightly more inclined to tinker & tweak their card to get a few extra frames per second

B) The 3D tests also depend on CPU performance to some degree. So 5500 XT user might on average have a slightly worse CPU (in terms of single thread performance). Or 1650 owners like overclocking more.

C) Vendors of the 1650 Super might have started overclocking their cards slightly since those reviews were done.

D) There might have been new drivers released since those older Anandtech reviews were done. So the 1650 Super might actually be slightly faster now than it was in 2019. Or the inverse, some of the performance impacting security patches might have changed performance over the last 4 months (on average across the installed base).

E) There might be a small sub-set of 5500 XT users who are thermally limited (i.e. not enough cooling). This doesn't happen on the big review sites, but does in real life.

F) There might be some common of video card and monitors that limit frame rates (via Freesync & G-Sync) more on one card.

Our benchmarks were designed many years before either of these cards were available. They don't deliberately favour one card or the other.

We could likely spend days digging into the small performance differences, running experiments and working out the true cause, but we've got higher priority projects.

[Qub]

Though i am subbed to this thread I didn't receive any email notifications when you replied.

Could some laptop users maybe ruin these stats? I don't know how you collect device names etc but maybe mobile 5500 has a role in this discrepancy? (just an idea which might help with similar cases)

Thanks for your detailed response and everything you do.

[David (PassMark)]

Yes, it is possible,
Normally it is possible to know what if a GPU is a mobile version or a desktop version. But from time to time there are machines with the desktop part in a laptop, or there are GPU parts that internally are the same, but they underclock the laptop setup. (or expect the desktop part to thermally throttle in installed a laptop after a short amount of use).

[Qub]

Its AMD's turn to do what AMD usually does.

<3

[Qub]

Script that gathers data for Value charts swallowed TITAN for $30 and its all warped now.

[Richard (PassMark)]

Script that gathers data for Value charts swallowed TITAN for $30 and its all warped now.

Amazon recently switched over their product search API and seems we have some extra filtering needed to be done with the results returned.

We clean up most of the erroneous prices. If you spot anything we may of overlooked, please let us know.