Pass mark performance 3d mark score difference inquiry

[uniconkj25]

1. Why are the 3D makr (UL) results and passmark results different when comparing 3 types of graphics cards?
2. Please tell me the difference between Direct 9, 10, 11, 12, and GPU Compute in Passmark 3D mark.
3. Is the 3DMark measurement score a reliable score?
4. I've seen some users lose confidence in the result score. What do you think?

[David (PassMark)]

1) You can find a description of the 3D tests in PerformanceTest here.
Different benchmarks can and do produce different results. Same for testing with different games and 3D content creation software. This is normal. You should always use a range of different benchmark software (unless, for example, you only use the computer for Photoshop, then in that case it only makes sense to test with Photoshop).

It should also be noted that poorly setup computers are unfortunately pretty common. Wrong device drivers, wrong PCIe slot, XMP not being used, single channel RAM, cooling issues, etc, etc... this can result in a range of benchmark results for the same CPU & GPU. But this is part of what a benchmark is for, to find those setup issues and fix them up.

2) DirectX is Microsoft programming interface for 3D. It is the most common interface used for Windows gaming. A DirectX overview can be found here. Compute refers to computational tasks that don't result in 3D rendered output. Bitcoin mining and physics are examples, but there are 100s of others.

3) Yes. But as noted above it depends on how you use your computer as to if it is relevant to you.

4) We aren't aware of any issues. We think our GPU charts are a reasonable reflection of general GPU performance. I am sure some people will argue about the exact ordering when cards are close in performance. Many people like to cherry pick results from certain games, so there will never be agreement on any ordered list. But it is an attempt to represent the entire performance of a video card with a single number, so all attempts to do this will be to some degree imperfect.

[uniconkj25]

Thank you for your response.


Additionally, I have more questions.

1. When you rotate several benchmark programs (3D mark, cine bench, nova bench, superposition ...), you get the same result as 3D mark.
By the way, it only appears differently in Passmark. What do you think about this?

3D mark(UL)
GTX1660 super > RTX3050(8GB) > RTX3050 (6GB)

Passmark
RTX3050(8GB) > GTX1660 super > RTX3050 (6GB)

2. Is Passmark 3d graphic mark score calculated based on direct 9,10,11,12 gpu compute score?

3. How can I interpret the difference between Direct11 and GPU Compute in the RTX 3050 (8GB) and GTX 1660 Super results in the picture?

[David (PassMark)]

3D mark, cine bench, nova bench, superposition ...), you get the same result as 3D mark

This isn't true if you look at a wide range of systems.

3D mark(UL)
GTX1660 super > RTX3050(8GB) > RTX3050 (6GB)

You contradicted yourself with your own screen shot
Your screen shot shows the 3050 8GB as being faster.

Passmark
RTX3050(8GB) > GTX1660 super > RTX3050 (6GB)

We rank them as super close on average. The difference is under 1%. Which is below the margin of error.
On a different day they could easily be reversed. (we recalculate the average result every day, based on new benchmarks)


But again, it isn't a reasonable expectation that all benchmarks produce the same rankings for all systems. It doesn't work like this in real life. Different games, apps & benchmarks, place different types of load on the hardware. One app might be more CPU bound, or it might be more dependant on video RAM speed, or might make more heavy use of shaders, etc...

Is Passmark 3d graphic mark score calculated based on direct 9,10,11,12 gpu compute score?

Yes.

How can I interpret the difference between Direct11 and GPU Compute...

Hard to comment without know more about the machines. For example it isn't clear if they are running the same device driver, using the came CPU, running the same resolution or even in the same motherboard. But all else being equal, 111fps is faster than 88fps in DirectX11. Just because the two cards are very close overall, doesn't mean they will be identical for all different types of workloads.

[uniconkj25]

Thank you for your reply.

There was an error in the content.
3D mark(UL)
RTX3050(8GB) >GTX1660 super > RTX3050 (6GB)
Passmark
GTX1660 super >RTX3050(8GB) RTX3050 (6GB)

So, which side of the passmark program's 3d mark bench has more load? vram? shaders compute? CPU?

[David (PassMark)]

Passmark
GTX1660 super >RTX3050(8GB) RTX3050 (6GB)

This isn't correct. Ranking as of today shows the 3050 (8GB) is faster. Exactly the same order as UL.
But as mentioned above they are so close (in real life systems) that any difference is meaningless.

....more load? vram? shaders compute? CPU?

The quantity of video RAM doesn't matter too much. Once you get past 2GB of video RAM there isn't much, if any, gain in performance. The speed (bandwidth) of the RAM matters so iGPUs should be teamed with fast main memory. Vertex and pixel shader speed matters. Compute matters. CPU matters to some extent (especially on the low resolution DirectX9 & 10 tests). Anti-aliasing matters. Ray tracing and upscaling (DLSS) aren't used, so they don't matter in V11.

There is also a dependency on DirectX versions, 9,10,11, 12, DirectCompute and OpenCL. So cards that are weak with one or two APIs may suffer. (The initial Intel ARC cards were an example of this, DX12 was good, but DX9 was weak on the initial release)

[uniconkj25]

Ranking doesn't matter to me.
This is an inquiry regarding the difference in the result scores shown by Passmark. (UL score, passmark score)
All tests were conducted in the same environment, and three types of graphics cards (1660super, RTX3050 6GB, RTX3050 8GB) were compared.
We tested with various benchmark programs, and only Passmark showed different results, so we asked a question.


In Passmark results, 1660super's compute 4974, direct11 111, total score 11865,
3050 8GB model compute 5985, direct11 88, total score 10679
I would like to inquire whether this effect can be interpreted as being related to vRam and iGpu.

Thank you

[David (PassMark)]

and only Passmark showed different results

Except that we didn't. Cards were ranked in same order, when you look at the average result.

GTX 1660 Super seems to be better when using the older API, DirectX9,10 & 11.
RTC 3050 8GB seems to be better when when using Direct12 and compute. So if you just look at DX12 and compute tests, the 3050 is way better. Many games still run on DX9 however, this will fade over time however. Having more RAM and a PCIe Gen 4 bus also makes the 3050 more future proof. So IMHO the 3050 8GB is a significantly better card, despite the very close benchmark.