Jump to content
3DCoat Forums

GPU spec to look at


davide445
 Share

Recommended Posts

  • Member

Just finishing my first serious work on 3DC, the remodelling and animation of a 3d-scanned butterfly.

Mine 4 years old HD7950Boost GPU didn't have problems managing the 1.5 Mil tris model, use various brush and render the animation.

Future work will increase the poly count and complexity, also the GPU is old and not sure how much can last.

Purchased this GPU since had the biggest bus and VRAM I can afford at that time, supposing just DX/GL performances isn't enough working with 3DCC tools and not just games.

Now with bus shrinking using texture compression, unif shaders count changing his value depending from the architecture, I hypotized the bandwidth can be a main factor to choose from. I looked also to pixel fill rate data but did find too much variation, I also didn't think GFLOP is reevant for our use.

I did create a graph with comparable GPU with mine in term of bus width, bandwidth or DX performances. Left axis the bandwidth, right axis $/bandwidth. Red dots the cards with bus narrower than mine 384 bit. (Apologize if all of that it's a bit brainy, I did share with Andrew the physics education - but not his experience - so I did like to play with numbers :)).

Conclusions from the data and this graph

- In every new generation you need to go up to find the same bus width, with Nvidia especially aggressive on that. On current Pascal arch only Titan X show the same 384 bit bus mine old card

- Nvidia does ask normally about 75% more than AMD for the same bandwidth

- Currently the best bang for the buck is coming from R9 390, followed by RX 480. Future RX 580 (Vega 11) might be the third in the row, if the specs I read are true.

My GPU usage is among few programs: 3DC for modelling, HitFilmPro for VFX, some specialized for scanned mesh editing, hopefully I will add some CAD, Unity for animation rendering, Akeytsu for rigging and animation. I won't use GPU renderer.

This are just numbers and not first hand experience among various GPU, will be glad if anyone will share his experience to confirm what models are more valuable to work with 3DC and other 3DCC.

If I need to choose: a GTX 1060 does have -25% bandwidth vs a RX 480, being roughly at the same level in DX performances. A 1070 does have exactly the same bandwidth of the RX 480, but +50% better DX performance (in DX11, different story in DX12). I can still find an R9 390 with +50% bandwidth vs RX 480, with about the same DX performances. Will I notice a real difference among all of them?

 

GPU evaluation.JPG

Edited by davide445
  • Like 1
Link to comment
Share on other sites

  • Member

Ok digging a bit more, I didn't find benchmark in 3DC, but tried to find bench on 3D Content Creation (3DCC) sw or bench able to evaluate the effect of bandwidth. Also cross check wih standard DX game test.

Reading AutoDesk 3ds Max 2017 GeForce GPU Performance from Puget Systems, HWBench RX 480 s R9 390X (390X sharing with 390 the same arch and bandwidth, just adding some shaders) and also the same HWBench RX 480 vs GTX 1060 or GTX 1070 comparision, Maya 2012 SPECviewperf12 results

- Only with 6-7 Mil poly ou notice the difference btw a mid range and a powerful GPU

- High res texture are more CPU and RAM bound

- Large scenes can benefit from more VRAM and bandwidth

- In DX tests btw AMD more resolution from FullHD (2 Mil pixel), 1440p (2.7 Mil pixel) and 4k (8.3 Mil pixel) benefit more and more the R9 390X vs the RX 480. But it's on average just 7% (12% in 4k) faster with 22% more shaders and 50% more bandwidth.

- Always in DX tests AMD vs Nvidia didn't show the same result. 480 vs 1060 is on average 5% slower (in DX11), despite having -25% bandwidth and -44% shaders. DX12 show them on par. 1070 show on average 45% better performances in DX11 and 14% in DX12, having -16% shaders and same bandwidth.

- Maya bench W8100 vs W7100 (+43% shaders and x2 bandwidth) show a +23% perf increase.

- Maya bench M6000 vs M5000 (+50% both shaders and bandwidth) show a +19% increase

So need to say I didn't find a strong evidence bandwidth itself it's a reliable indicator for performance. Architecture, shaders num, bandwidth, graphics API all play some role. Interesting to find confirmation Nvidia ask more $ per perf than AMD (1060 +25% for about same perf, 1070 +59% for 45-14% increase).

Interested to know if there is any bench for retail (not workstation) cards that can drive a decision also for 3d-Coat.

Edited by davide445
Link to comment
Share on other sites

  • Member

Just to close my brainy digression, looking at RX 480 performances in professional applications and using the same for GTX 1080 as a proxy for GTX 1060 with a -50% performance correction (coming from this test) I conclude definitely that bandwidth btw different arch generation it's less a good indication of performance.

A RX480/GTX 1060 with 256GBps bandwidth does have about the same performance of a R9 390X with 512GBps even on professional applications with fairly high workload.

Bandwidth tend naturally to growth due to the growing demand in term in textures size, model poly count, screen resolution, so having a bigger bandwidth is definitely an advantage, but the gain it's not linear at all.

After all this analysis what I'm sure now is I have still a lot of road to cover before reaching the limit of my current hw. After that a modern midrange GPU will be probably a good enough update to have some more easy time working with 3DC and other packages.

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

 Share

×
×
  • Create New...