I was wondering if you could add additional system hardware info in future releases:
1. Not seeing win 11 at all.
2. Pcie version (v3.0, v4.0, etc.)
3. Onboard ram.
4. Overclocking
5. Average results
6. Sortable fields.
7. motherboards, bios versions, nvme types, SSRs,PCIe lane availability, etc. perhaps a way to integrate something like cpuid or hwinfo
I mention this because I believe the results (showing the high-end) are extremely skewed. There’s definitely crypto miners using this tool and overclockers which don’t necessarily represent average results. While the tool is useful, it’s difficult for an average person focusing on stability (not overclocking) to determine if their hardware is within a certain acceptable range.
This is possible now, a colleague of mine showed me in the past but I cannot remember now. I will ask again and get back to you
But yeah this should be easier to find.
This will not be possible and probably makes things complicated
PCIe Gen 4 x4 is the same speed as PCIe Gen 3 x8 , so the generation alone doesn’t tell the whole story.
And depending on the number of devices connected, this link speed will change. And this depends on the platform and board
There is also PCIe Gen 5 with some Z690 boards now, this will be a complicated task to do
I have done some PCIe bandwidth testing for V-Ray GPU, let me know if you are interested. I can share the data
Do you mean system memory?
It should be possible yes, although it doesn’t really tell much(unless we report additional information like memory channels, frequency and latency, then this becomes very complicated)
Not possible,
There is no easy way to tell if a CPU or GPU are overclocked. For GPUs some 3rd party cards come factory overclocked and you will have many vendors competing like Asus, MSI, Zotac..etc
They all try to push GPU frequency higher, there is no standard really to match against and tell if there is Overclocking
For CPUs it is a similar story, even the reported frequency is not gauranteed to be accurate. With PBO, AMD CPUs will boost higher as long as temperatures are within limit
Too much work honestly, HWinfo and/or GPUz exist and are free
PCIe lane availability is something very hard to report, most softwares now are not able to verify correctly
What you can do is check a device’s link, like a GPU or an NVMe
I agree that Extreme results are misleading, hence why we added the average results
Extreme results are indeed overclocking(but not miners)
Mining hardware will have bad benchmarks actually, as miners will undervolt it to save electricity.. this undervolting is quite extreme and mostly through a custom bios
Thanks for the feedback, we will consider for the upcoming versions
Extreme results are indeed overclocking(but not miners)
Mining hardware will have bad benchmarks actually, as miners will undervolt it to save electricity.. this undervolting is quite extreme and mostly through a custom bios
Interesting. I seen a few results in which someone or some company had machines with 8+ (even 12) GPUs and all I could think was that it was perhaps someone using their rigs to mine. I just don’t even know how it is possible to run this many cards.
Many consumer boards could equip up to 7 GPUs, examples are Asus X299 WS Pro and Asus Pro WS WRX80E
Each of these will be able to equip 7 GPUs natively and they will run at the proper bandwidth, the second board will run all GPUs at x16 PCIe Gen 4
For this to work, you need to use riser cables or you watercool the GPUs(removing the massive air cooling shrouds)
On another note, I built 8 and 10 GPU servers in the past. They exist and are fairly common in GPU rendering
Supermicro offers many boards and cases that equip 10 GPUs natively.. it uses dual Xeon CPUs and provides x16 bandwidth as well
More than 10 GPUs, you will need to use clusters. I have seen up to 18 GPUs per machine using clusters
Anything more than 4 GPUs will not work out of box and will require bios tweaking
This is very different than mining setups, where PCIe bandwidth don’t matter at all. Mining rigs are much cheaper than this
For this to work, you need to use riser cables or you watercool the GPUs(removing the massive air cooling shrouds)
I have riser cables and had done this with my current intel board. Also watercooled my previous cards.
It’s something I got out of due to stability and power requirements. I had a difficult time trying to find reasonably priced riser boards for a separate chasis. It did seem like the majority of them and info years back were mostly geared towards crypto.
It’s something I had brought up with Vlado a long time ago and I remember he was curious then ran some experiment. It must have been in the infancy of Vray GPU rendering.
Yeah, it is a bit different now
Riser cables or watercooling can be very reliable at the moment. So 7, 8 and 10 GPU machines are not odd on the GPU Rendering community. I had access to 2 of these when I worked at Carlex a few years back
7 GPUs with Riser cables is possible, and kind of easy to pull off using Riser cables and a variety of boards like the Asus X299 Pro WS.. the setup itself is not expensive(without the cards)
We have quite a bunch of GPU rigs, most of them have their GPU’s on x1 risers.
The pure renderpower is fine and the GPUs are excellent to cool on risers.
The downside is that building the light cache for example, or using a denoiser makes a lot of busload and slows down the rendertime in comparisson to onboard GPUs.
Yeah, exactly
You need x16 shielded risers for this to be optimal, it will be identical to using the PCIe lane on the board. You will find some fancy Gen 4 risers now as well in case you have high-end Ampere GPUs
Yeah this is why people use risers, they offer really good thermal performance as you can space the cards in a GPU rack
And for something like the X299 Pro WS, you can equip up to 7 GPUs on this board using risers
And yes, the Render power is fine once the data is loaded in the GPU memory. should be identical to running natively
In my testing, x8 Gen 3 is the same speed as x16 Gen 3. This bandwidth is usually enough for GPU rendering
Then with x4 Gen 3 you will start losing performance.. like 5-10% depending on how heavy the scene is
x1 is much slower, specially with something like a 3090/3080 .. heavy scenes will take a while to load into GPU memory before the actual rendering starts
x4 Gen 4 is similar to x8 Gen 3 and so on
