Curious benchmark result

[pavoda]

Hi,
I tried RTX 2080ti to Indigobench.
Result is Ok, but if I render together with my second GTX 1070 result is worse than single RTX 2080ti.

Thisis my results at indigobench (bedroom)

RTX 2080ti....8.511
GTX 1070.....3.855
RTX 2080ti + GTX 1070...7.396

before 2x GeForce GTX 1070...6.861

When rendering RTX usage is only about 50-60 percent. GTX 1070 go to full power 90-100 perc.

[Zom-B]

Out of curiosity: load the benchmark scene in the latest Indigo build and test if GPU utilization is as bad as in the benchmark.

[pavoda]

It is totaly same in latest Indigo.

[Paraket]

I have a similar problem. I have 1060 and 1080Ti at the same time they work as one 1060. I think that the problem is in the motherboard. I have one PCIe 3.0 x16 slot, and the second PCIe 2.0 x16 slot. I am not very good at video cards.
And one more little poop: the size of the video memory is limited to the smallest card. At 1070, this is 8GB, so the render will be turned off at 6-7GB with a large number of textured objects.
I used to think that the video memory size is summed up (1070 8GB + 1080Ti 11GB = 19GB), but, unfortunately, practice has shown that this does not happen (1070 8GB + 1080Ti 11GB = 8GB).
I would be glad if I am mistaken and I just need to buy a not-peckerwood-motherboard.

[Zom-B]

I used to think that the video memory size is summed up (1070 8GB + 1080Ti 11GB = 19GB), but, unfortunately, practice has shown that this does not happen (1070 8GB + 1080Ti 11GB = 8GB).

The new NVIDIA 2080 (and upwards, thanks pixie!) Cards have a feature where you could connect them with a special bridge and they should add up Memory for working on the same stuff!
BUT this has to be tested with Indigo, no idea if it works out of the box at all!

Hitting bottlenecks with a 1080ti + 1070 sounds strange, both cards aren't that far away from each other regarding performance. PICEx8 is fine to run up to 5 x 1080Ti with full force, adding card No 6 starts to bottleneck a system (atm)

I assume you guys have full GPU utilization on OctaBench?!
https://render.otoy.com/octanebench/

[pixie]

@Zom-B
It work in tandem with 2080 upwards. That's why I bought a 2080 and not a 2070

[Zom-B]

Thanks for the info pixie, I updated my post.
Now please buy a second one and test if it works within Indigo

[Juju]

Now please buy a second one and test if it works within Indigo

ROTFLOL

[pixie]

Thanks for the info pixie, I updated my post.
Now please buy a second one and test if it works within Indigo

Please don't tease me, I might well do it!

[pavoda]

I have a similar problem. I have 1060 and 1080Ti at the same time they work as one 1060. I think that the problem is in the motherboard. I have one PCIe 3.0 x16 slot, and the second PCIe 2.0 x16 slot. I am not very good at video cards.
And one more little poop: the size of the video memory is limited to the smallest card. At 1070, this is 8GB, so the render will be turned off at 6-7GB with a large number of textured objects.
I used to think that the video memory size is summed up (1070 8GB + 1080Ti 11GB = 19GB), but, unfortunately, practice has shown that this does not happen (1070 8GB + 1080Ti 11GB = 8GB).
I would be glad if I am mistaken and I just need to buy a not-peckerwood-motherboard.

Yes, it makes sense, but I tested it and that is not motherboard fault. I can render 2 indigo scene together, every scene run on one card. Both cards go to full power. Also I have octan render and there results are correct .
It seems that is fail of indigo.
I think the Indigo is limited by first card.
My sequence of card is first 1070, second 2080ti. If I switch to first position 2080ti it maybe will be correct...but it is not necessary for me now. I can test it later.

[pavoda]

@Zom-B
It work in tandem with 2080 upwards. That's why I bought a 2080 and not a 2070

I have Information RTX 2080 can use nvlink, but only two Titans RTX can count memory size. Therefore nvlink with 2 RTX 2080(ti) have still 8(11)GB. 2 Titans RTX have 2x24 = 48Gb ...just pricing policy

[pixie]

It is said:

On the GeForce RTX 2080 Ti, NVLink allows for up to 100GB/s bidirectional bandwidth between the cards (the GeForce RTX 2080 offers 50GB/s). When fully utilized, NVLink will minimize inter-GPU traffic over the PCI Express interface and also allows the memory on each card to behave more as a single, shared resource. Traditional SLI was effectively a display interface, where the output from two GPUs was linked together and sent out to the display. To maintain compatibility, NVLink does the same thing, but it is also a high-bandwidth, low-latency memory interface, which opens up the possibility for new multi-GPU modes, not only for gaming but for scientific analysis and big data workloads as well.

on https://hothardware.com/reviews/nvidia- ... next-level

[pixie]

It is said:

On the GeForce RTX 2080 Ti, NVLink allows for up to 100GB/s bidirectional bandwidth between the cards (the GeForce RTX 2080 offers 50GB/s). When fully utilized, NVLink will minimize inter-GPU traffic over the PCI Express interface and also allows the memory on each card to behave more as a single, shared resource. Traditional SLI was effectively a display interface, where the output from two GPUs was linked together and sent out to the display. To maintain compatibility, NVLink does the same thing, but it is also a high-bandwidth, low-latency memory interface, which opens up the possibility for new multi-GPU modes, not only for gaming but for scientific analysis and big data workloads as well.

on https://hothardware.com/reviews/nvidia- ... next-level

Do GeForce RTX Cards Support Memory Pooling in Windows?
Not directly. While NVLink can be enabled and peer-to-peer communication is functional, accessing memory across video cards depends on software support. If an application is written to be aware of NVLink and take advantage of that feature, then two GeForce RTX cards (or any others that support NVLink) could work together on a larger data set than they could individually.

on :https://www.pugetsystems.com/labs/artic ... s-10-1253/

[pavoda]

It is said:

On the GeForce RTX 2080 Ti, NVLink allows for up to 100GB/s bidirectional bandwidth between the cards (the GeForce RTX 2080 offers 50GB/s). When fully utilized, NVLink will minimize inter-GPU traffic over the PCI Express interface and also allows the memory on each card to behave more as a single, shared resource. Traditional SLI was effectively a display interface, where the output from two GPUs was linked together and sent out to the display. To maintain compatibility, NVLink does the same thing, but it is also a high-bandwidth, low-latency memory interface, which opens up the possibility for new multi-GPU modes, not only for gaming but for scientific analysis and big data workloads as well.

on https://hothardware.com/reviews/nvidia- ... next-level

Interesting!

[pavoda]

Now I have even more curious thing in indigo.

I was try to change position of graphic card in motherboard. I was reinstall drivers and now in indigo list of GPU devices include followings

(GPU) GeForce GTX 1070
(GPU) GeForce GTX 1070
(GPU) GeForce RTX 2080 ti
(GPU) GeForce RTX 2080 ti

If I check it one GPU, this GPU render on full power. If I Check it two same GPU, it is same like before. But If I check 2 different GPUs the RTX render on half power and GTX on full power.
Now I can check one GTX and two RTX and both cards render on full power So curious...(bug maybe?)

For me it is ok...I can choose between full and half power of RTX