OTOY Forums

PaGz wrote:Hello

What configuration is faster?
Two RTX 2070 or one RTX 2080 TI ?

Thanks

Two RTX 2070 should be faster than one RTX 2080 Ti based on some benchmarks.

BUT

Keep in mind that NV Link will not be available for RTX 2070 cards.

It is still unclear what benefit the NV Link will bring for two RTX 2080 or two RTX 2080 Ti.

Some sources say that it might be possible to get stacking VRAM of 22 GB at a slower render speed by using NV Link.

-> If you are not in a hurry wait until it is fully clear what RTX and NV Link features can be supported in Octane.

Thanks for the advice

I am actually searching for the better price/performance.
The graphic cards are so expensive since that bitcoin mania...
I am still stuck with 1 GTX 780 + 1 GTX 980 and this is not enough fast for me...

I think about keeping the 980 and add something like a 1080 TI or a 2080 TI that is too expensive
or
just trow away all my older cards and buy two 2070 even there is no NV Link

Now with Octane 4 if i understand the Out of Core Geometry will get the motherboard ram so for the rare situations where i will need more memory i will just wait longer because i think that this will slow down the rendering...

Hi,
OctaneBench 4 results in single GPU mode with Volta and RTX scores:
https://render.otoy.com/octanebench/res ... ingleGPU=1
ciao Beppe

PaGz I had two GTX 780 and replace with two RTX 2070 for less than the price of RTX 2080ti. Currently with the settings at OC Mode it runs 50% faster than a RTX 2080ti. The OctaneBench 4 score I get is 450 for my 2x RTX 2070. In OctaneBench 4 average results for one RTX 2080ti is 300 and for 2x RTX2080 440.

Currently for octane this is the best price/performance available. If the 8gb memory limit with no nvlink is acceptable to you.

These results are pretty odd…

Nothing odd they are the OC edition of the rtx 2070 running with the OC Mode settings. In the octanebench4 the average results for 2x rtx 2070 is 430. I just did the test again and got an octanebench4 results of 446 (see attach picture).

Over clocked ! thank you for these informations…

I did not have to overclock the 2070 OC edition card by that much it came with default boost clock of 1740 MHZ. I just increase it to 1800 MHz and increase the memory speed 15500. For faster RTX 2070 is important to get a Overclock / Gaming edition that has the 400A GPU and good cooling.

eklein wrote:Nothing odd they are the OC edition of the rtx 2070 running with the OC Mode settings. In the octanebench4 the average results for 2x rtx 2070 is 430. I just did the test again and got an octanebench4 results of 446 (see attach picture).

It is actually interesting, i mean the OctaneBench average result as well, given the fact the 2070 chip (TU106) is half of the 2080Ti one (TU102) - even though not exact half, since 2080Ti is slightly cut, its 2304/4352 = 0,53...

Anyway, if the 2080Ti scores on average 300 points, which i can confirm, one would expect the 2070 to score somewhere in the 160 -170 range, not 220.

This most likely points out to the way higher GPU clocks for 2070, probably consistently above 2GHz, and i have to wonder about VRAM speeds as well. From my limited experience with 2x 2080Ti on stock settings, during OctaneBench test they hover aroud 1800 MHz on avg (the upper one more in the 1700-1800s, the bottom one 1800-1900)...there is not room for trying any OC in my case, as the temps are already 78C and 84C respectively...so its very likely i am leaving quite a lot of performance on the table and could be looking at maybe close to 700 points with better cooling, rather than current 600-610 - judging by those 2070 results. Cant really see any other explanation.

Actually looking at the OctaneBench now, i see 2x 2070 scoring those 433 OB points, while result right above is 2x2080 with 439 . Since those 2070s have 79 percent of 2080´s cores, they need to run 20 percent faster than those 2080 to make up for that. I guess, that is plausible, with 2080 at cca 1800, and 2070 at 2150....

On another note, i tested 2018.1 XB1 speed against 4.00 in one of my own scenes, not really complex exterior architecture model with daylight/HDR environment without any foliage or any other light sources...and the result was once again interesting. I had set up 2 cameras, each one from different angle and tried how long it takes to get to 500Ms for each camera. It turned out one of them was actually very slightly slower, 35 seconds to 36s in case of 4.0, the other one though saw the time cut by 10s (from 48s to 38s in XB1). Tried only single run, so not really most scientific, but still, perhaps i would get similar result if i repeated the test. Have to wonder if the speed increase in the second camera angle is down to RT cores coming into play...and why only one of the cameras got faster....

The issue of performance with 2070 vs 2080ti. One factor is the GPU for 2080ti is much larger so it will generate more heat the GPU won't be able to easily reach the higher clock rates as a 2070. The second factor is the amount of cuda cores. A third factor and just as important is data bandwidth to the GPU cuda cores. It looks like all these cuda cores from 2080ti don't have enough data to be fully utilize affecting it's performance.