RTX Benchmark thread...show me the power

13»

Comments

  • outrider42outrider42 Posts: 2,775
    edited March 1

    I uploaded a copy to sharecg.

    https://sharecg.com/v/97260/view/21/DAZ-Studio/Iray-RTX-Benchmark-Scene

    I also ran some new tests.

    Daz 4.15.0.2

    Nvidia driver 461.72

    One 1080ti:

    2021-02-28 18:49:19.254 Total Rendering Time: 2 minutes 9.66 seconds

    2021-02-28 18:49:26.150 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2021-02-28 18:49:26.150 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce GTX 1080 Ti): 600 iterations, 1.301s init, 127.492s render

    That comes to 4.7 iterations per second.

    TWO 1080tis:

    021-02-28 18:43:40.255 Total Rendering Time: 1 minutes 8.56 seconds

    2021-02-28 18:44:10.146 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2021-02-28 18:44:10.146 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce GTX 1080 Ti): 301 iterations, 2.075s init, 64.732s render

    2021-02-28 18:44:10.146 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce GTX 1080 Ti): 299 iterations, 1.424s init, 65.913s render

    That works out to 9.2 iterations per second.

    Interestingly I don't see the performance improvement that the raydiant benchmark scene has with 4.14 and 4.15. Just to be sure, I ran the raydiant bench one more time and got the same exact time I did yesterday.

    I think I see why, this scene has no normal maps at all. But the rayDant scene does have normal maps on the outfit. The notes for 4.14 does mention that a change was made to make normal maps more efficient. So I believe this confirms that the speed increase observed in Daz 4.14 is directly tied to the scene using normal maps.

    This also means that we can indeed directly compare any new numbers posted to the old numbers posted when this thread was created. That is excellent.

     

    Post edited by outrider42 on
  • Matt_CastleMatt_Castle Posts: 1,342
    edited March 1

    As far as keeping the results here as well, this was my test with the new 3060:

    GPU: ASUS 3060 TUF OC
    Driver: 461.72
    DS: 4.15.0.2

    2021-03-01 01:10:57.671 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3060):      600 iterations, 3.504s, 31.993s render

    Which equates to 18.75 iterations per second.

    This was done mostly as far as seeing whether there was any relative difference with the new Ampere architectures in a scene with different geometry (as compared to the RayDant benchmark), but the relative performance against other cards seems to stay about the same. (We were mostly comparing to the 2080 Ti, which the 3060 gets surprisingly close to, at about ~90% of the performance in Iray).

    Post edited by Matt_Castle on
  • outrider42outrider42 Posts: 2,775

    Interesting, so the 3060 actually loses a bit of ground in this benchmark. It is still very close, and ultimately puts the $1200 2080ti to shame. It would be pretty cool to see more Ampere cards take on this scene.

    And compared to the 1080ti, you ran the scene in 32 seconds that took a 1080ti over 2 minutes. So the flagship of 2017 took 4 times longer. That is crazy, and in that comparison shows the difference between non RTX and RTX can get much wider than what is observed with rayDant's scene.

  • takezo_3001takezo_3001 Posts: 1,089

    Wow, the 3090 is a beast, so glad I saved up for nearly a year for this bad boy!

    GPU: RTX 3090

    DRIVER: 461.72

    DS: 4.15.0.13
    2021-02-28 21:22:41.579 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090):      600 iterations, 1.871s init, 13.149s render

    The second test, default rendering at 5000 iterations:

    IRAY   rend info : Device statistics:
    2021-02-28 21:42:48.218 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090):      5000 iterations, 1.533s init, 106.997s render

    Both are still infected with fireflies though...

     

  • S.CaronS.Caron Posts: 8
    edited March 1

    GPU: RTX 3070

    DRIVER: 461.72

    DS: 4.15.0.2
    2021-03-01 12:55:16.567 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Received update to 00600 iterations after 22.655s.

    My new rig as of last Thursday.

    And now a small PSA.

    When I first started using this new machine my CPU temps were though the roof if I added in CPU rendering to the mix (as in 85-95C)!
    Come to find out there is something (new to me) called Multicore Enhancement that from what I understand is enabled be default on a lot of the Z490 series boards from manufacturers.  So by default the motherboard was sending way more voltage to the CPU (essentially overclocking) the processor to 5.1Ghz.  I kid you not.  Anyways I found where to disable it in the BIOS and I've been running now at roughly 65 to 75C.  So just a small warning if you get setup with the 10th Gen chip sets.  I can't fathom why board makers would have that enabled by default just to eek out a few more CPU cycles at the cost of possibly doing damage to your processor or motherboard. A quick Google search will show many discussions about this, and how really it should be off by default.

    Ok end rant!

    Post edited by S.Caron on
  • outrider42outrider42 Posts: 2,775

    takezo_3001 said:

    Wow, the 3090 is a beast, so glad I saved up for nearly a year for this bad boy!

    GPU: RTX 3090

    DRIVER: 461.72

    DS: 4.15.0.13
    2021-02-28 21:22:41.579 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090):      600 iterations, 1.871s init, 13.149s render

    The second test, default rendering at 5000 iterations:

    IRAY   rend info : Device statistics:
    2021-02-28 21:42:48.218 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090):      5000 iterations, 1.533s init, 106.997s render

    Both are still infected with fireflies though...

    That comes out to 45.6 iterations per second, LOL.

    The second test is interesting, as its rate comes to 46.73. So no drop off for rendering longer, it actually got faster. It is great to see the scaling work perfectly. I am thinking that the 3090 rendered the 600 iterations so fast that the initial ramp up to full clockspeed might have penalized it slightly. But it is so minor in the overall scope.

    And just for comparison, that is exactly 10 times faster than my 1080ti, and a great demonstration of what RTX and Ampere is capable of. Just think, if you built a scene that matched this complexity and it took you 30 minutes, it would take me 5 hours to render. And the 1080ti was the fastest GPU in 2017. It is not that freakin' old.

  • takezo_3001takezo_3001 Posts: 1,089

    outrider42 said:

    takezo_3001 said:

    Wow, the 3090 is a beast, so glad I saved up for nearly a year for this bad boy!

    GPU: RTX 3090

    DRIVER: 461.72

    DS: 4.15.0.13
    2021-02-28 21:22:41.579 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090):      600 iterations, 1.871s init, 13.149s render

    The second test, default rendering at 5000 iterations:

    IRAY   rend info : Device statistics:
    2021-02-28 21:42:48.218 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090):      5000 iterations, 1.533s init, 106.997s render

    Both are still infected with fireflies though...

    That comes out to 45.6 iterations per second, LOL.

    WoW! That is fast, faster than I thought!  Yes, it's the best PC parts investment that I ever made, I really lucked out at getting one, it really sucks that the market is pretty s*#|76 right now; curse those scalpers, hope their PCs gets a worm infection!

    The second test is interesting, as its rate comes to 46.73. So no drop off for rendering longer, it actually got faster. It is great to see the scaling work perfectly. I am thinking that the 3090 rendered the 600 iterations so fast that the initial ramp up to full clockspeed might have penalized it slightly. But it is so minor in the overall scope.

    And just for comparison, that is exactly 10 times faster than my 1080ti, and a great demonstration of what RTX and Ampere is capable of. Just think, if you built a scene that matched this complexity and it took you 30 minutes, it would take me 5 hours to render. And the 1080ti was the fastest GPU in 2017. It is not that freakin' old.

    I did not even notice that thanks for pointing it out, heh, it would have been at least 12 times faster than my old faithful 1080, I'm so glad I never listened to the "pro-3080/anti-3090/wait-'til-the-prices-come-down crowd!"

  • outrider42outrider42 Posts: 2,775

    I decided to run this test with my Ryzen 5800X and see what happened.

    021-03-02 20:28:17.103 Total Rendering Time: 5 minutes 6.40 seconds

    2021-03-02 20:28:48.362 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CPU: 600 iterations, 1.422s init, 303.043s render

    That works out to 1.9799 iterations per second.

    Now why is this interesting? Because in the rayDant bench scene it only scored a pitiful 0.732 iterations per second. So, a lot of cards saw higher numbers. But those are RTX cards. My GTX 1080ti actually scored a very similar iteration speed as it did with the rayDant test, which was right at 4.7 iterations.

    This does alter the perception a bit. In the rayDant scene, the 5700X is so slow that just about any GPU is faster. But in this benchmark, it is not quite THAT slow. While my 1080ti is still clearly faster, and faster by several times, it is not the almost absurd difference seen with rayDant's bench.

    So it appears that CPU rendering might have one small trick up its sleeve. My CPU is not the only one to behave this way, on 2nd page an 18 core Intel chip saw a very similar uptick in speed (and humorously my 8 core 5800X is just a hair slower than that 18 core Intel that uses almost twice the power). Still, this does not really make CPU rendering much more viable than before, as the GPUs are still faster, easpecially the RTX cards that make this benchmark laughable.

Sign In or Register to comment.