I had a chance to meet with UL at CES 2019 and I asked a few questions about the ORB, DLSS, PCMark and remaking 2k1.
We have been waiting for a DLSS benchmark and you give us this demo, Any reason this was not a benchmark?
The NVIDIA DLSS feature test is more than a demo. It provides framerates for the test with DLSS on and off, as well as allowing for changing certain settings to affect performance. It was not made it?s own benchmark for two reasons.
DLSS is NVIDIA proprietary tech and thus can only run on NVIDIA hardware. We strive for neutrality in everything we do, thus a benchmark with 3DMark like scoring would not be in good faith to other vendors.
As it utilized Port Royal as the starting workload to then be upscaled by DLSS, it would be all too easy to confuse Port Royal benchmark scores with DLSS scores.
I would like to know more about the and the process involved with producing the Port Royal DLSS demo.
The NVIDIA DLSS feature test uses the same engine and assets as the standard Port Royal. Production process only involved integrating DLSS into the Port Royal engine (a simple process), and then distributing the modified binary with DLLs that NVIDIA provided. These DLLs contain the NVIDIA?s proprietary DLSS implementation. The approach taken Is identical to the approach a game developer would take to incorporate DLSS into their game.
AI was used to port the scenes from Port Royal raytracing benchmark to DLSS? Can you tell us anything about this machine?
A machine (or, a cluster of machines) on NVIDIA?s side was used to train the neural network that does both anti-aliasing and upscaling. This was done by NVIDIA and thus we cannot share more details on the training process ? exact details are their trade secret.
How long did it take to get from original scene to DLSS optimized? Any hurdles in the process?
The scene itself was not changed in any way ? it is exactly the same that is used in Port Royal (DLSS feature test references the exact same scene). Therefore, no artist work was required, and as stated earlier, engine integration was easy.
Can we expect a DLSS benchmark?
As clarified above, the NVIDIA DLSS test already functions as one. The test provides two scores ? one without DLSS, and one with it ? which can be used to measure performance delta between the two runs.
How long would it take to update and port an older benchmark like 3DMark 2001 to DLSS with new textures, we would really love to see a remake of 2001?
Effort to do this for one of our recent benchmarks would be roughly the same as it was for this Feature Test. Doing it for a legacy benchmark would require us to first update it to work with modern tools which would likely be a considerable effort.
Given that remastering old games is quite popular now this kind of an idea has a certain charm, but honestly it?s very unlikely.
Any word on a PCMark update? Hopefully better memory and storage tests?
New tests for PCMark 10 coming in 2019 include a Battery Life, Microsoft Applications, and Storage benchmarks. We expect battery to be available soon, with Microsoft Applications coming next, and storage being the last to release later this year.
The ORB has been a favorite for so many for two decades now, can we get some search functionality and possibly Hall of Fame for all benchmark?
Updating various features on 3DMark.com is a frequent topic at our office and an upgraded search is on top of the list. Can?t promise a schedule but we are well aware of the limitations of the current system and have a strong desire to improve upon it.
Is there anything else you can share with us coming up?
Not quite yet, but look to GDC 2019 for some very cool announcements from us.
Great work, great products, and thanks for your time.
Thanks Ryan, Joki and Jani!
Reply With Quote
rool:
Bookmarks