Upgrade to 10920X computer – Experience

Report · Apr 09, 2020

I made a new computer for my Premiere Pro processing based on the Puget Systems 6K/8K Editing Workstation. I recycled two NVMe M.2, power supply, Blu ray writer and graphics card from my old computer.

The specs are as follows:

CPU Intel I9-10920X 3.5GHz, 12 cores, 24 hyper threads

Gigabyte X299X Designare 10G which included an AORUS Gen4 AIC card for four NVMe M.2 drives

128GB Crucial Ballistix 2666 8x16 RAM

1TB Samsung 970 EVO Plus NVMe M.2 (System)

2TB Samsung 960 Pro NVMe M.2 (Media)

500GB Samsung 950 Pro NVMe M.2 (Cache)

Noctua NH-15 Air Cooler

Fractal Design Define R6 USB-C Case

EVGA Geforce GTX1080 Graphics Card

Seasonic Platinum 860W Power Supply

LG BluRay writer

I usually make a new computer for my Premiere processing every 3-4 years. I waited an extra 6 months to get the new Intel I9-10920X CPU. Initially, they were offered for $650, but they were very hard to get. I saw Amazon selling them for $1999 by third party. I got on a waiting list with BH Photo and got it for $812.

I wanted to upgrade because my old computer (Intel i7-5960X (3.0 GHz, 8 cores), Asus X99 Deluxe MB, 32GB Crucial Ballistix RAM, 2TB Samsung Pro NVMe M.2, GTX1080 Graphics) was having trouble with the Premiere Pro CC 2020. It was crashing very frequently. It was fine with PP CC 2019.

I was really impressed by the Fractal Design case. All of the components were high grade. Even the screws were black oxide coated and not chrome. It uses a lot of thumb screws. It was very easy to assemble my computer due to good access. Most of cable runs were hidden due to the false bottom. It was very quiet, but pretty heavy.

I got a Noctua NH-15 cooler instead of the Puget Systems recommended NH-12, because the NH-15 used 140mm fans, instead of NH-12’s 120mm fans. I thought the NH-15 would be quieter and it was cheaper. The NH-15 fit in the Fractal case with about ¾ inches to spare.

I chose the Crucial Ballistix RAM because it had low profile heat sinks. Otherwise, I couldn’t fit one of the Noctua 140mm fans.

I got a new 1 TB Samsung EVO Plus, so that I could have three NVMe M.2 drives (System, Media, Cache/Scratch).

Originally, I intended to put my three m.2 drives on the motherboard. One of the mounts didn’t have a screw in socket. So I decided to use the AIC board. I think it was intended for a Gen4 m.2 RAID drive. It can do 15,000 MB/S with four 2TB Gen 4 m.2 drives. Fortunately, it can also run multiple Gen3 m.2 drives as individual drives. It has a very large heat sink and a blower. It is easier to change out the M.2 drives than if they were on the MB.

I transferred my old computer’s system m.2 drive to the new computer and ran Puget System’s Puget Bench for Premiere Pro using PP CC 2020 4.0. There was a 73% improvement (340 to 583). There were several factors that improved the performance. My old cpu ran at only 3.0MHz. Originally, I overclocked it, but it was crashing too much. My new cpu had automatic over clocking. It was usually at 4.3MHz, but would do occasional bursts to 4.7MHz. Cpu temps were as high as 80C. My old cpu had only 8 cores vs 12 cores for the new CPU. I had 128GB of ram vs 32GB for the old one. My Samsung 960 Pro 2TB increased in speed to 3000MB/S from 2500MB/S with the new MB.

I did a clean install and made a Samsung 970 EVO Plus the system disk. The performance improved from 583 to 611. Part of this increase may be due to increasing the system/media disk speed from 3000MB/S to 3500 MB/S.

I found that PP CC 2020 4.04 Puget Bench performance was a lot lower (532) than PP CC 2020 4.0 (611).

My system’s Puget Bench performance was a lot lower than the Puget Systems made 6K/8K Editing Workstation (748.5). The major difference was they are using a RTX 2080 Ti 11GB and I had a GTX 1080 8GB. One of their previous publications showed that the RTX 2080 Ti 11GB was only 3% better than my GTX 1080. I don’t know why there is such a difference.

I am happy with my new computer. The response is much better and it crashes a lot less frequently.

Report · Apr 10, 2020

Were your scores attained from using the "standard" preset? Or the "extended" preset?

If those scores were achieved using the "standard" preset, your old system's score was actually as slow as, if not slower than, my reserve i7-7700 4-core/8-thread CPU-based mini-ITX PC with only SATA SSDs, 16 GB of RAM and a GeForce GTX 1050 Ti GPU (a GeForce GTX 1060 6 GB GPU made only a minor improvement to that system's overall score). And your 12-core/24-thread new setup is no faster overall than my more humble 8-core/16-thread AMD Ryzen R7 3800X PC with only 32 GB of RAM, a Samsung 970 EVO Plus m.2 SSD plus two SATA SSDs and a GeForce RTX 2060 SUPER GPU. That latter result is not surprising, given the GTX 1080 being somewhat weak in comparison to most of the RTX 20-series GPUs.

And your comparison between a GTX 1080 and an RTX 2080 Ti was based on the results with an earlier version of the Puget Systems benchmark, which used a frame rate of 29.97 fps for the Standard preset. Puget Systems has since changed the benchmark presets to now use 59.94 fps video in its Standard preset instead of the 29.97 fps that was used previously. At 59.94 fps, the differences become significantly greater than the 3% that was reported before. So in other words, all Pascal GPUs, including yours, as well as everything that Nvidia had produced prior to and including Pascal, turned out to be choke artists compared to even the absolute lowest-level Turing (a non-SUPER GTX 1650). Or put it another way, even a GTX 1080 Ti would have pulled a choke job in these 4k/59.94 fps video tests compared to even the aforementioned GTX 1650.

For the record, my R7 3800X PC, despite its CPU having only 8 cores, scored a 611 in my most recent Puget Systems test with the 14.0.4 release of Premiere Pro 2020. My i7-7700 reserve PC averaged only a 344 on that same test - and that's with that CPU's QuickSync enabled.

Randall

Report · Apr 12, 2020

Hi Randall,

Thanks for your comparison.

I was using the standard Pugetbench test.

It looks like my GTX 1080 is bottle necking my system.

I will upgrade to a RTX 2070 Super. I don't think my requirements require a RTX 2080 Ti.

Roy

Report · Apr 12, 2020

That RTX 2070 SUPER should bring that standard overall score from the PugetBench 0.9 tests closer to 700 with that CPU. You will not be able to match the 714 "standard" overall score that Puget Systems itself had gotten with that CPU but with an RTX 2080 Ti.

Report · Apr 16, 2020

Hi Randall,

I replaced my EVGA GTX1080 with a Gigabyte Aorus RTX2070 Super and increased my PugetBench 0.9 Standard Score from 607 to 769! This beats the Puget Systems 10920X with a RTX2080 Ti score of 738.

The Aorus has auto overclocking to 2100 MHz.

Also I just updated the NVIDEA Studio Driver to Version 442.92 that was released today (4/16/2020).

Thanks for your suggestions.

Report · Apr 11, 2020

A follow-up to this:

Puget Systems has updated the benchmark since the version 0.88 that we ran. We had to run the "standard" preset because our systems crashed whenever the 8k H.265 part loaded. The new version 0.9, which has been available since the first week of this month, no longer includes the 8k H.265 timelines in its Extended preset. And the GPU and CPU effects tests have also changed.

With these changes, my R7 3800X system scored a 637 and a 608 in the "Extended" and the "Standard" tests, respectively. This clearly indicates that something, primarily the GPU, is holding down your system's performance.

I will be testing my i7-7700 PC with the GTX 1060 with the new tests, with QuickSync both enabled and disabled.

The results from that dimunitive system will be reported on shortly.

Report · Apr 12, 2020

I have just run the new Puget benchmark with my system.

Windows 10

Intel i9-9900K

64 GB RAM

RTX 2070 Super

1Tb Samsung SSD 860 boot drive

1Tb Samsung M2 projects and Export

500 Gb Samsung M2 Cache

Puget Results:

Standard Overall 692

Standard Live Playback 70.3

Standard Export 68.1

GPU Score 59.5

Report · Apr 12, 2020

Richard,

Those results just tell me that the thread starter's overall results (using the previous version) to be mostly GPU-bound. And the newer tests really bear that out. In fact, I actually compared my system's score to the ones that had already been posted online, and discovered that one of the results in that database was a "standard" overall score of an abysmal 15 from a 11-year-old i7-860 system with 8 GB of RAM and an ATi Radeon HD 5000 series GPU - I was very surprised that Premiere Pro 2020 even ran at all on that dinosaur.

A much more recent AMD Ryzen R3 3200G with 16 GB of RAM and only integrated Radeon Vega 8 graphics scored almost as badly as that obsolete Lynnfield system, with a standard overall score of around 30-ish.

I took out my reserve i7-7700 system with 16 GB of RAM and a GeForce GTX 1060 6 GB card, and my testing with QuickSync disabled was completed early this morning. It came out with an extended overall score of 346 and a standard overall score of 309. The results from that same system with QuickSync enabled will be reported shortly. However, the GTX 1060 still bottlenecks that i7-7700 greatly, with a GPU score that was barely more than half the overall live playback and export scores for that system. This micro desktop system really needs a GTX 1660 SUPER just for it to be balanced in performance.

Report · Apr 14, 2020

I didn't realise until I ran the Puget test with the hardware monitor open that Premiere uses QuickSync for H264 decode as well as encode. Playing back the Policecar scenes the inbuilt intel graphics was running at about 50%

Report · Apr 14, 2020

Just installed todays Premiere update 14.1.0 Build 116 and the Puget benchmark now crashes every time.

Report · Apr 16, 2020

I just saw your new score, posted today, after you received your RTX 2070 SUPER. That 769 "standard" score is more like what I had expected from that system all along. There was a small boost with the 128 GB of RAM as well as another level of boost from your new replacement drives (versus a single drive plus OS drive in the Puget Systems' own test benched systems) whereas Puget Systems' scores were with 64 GB of RAM.

It turned out that everything GPU that Nvidia made prior to Turing choked at 4k/59.94 and above, and were no better in CUDA mode than AMD Radeons' same-generation GPUs in OpenCL mode at that resolution and frame rate.