- Joined
- Feb 22, 2009
- Messages
- 791 (0.13/day)
| Processor | Ryzen 7 5700X3D |
|---|---|
| Motherboard | Asrock B550 PG Velocita |
| Cooling | Thermalright Silver Arrow 130 |
| Memory | G.Skill 4000 MHz DDR4 32 GB |
| Video Card(s) | XFX Radeon RX 7800XT 16 GB |
| Storage | Plextor PX-512M9PEGN 512 GB |
| Display(s) | 1920x1200; 100 Hz |
| Case | Fractal Design North XL |
| Audio Device(s) | SSL2 |
| Software | Windows 10 Pro 22H2 |
| Benchmark Scores | i've got a shitload of them in 15 years of TPU membership |
How many of you who have HEDT platforms have their memory placed in quad channel layout, if not because this is needed, but rather because it is ''cool'' to fill up your slots? Most folks doing everyday tasks do not actually need quad channel RAM - basic rendering, video encoding and gaming tasks don't really benefit from extra bandwidth and will see minimal to no gains. This is why the Ryzen 9 3950X & 5950X, released for mainstream platforms were so successful in the first place - most people who were building workstations or anything that ''sort of gimmicky'' would mostly do rendering and video production stuff, where the Ryzen chips suffered no penalty with their dual channel memory layout and made the LGA2066 competition look very bad price/performance wise, just as Intel processors did not benefit from quad channel memory in the applications which were selected for the reviews. However, be sure that there are plenty of less common, but important applications, which will benefit massively from quad channel layout, and i ain't just talking server stuff. This time around i am not going to rant how the reviewers of Ryzen 9 3950X did not include enough proper workstation apps in their benchmarks those would actually benefit quad channel systems - i will leave this for another thread. For now it is a pure RAM vs. RAM fight, and a benchmark for Techpowerup, which i have not done in a while.
That being said let's move on to the test which will be based on:
Windows 10
Core i9 9960X 16 cores/32 threads (4.0 GHz all core turbo)
Asrock X299 Taichi
2X16 GB DDR4 3200 MHz CL16-18-18-36-2N SINGLE RANK 1Rx8 RAM in dual channel mode
4X8 GB DDR4 3200 MHz CL16-16-16-36-2N SINGLE RANK 1Rx8 RAM in quad channel mode
It is very important that both memory configurations would only differ in their final bandwidth (128 bits vs 256 bits) and nothing more, meaning that memory size, memory ranks and memory timings have to be identical. The timings are for all sticks (primary, secondary and tertiary): 16-18-18-36-2-24-560-7-7-4-12-34-16-12480-8-4-7-4-7-1-4-1-1-5-6-6-6-1-1-1-3-5-3-3.
3D PARTICLE MOVEMENT 2.1 {higher score = better}
Six repeated tests, from which the best runs are selected show no real diversity with each run being some 1 % different from the previous.
7-ZIP 19.00 {higher score = better}
5 to 10 runs selected, results varying within 1 to 2 % show massive performance gains in compression tasks when using quad channel RAM.
7-ZIP 19.00 {higher score = better}
When doing decompression tasks this time around there is a constant difference (up to 2%) between dual and quad channel configurations, slightly favoring quad channel RAM.
BLENDER 3.5.0 {higher score = better}
Blender benchmark 3.1 uses 3 scenes for both CPU or GPU rendering and can simulate older or newer Blender versions. Using the current 3.5.0 version all 3 tests show absolutely no difference between dual and quad channel RAM, and all 3 scenes display consistent equal results, therefore i won't even bother showing the other two scenes.
V-RAY 4.10.7 {higher score = better}
V-ray, like the name suggests, is a ray traced renderer, that can be used for CPU, GPU or heterogeneous. This time around when using CPU we see the results within the margin of variation - less than 1 %.
LUXMARK 3.1 {higher score = better}
Using another proprietary renderer with the help of C++ libraries tells the same story. The ''ball'' scene, obviously, is there too, but i won't bother showing it as nothing will change.
HANDBRAKE 1.6.1 {lower score = better}
In this case i am encoding a custom RAW 8 GB 3840x2160 video to an output file using H.265, 8-bit, 3840x2160, 64 Mbp/s compression settings. I love handbrake as it is simple to use and provides incredibly consistent results with no margin or error - if the first run finishes at 840 second mark, the following runs end at the same time! This actually means that quad channel RAM config did the job 5 seconds faster here, but this clearly does not mean shit as it is a small difference.
Passmark 11 {higher score = better}
I am selecting some of the Passmark's tests which won't be duplicated in other benchmarks. After many runs CPU exteneded instructions settled the best results at over 2 % difference in a rather negligable favor for quad channel RAM
Passmark 11 {higher score = better}
Now, i am not sure what kind of physics are we talking here, but clearly this is important and the quad channel RAM users would benefit almost twice fold.
Passmark 11 {higher score = better}
I don't really have any comment on this, but who's ever the ''big time'' mathematician should know what's best for him.
Passmark 11 {higher score = better}
After repeated sortings the top scores settled at around 2.5 % in favor of quad channel RAM.
Sandra Lite 2021 {higher score = better}
Another synthetic benchmark doing some CPU arithmetics - pretty consistent results after several repeats and no difference in performance.
Sandra Lite 2021 {higher score = better}
This is a case where floating point operations are not influenced by the increased memory bandwidth, but more to come.
Sandra Lite 2021 {higher score = better}
And this is the case where floating point operations are heavily influenced by the increased memory bandwidth.
Sandra Lite 2021 {higher score = better}
Encrypting/decrypting files is not exactly a productivity workstation task, it's rather one of the most important server features, ''glad we have other Intel systems that support Xeon CPU and ECC RAM''...
Y-cruncher 0.7.7.9501 {lower score = better}
I wonder how would the Ryzen 9 processors perform here with their dual channel mainstream motherboards? Another reason for HEDT.
We are now moving on to the most important benchmarks in this thread - the SPEC Workstation standard, the proper tool to distinguish the HEDT platform from the mainstream. I am not going to comment under the pictures and will let the results speak for themselves. After all i do not posses the needed competency to explain what each SPEC benchmark does. I did exclude some of SPEC's tests as they were redundant and present in the previous programs. These tests are made from 3 runs, with the first run results always ignored as during the first run information and data is being filled into CPU cache, therefore the following runs are always more similar among themselves and faster too - i've noticed this behavior with the majority of SPEC's tests, and that kinda makes sense in a way.
SPEC WORKSTATION 3.02 {higher score = better}
The benefit of having quad channel RAM in your computer can not be denied, particurlarly looking at the SPEC Workstation results, where all the tests showed improvements vs. dual channel RAM layout - from minor to major. But such systems are not for everyone. If you are a casual photo & video editor, or a 3D animator, you don't need to look at these HEDT platforms - they will offer no real advantage. But if you are doing anything that the SPEC standard offers in it's benchmark suite, heck if you are archiving and encrypting/decrypting massive chunks of data, working with whatever the hell those Pi, Prime and floating point numbers mean - it's a no brainer to assemble a quad channel HEDT platform.
To be edited...
That being said let's move on to the test which will be based on:
Windows 10
Core i9 9960X 16 cores/32 threads (4.0 GHz all core turbo)
Asrock X299 Taichi
2X16 GB DDR4 3200 MHz CL16-18-18-36-2N SINGLE RANK 1Rx8 RAM in dual channel mode
4X8 GB DDR4 3200 MHz CL16-16-16-36-2N SINGLE RANK 1Rx8 RAM in quad channel mode
It is very important that both memory configurations would only differ in their final bandwidth (128 bits vs 256 bits) and nothing more, meaning that memory size, memory ranks and memory timings have to be identical. The timings are for all sticks (primary, secondary and tertiary): 16-18-18-36-2-24-560-7-7-4-12-34-16-12480-8-4-7-4-7-1-4-1-1-5-6-6-6-1-1-1-3-5-3-3.
3D PARTICLE MOVEMENT 2.1 {higher score = better}
Six repeated tests, from which the best runs are selected show no real diversity with each run being some 1 % different from the previous.
7-ZIP 19.00 {higher score = better}
5 to 10 runs selected, results varying within 1 to 2 % show massive performance gains in compression tasks when using quad channel RAM.
7-ZIP 19.00 {higher score = better}
When doing decompression tasks this time around there is a constant difference (up to 2%) between dual and quad channel configurations, slightly favoring quad channel RAM.
BLENDER 3.5.0 {higher score = better}
Blender benchmark 3.1 uses 3 scenes for both CPU or GPU rendering and can simulate older or newer Blender versions. Using the current 3.5.0 version all 3 tests show absolutely no difference between dual and quad channel RAM, and all 3 scenes display consistent equal results, therefore i won't even bother showing the other two scenes.
V-RAY 4.10.7 {higher score = better}
V-ray, like the name suggests, is a ray traced renderer, that can be used for CPU, GPU or heterogeneous. This time around when using CPU we see the results within the margin of variation - less than 1 %.
LUXMARK 3.1 {higher score = better}
Using another proprietary renderer with the help of C++ libraries tells the same story. The ''ball'' scene, obviously, is there too, but i won't bother showing it as nothing will change.
HANDBRAKE 1.6.1 {lower score = better}
In this case i am encoding a custom RAW 8 GB 3840x2160 video to an output file using H.265, 8-bit, 3840x2160, 64 Mbp/s compression settings. I love handbrake as it is simple to use and provides incredibly consistent results with no margin or error - if the first run finishes at 840 second mark, the following runs end at the same time! This actually means that quad channel RAM config did the job 5 seconds faster here, but this clearly does not mean shit as it is a small difference.
Passmark 11 {higher score = better}
I am selecting some of the Passmark's tests which won't be duplicated in other benchmarks. After many runs CPU exteneded instructions settled the best results at over 2 % difference in a rather negligable favor for quad channel RAM
Passmark 11 {higher score = better}
Now, i am not sure what kind of physics are we talking here, but clearly this is important and the quad channel RAM users would benefit almost twice fold.
Passmark 11 {higher score = better}
I don't really have any comment on this, but who's ever the ''big time'' mathematician should know what's best for him.
Passmark 11 {higher score = better}
After repeated sortings the top scores settled at around 2.5 % in favor of quad channel RAM.
Sandra Lite 2021 {higher score = better}
Another synthetic benchmark doing some CPU arithmetics - pretty consistent results after several repeats and no difference in performance.
Sandra Lite 2021 {higher score = better}
This is a case where floating point operations are not influenced by the increased memory bandwidth, but more to come.
Sandra Lite 2021 {higher score = better}
And this is the case where floating point operations are heavily influenced by the increased memory bandwidth.
Sandra Lite 2021 {higher score = better}
Encrypting/decrypting files is not exactly a productivity workstation task, it's rather one of the most important server features, ''glad we have other Intel systems that support Xeon CPU and ECC RAM''...
Y-cruncher 0.7.7.9501 {lower score = better}
I wonder how would the Ryzen 9 processors perform here with their dual channel mainstream motherboards? Another reason for HEDT.
We are now moving on to the most important benchmarks in this thread - the SPEC Workstation standard, the proper tool to distinguish the HEDT platform from the mainstream. I am not going to comment under the pictures and will let the results speak for themselves. After all i do not posses the needed competency to explain what each SPEC benchmark does. I did exclude some of SPEC's tests as they were redundant and present in the previous programs. These tests are made from 3 runs, with the first run results always ignored as during the first run information and data is being filled into CPU cache, therefore the following runs are always more similar among themselves and faster too - i've noticed this behavior with the majority of SPEC's tests, and that kinda makes sense in a way.
SPEC WORKSTATION 3.02 {higher score = better}
The benefit of having quad channel RAM in your computer can not be denied, particurlarly looking at the SPEC Workstation results, where all the tests showed improvements vs. dual channel RAM layout - from minor to major. But such systems are not for everyone. If you are a casual photo & video editor, or a 3D animator, you don't need to look at these HEDT platforms - they will offer no real advantage. But if you are doing anything that the SPEC standard offers in it's benchmark suite, heck if you are archiving and encrypting/decrypting massive chunks of data, working with whatever the hell those Pi, Prime and floating point numbers mean - it's a no brainer to assemble a quad channel HEDT platform.
To be edited...
Last edited:
