Test: CPU Core count and RAM scaling in ACC, AC1 and R3E
Awesome thread! Very interestibg results. Wpuld be interesting to compare different cpu clocks with different ram speeds. I have my 6700k at 4.5 (might be able to do more if I go for a serious OC) and 3000MHz ram. Might be interesting to oc both a bit more to get some easy gains. By the OP testing, it seems to be diminishing returns once you go pver 3200MHz ram speed which is interesting.
That's all I have.. Sorry, not really well executed. It's cinebenech r15 with 4 threads and 8 threads for the HT ones.
Or simply the multi test. But it's only really interesting for us to see other CPUs with 4 threads too (cinebenech, options, advanced, set specific threads and then putting in "4" and hit the "single core test").
Or simply put: click around until you see 4 squares at once
i7 2600k:
4.2, 1333: 533
4.2, 1600: 553
4.4, 1066: 570
4.4, 1333: 572
4.4, 1600: 582
4.4, 2133: 598
Without HT, 200 MHz CPU have a lot higher gain than higher memory speeds. 4.4 ghz + 1066 ddr3 is still clearly faster than 4.2 ghz + 1600. It's about 20 points.
However going from 1066 at 4.4 ghz to 2133 at 4.4 ghz gives another 30 points.
Interestingly the gain seems to increase at higher memory speeds.
4.4, 1066 + HT: 737
4.4, 1333 + HT: 738
4.4, 1600 + HT: 754
4.4, 2133 + HT: 771
4.9, 2133 + HT: 853
With HT on and 8 threads instead of 4 the gains are still there but not as high, percentage wise.
Overall synthetic benchmarks react more to ram speeds than games. BUT newer CPUs, especially amds react a lot more to memory speeds than my old i7
Yeah. I need to test this in game once i return. I dislike synthetic benchmarks for gaming results. I mean, they are useful but the best metric is to run the game imho.That's all I have.. Sorry, not really well executed. It's cinebenech r15 with 4 threads and 8 threads for the HT ones.
Or simply the multi test. But it's only really interesting for us to see other CPUs with 4 threads too (cinebenech, options, advanced, set specific threads and then putting in "4" and hit the "single core test").
Or simply put: click around until you see 4 squares at once
i7 2600k:
4.2, 1333: 533
4.2, 1600: 553
4.4, 1066: 570
4.4, 1333: 572
4.4, 1600: 582
4.4, 2133: 598
Without HT, 200 MHz CPU have a lot higher gain than higher memory speeds. 4.4 ghz + 1066 ddr3 is still clearly faster than 4.2 ghz + 1600. It's about 20 points.
However going from 1066 at 4.4 ghz to 2133 at 4.4 ghz gives another 30 points.
Interestingly the gain seems to increase at higher memory speeds.
4.4, 1066 + HT: 737
4.4, 1333 + HT: 738
4.4, 1600 + HT: 754
4.4, 2133 + HT: 771
4.9, 2133 + HT: 853
With HT on and 8 threads instead of 4 the gains are still there but not as high, percentage wise.
Overall synthetic benchmarks react more to ram speeds than games. BUT newer CPUs, especially amds react a lot more to memory speeds than my old i7
Just been playing my 3600 (sans X). 4300mhz for CCX 0, 3600 mhz for CCX 1. SMT ON, limit ACC (for example) to first 6 cores in Process Lasso. Seems to work really well.
When I said ACC optimization guide, this is what I meant:
https://www.nvidia.com/en-us/geforce/guides/control-graphics-and-performance-guide/
https://www.nvidia.com/en-us/geforce/guides/control-graphics-and-performance-guide/
Reflections, shadows etc are also increasing the cpu load, not only the graphics card's load.
Since you don't really know how the rendering thread(s) are split, maxing out everything will guarantee you the highest cpu load possible.
The resolution on the other hand has barely any impact apart from loading the pci bus and influencing the GPU load directly.
But you have a point. If the GPU load still goes above 90% it would make sense to lower settings too to keep being cpu limited.
StefanoCasillo
Jaxx Vane Studio
To hit the CPU in AC I would keep reflections at 6 per frame, at the lowest resolution.
Shadows, lowest resolution, CPU load won't change, it's still 3 passes.
Post processing all to the minimum or off.
MSAA off.
This should force the CPU to go over the scene 11 times per frame:
6 passes for reflections
3 passes for shadows
1 pass for mirror
1 pass for the main scene
Shadows, lowest resolution, CPU load won't change, it's still 3 passes.
Post processing all to the minimum or off.
MSAA off.
This should force the CPU to go over the scene 11 times per frame:
6 passes for reflections
3 passes for shadows
1 pass for mirror
1 pass for the main scene
D
Deleted member 197115
-
Deleted member 197115
6 passes for reflections in number of faces settings.
Why 3 passes for shadows, does it depend on shadows resolution?
Why 3 passes for shadows, does it depend on shadows resolution?
StefanoCasillo
Jaxx Vane Studio
no it's always 3 regardless of resolution.
So I did some more testing with HT on vs off. This isn't FPS related directly. It's more about how much burden HT puts on the cooling system.
- Test done using Cinebench R20 since it has AVX and runs longer than R15
- Haf X case with fans on max and Dark Rock 4 air cooler (non pro)
- Scenario: HT OFF. 9900k@5.2ghz 1.32v in bios with llc high. Result: vrvout = 1.235v mid run. core max = 79c cpu package = 77c
- Scenario: HT OFF. 9900k@5.3ghz 1.39v in bios with llc high. Result: vrvout = 1.290v. cpu package and core max = 87c
- Scenario: HT ON. 9900k@5.1ghz 1.32v in bios with turbo llc. Result: vrvout = 1.254v thermal throttling during test. core max 98c cpu package 97c < incomplete result
Summary: To match the thermal envelope, I'd need to run the HT version at 5.0ghz while the NON HT setup runs at 5.3ghz. For most games, especially sim racing and VR, the 300mhz is going to be a notable improvement in minimum FPS over having HT on which will often hurt performance even if all things are equal.
- Test done using Cinebench R20 since it has AVX and runs longer than R15
- Haf X case with fans on max and Dark Rock 4 air cooler (non pro)
- Scenario: HT OFF. 9900k@5.2ghz 1.32v in bios with llc high. Result: vrvout = 1.235v mid run. core max = 79c cpu package = 77c
- Scenario: HT OFF. 9900k@5.3ghz 1.39v in bios with llc high. Result: vrvout = 1.290v. cpu package and core max = 87c
- Scenario: HT ON. 9900k@5.1ghz 1.32v in bios with turbo llc. Result: vrvout = 1.254v thermal throttling during test. core max 98c cpu package 97c < incomplete result
Summary: To match the thermal envelope, I'd need to run the HT version at 5.0ghz while the NON HT setup runs at 5.3ghz. For most games, especially sim racing and VR, the 300mhz is going to be a notable improvement in minimum FPS over having HT on which will often hurt performance even if all things are equal.
Last edited:
D
Deleted member 197115
-
Deleted member 197115
Nice results. Would be interesting to see how that translates to real performance in games.
I'd assume most synthetic CPU benchmarks will be low with HT off despite higher core clock, curious about single thread benchmark though.
I'd assume most synthetic CPU benchmarks will be low with HT off despite higher core clock, curious about single thread benchmark though.
Single thread would naturally be higher as frequency scales in a linear fashion.
In older games, HT can create issues where as straight cores are rarely a problem. If you review the results Dan posted in the OP, you can see that adding cores never creates a problem. However, adding HT can and does.
You could start reducing the number of physical cores to get more frequency as you'll have more thermal headroom and require a lower vcore.
D
Deleted member 197115
-
Deleted member 197115
Even with Win10 quirky thread scheduling and presence of all those other things going on on computer at the same time?
Which part?
What's "quirky" about the scheduler? Also, what "other things" are you talking about?
D
Deleted member 197115
-
Deleted member 197115
Quirky in a sense that it moves single thread around cores.
Other things, well all those 100+ background processes running on normaly operated OS.
You trade number of workers for their strength.
HT on or off was always controversial subject, results might not be what you expect.
Plus with silicone lottery there is no guarantee you can push clock with HT off that much higher to compensate for lost cores.
Other things, well all those 100+ background processes running on normaly operated OS.
You trade number of workers for their strength.
HT on or off was always controversial subject, results might not be what you expect.
Plus with silicone lottery there is no guarantee you can push clock with HT off that much higher to compensate for lost cores.
Read the op and then others have done more testing with sims.
There's no absolutely simple answer to HT but on intel, on 1151, you'll certainly get more headroom as you'll have more thermal headroom and require less vcore.
