3700X SysSetHack

[Adi2071]

Hello,

I am trying to use this program but unfortunately I am not sure if it works.

I did various things to get SysSetHack to show at least 1/16. Unfortunately, I can't force to check if this program actually works as it should. Please help.

[milkmaman]

Did you try 1.1.7 b6? Linked in this thread on overclock.net.

https://www.overclock.net/threads/cpudoc-little-cpu-helper-tool-with-some-exclusive-features.1802081/post-29104334

[mann1x]

Sorry I missed the notification Please check with the beta as said by @milkmaman

[Adi2071]

What are the best PBO settings and voltage for the 3700X for the best gaming performance? I can set my CPU to 4450MHz 1.368V

[mann1x]

I use PPT 135 and TDC 90 for my 3800X at EDC 1A. You can use ZenControl with these values leaving EDC 1A not Auto. ZenControl will lower the PPT/TDC based on load to get the best gaming perf.

BTW there's no reason to enable NumaZero with a 3700X unless it's a dual CCD and yours is not.

[Adi2071]

Is voltage reduction needed or will it reduce gaming performance? For EDC 1 I used -0.01875V.

[mann1x]

It depends on your system; CPU, board, bios version, etc It's not needed, only set an offset if it helps increasing the clocks or performances

[Adi2071]

For EDC 1, the voltage offset has no major effect. I tested -0.0125, -0.01875 down to 0.0375V. The highest efficiency is at -0.01875V but these are 0.1% values.

[mann1x]

It looks good. Even if very small the offset can have a very big impact.

Run some game benchmarks; for gaming it's probably better to keep the PowerTweak at AutoPower and ActiveMode/GameMode in Auto. But it could be different by system and also I haven't tested on Win11.

[Adi2071]

Nice program, satisfactory performance. Where you can check CPU performance at 1440p and low details (to get 165 fps+). I would like to test the settings with your program vs 4450MHz 1.368V.

[mann1x]

Depends on the game; best is Shadow of the Tomb Raider to test CPU performances.

[Adi2071]

Ohh nice. AutoOC with your powerplan and software I see on 1080p ultra low 197fps. On 4450MHz 1.375V i see 201 ( only once) later 183-184. The first time I ran the test in 1080p I saw 192fps. Second time around 197fps on average. Probably the program put my processor in Booster mode. Or maybe it was better able to use the GPU.

[Adi2071]

I have an asus motherboard. In the PBO settings I have an option like FMax Enhancer. This is the option responsible for EDC. According to specialists in this function, the 3700X should achieve clocks higher than 4400MHz. Exactly this function is responsible for the EDC bug but it could be better optimized. I have a feeling that the cpu is throttled when EDC is set to 1. I would have to enable the Fmax function. Adjust the voltage offset to be exactly the same as EDC 1.

[mann1x]

The FMax Enhancer was a nice idea but most of the times doesn't work and makes the system unstable or slower due to massive clock stretching. You need to be careful with it.

Don't trust feelings and use benchmarks; Unigine Heaven, 3DMark, Geekbench 5, CPU-z, etc Use my BenchMaestro to check if you have clock stretching. The 3700X could boost above 4400 MHz but it's not always possible. My 3800X without the EDC bug doesn't go over 4550 MHz, while with EDC 1 tops 4625 MHz. The best you can get with the EDC bug is probably 4550 MHz but that depends on the binning quality of your CPU.

Use the benchmarks, especially Geekbench 5, to find the best setting for EDC. It's not always 1 the best; I usually test 1-2-3-5-7-10. You may also need a little bump up in SOC Voltage. Very often with the EDC bug to get the clock boosting higher you also need a small negative or positive Vcore offset. It all depends on the specific sample so you have to test yourself.

[Adi2071]

My play maxes out at 4450MHz. My sister's computer is similar, 4425MHz. I haven't seen bug 3700X reach more than 4425-4450MHz on EDC. The 3800X can actually go up to 4600MHz. My 3700X unfortunately not.

[mann1x]

The 3800X has only 100 MHz more boost clock but it's much better binned than the 3700X so it's much easier to get a better one. I'd check and experiment more with the settings cause there's a chance that it can go better with the EDC bug, at least up to 4500 MHz.

[mann1x]

You may find something interesting reading this article:

https://www.hisevilness.com/articles/technology/ryzen-7-overclocking-the-3700x.html?showall=1

Consider testing with Global C-State control to disabled as often is causing issues with the EDC bug. Don't disable Core Performance Boost or PBO of course.

Very important also to tweak DIGI+ VRM settings but be careful with the Power Duty/Phase to Extreme as often are causing instability. Also the options that can be set at 120-130% can be messy and more stable with just a notch below.

[Adi2071]

If you can improve the program, setting the PBO so that I can get a better result, I would ask for help. I currently have 145/95/1. CPU draws 135-138W under stress in cinebench r23.

[mann1x]

Just adjusting PBO limits it's probably not going to help. You need to experiment and benchmark with different BIOS settings. Higher limits and power draw doesn't mean necessarily better performances. Use Geekbench 5 and compare the scores with different settings. I use 135/90/1 on my 3800X so it's probably too high at 145/95.

[Adi2071]

Global C-state Control = Disabled Power Supply Idle Control = Low Current Idle CPPC = Enabled CPPC Preferred Cores = Enabled AMD Cool'n'Quiet = Disabled PPC Adjustment = PState 0

I changed PBO to 135/90/1.

[mann1x]

Try with C-state and CnQ enabled if you didn't already. The LLC seems too high; try with CPU set to Auto or 3. SOC it's probably fine at 4, try with 3.

[Adi2071]

SOC not possible at level 3. Then the RAM is unstable.

[Adi2071]

When these options are enabled, the processor is more stable. I set the CPU LLC to Auto. Now CPU eats 133W so PPT 135W is enough. Previously at 135W it was throttled because it used 138-139W.

[Adi2071]

After enabling C-State and CnQ, single-core performance increased from 527 to 531 points in CPU-Z. Manual OC 4400MHz gets 534 points, 4450 gets 541 points, 4500MHz over 550 points. Unfortunately, manual OC has much worse performance in games than AutoOC with PBO and EDC bug.

[Adi2071]

Version 1.1.8 is broken. The processor shuts down at various times. Not only on my computer, but also on another computer. Single-core performance in cpu-z is around 529-531 points. In the new version 1.1.8 it drops dramatically.

[mann1x]

Very odd, it's almost identical to 1.1.7 except a few small fixes. Are you sure you didn't change anything else? Did you upgrade from a previous version? If that's the case try with a clean install.

Also upgrade CPU-z to the latest version, 1.91 is very old and CPU-z it's pretty finnicky.

If you still have issues try disabling SysSetHack and/or PowerSaverActive in settings to find out which one could be the root cause.

[mann1x]

Very often the CPU-z ST bench falling down is a symptom of Global C-state not working well with EDC bug. Sometimes works and then after a few reboots does this. Try if disabling it in BIOS helps.

[Adi2071]

Unfortunately not. On the second computer as well as on my version 1.1.7 everything was fine. When he updated to version 1.1.8 it started crashing on both computers.

When I disable SysSetHack nothing changes.

[Adi2071]

Disabling PowerSaver doesn't change anything. Only disabling the program or ThreadBooster option fixes the problem.

[mann1x]

Try disabling ZenControl. Stopping ThreadBooster it's almost as closing the program.

Did you try with the 1.1.7 you were using before?

[Adi2071]

Disabling ZenControl resulted in drops to 510 points. I thought it was ZenControl's fault and suddenly it dropped to the value as in the picture above.

On version 1.1.7 everything is fine. Everything works fine on both computers.

[mann1x]

Very puzzling... Did you update CPU-z to the latest version?

[Adi2071]

"Did you update CPU-z to the latest version?"

There is no need. No program matters. The problem is in your software. My sister played valorant. The clocks dropped to 3440MHz and the game was 70 FPS. Then the clocks increased to the correct value and FPS went up. I turned off this program for my sister so she could finish the match and 0 problems until the end of the match.

It doesn't matter if I'm testing it with cinebench, cpu - with or I'm on the desktop, HWinfo shows how the clocks drop during the test below 3.6GHz.

Most likely, you have messed something up and for the program to work properly, you need to turn off C-State because with it on clocks fall below 3.6GHz.

[mann1x]

CPU-z version matters a lot! It's one of the few programs that handles affinity on its own, not using Windows. There are very specific optimizations for it in the code and they could not work the same on the old versions. I only test it with the latest one.

If the clocks are going down during gaming then it's a whole different story. Probably the issue is not related to CPU-z but please use only the latest version.

Did you made a fresh install of 1.1.8? Do you have a 2nd monitor to check while gaming what is the PerfBias (Economizer/Standard/Booster) and if GameMode is enabled?

[Adi2071]

PowerSaverActive [Booster] - Even on the desktop, all the time.

There is no need to update CPU-Z, it does not affect what is happening with the processor. I tested the previous settings on this version even when I was using 1.1.7.

Something in this program is broken. I'll try disabling C-State and see if it still does the same thing.

This is not a clean install, just an upgrade from 1.1.7. It doesn't matter basically.

[mann1x]

You need a clean install, not upgrade. I ended up in this situation as well with 1.1.7 and I don't know why. Clean install fixed it. I must have missed something but I don't know where.

Before disabling C-state try it.

You can keep using an old CPU-z but if it's doing weird stuff I can't help.

One of the few things that I may have added after 1.1.7b6 could be the Vanguard protection detection, have to check. Maybe that could be an issue as well.

[Adi2071]

Disabling C-State fixes the problem.

I will enable C-State and do a clean install and let you know.

[mann1x]

No the Vanguard detection was already there; if it's detected ZenControl will not work as it stops the library

[mann1x]

Thanks for testing!

Sadly C-state with the EDC bug most often doesn't work. I was one of the few that could but at some point it stopped working on my 3800X and had to disable it as well, some crazy stuff started happening randomly all of a sudden. I think the version of the AMD Chipset Drivers plays a role. Update them if you didn't recently.

[Adi2071]

I have the latest drivers on both computers.

How to do a clean install? There is no .exe file in the download package to unload the program.

[mann1x]

Just don't use the previous folder and extract the zip file in a new folder. Re-configure the settings if different from the defaults.

If you had Autostart set at boot, remove and re-create the task.

[Adi2071]

There is no .exe file to run the program.

[mann1x]

There is... Maybe the AntiVirus somehow messing it up?

[Adi2071]

I'm getting this.

[mann1x]

You are downloading the source code to compile :)

Go here: https://github.com/mann1x/CPUDoc/releases/tag/v1.1.8

And download CPUDoc-v1.1.8.zip

[Adi2071]

A clean install didn't help. C-State must be disabled.

Then what about the option AMD Cool'n'Quiet ? Should it be enabled or disabled?

[mann1x]

The Cool'n'Quiet option is quite a mystery. There's no such thing in Ryzen 3000 and later... It does something but I don't recall exactly what.

On some boards if disabled PBO boosting will be limited. I would keep it enabled if it doesn't creates issues, worst should happen is that does nothing.

[Adi2071]

Everything works fine now but the performance in csgo is quite low.

I don't think cutting the cores in half has any good effect on performance. SMT and single CCD applies to new generation processors like 7900X/7950X, not age like 5900X and 5950X and most importantly only on Windows11 there is such a performance problem.

[mann1x]

The performance of a game with SSH depends on a number of factors. Some games does not show any improvement. Others lower or even stuttering. In general you need to configure the graphics settings in a way that the CPU usage does not go over 50%; other wise CPUDoc will start enabling the T1 threads and more are enabled and lesser and lesser the performance gain will be.

SMT/HyperThreading applies to almost all processors produced in the last 10 years; more cores, more gain. Using Single CCD/single Numa is a good way to reduce latency, if it doesn't cause issues which sometimes does. Lots of people with old Intel CPUs are using CPUDoc SSH to get more consistent fps with CS:GO or other high fps games.

I'm using it happily with a 3800X and a 5950X, you don't need a latest gen CPU.

If you want to test the performance gain on your CPU check the README, use 7-Zip File Manger benchmark mode with 6 threads, run it with and without SSH

[Adi2071]

I'm playing BF1 and the cpu usage on some threads reaches 70-80% but still 8 cores out of 16 are used. I'm using MSI afterburner and I have all the threads displayed there.

I don't really understand how this program is supposed to work. In dying light i think there is less FPS than on full 16 cores same in csgo. Performance with SMT OFF can only improve in a few % of games. There have been so many tests and in general it turned out that disabling the cores does not make much sense because the profit is meager or non-existent.

[mann1x]

There are 3 areas of improvements with SSH:

• Raw performances; it's about 1% per core, you can verify it yourself using any benchmark like 7-Zip mentioned above or with 3DMark CPUProfile. Doesn't look much but it is. If you consider the money and efforts and risks saved against an OC which often results in the same. On a 16 cores like a 3950X or 5950X it's about 15% gain in speed. It's massive if you are in the right conditions. Just the power consumption savings are worthwhile if you consider that normally a PC or a Server runs most of it time with a CPU usage around or below 50%.
• Processes Priorities: manipulating the CpuSet in Windows has the effect to alter the priority of process scheduling. If you start a game installation while playing a game your fps will go down massively. If you do the same with SSH enabled the fps will go down but much less.
• Low FPS improvements: almost all games, also those which will have the same average fps, will gain much better lows. You can use any in-game benchmark to verify this. If they don't show the P95/P99 fps you can use CapframeX. This is because of the nature of SMT/HyperThreading; they are not 2 cores but 2 threads on the same core and they share resources. Anything that is running on the 2nd thread can and probably impact the execution of what is running in the 1st one.

SMT Off is nor recommended because you effectively kill almost half of the max performances of the CPU. CPUDoc does it dynamically, that's the point. Having the advantages without the disadvantages. SMT Off is used by most pro-players because it's the only way to get really smooth gameplay fps. You just need an oversized CPU cause you are going to use half of it.

I suggest you look in the benchmarks I've posted in the AMD thread on Overclock.net, it's in the README.