Brief Report on Silicon Quality of 3700X, Batch 2036SUS

[Sabotaged_Enigma]

Preface​

Still a student, bearing the annoying terrible quality of my 2600X for more than a year, I finally got the money I needed to buy a 3700X and B550 mainboard.
This is not an extreme test, with a hope for it to be surprising.
The test was done in January of this year.
I hope this test could give you another aspect of Zen 2 overclocking.

Setup​

Mainboard: Asus TUF Gaming B550M-Plus (Wi-Fi), BIOS version 1202, all settings auto
CPU: AMD Ryzen 7 3700X (2036SUS), SMT on, all cores on
Radiator: Thermalright AS120 + stock TF7 toothpaste
System: Windows 10 Pro 1909, B550 chipset driver version 2.10.13.408

Testing Method​

Control the clocks and voltages via Ryzen Master, and test it with CPU-Z's version 17.01.64 benchmark and stability test. At the same time, focus on the ratio of single-threaded and multi-threaded benchmark scores to ensure there's no thread drop.
Don't blame me for not giving it the most stress. As I have stated at the beginning, this is not an extreme test, whose purpose is to explore the frequency-voltage relation.

Testing Process​

Benchmark at 3.6 GHz 1 V as a reference. After three days of carefully twicking, I've got the results in the chart below.

Notes:
1. Frequency and voltage in grey colour are not stable;
2. From left to right, the lables are: a. Frequency, b. Voltage of a stable status, c. Voltage of a relatively stable and extreme status, d. single-threaded score, e. theoretical single-threaded score, f. multi-threaded score, g. theoretical multi-threaded score, h. the difference of d and e, i. the difference of f and g;

From 3.9 GHz to 4.6 GHz, the ratio kept almost stable with an average of 11.07, a maximum of 11.08 and a minimum of 11.05, indicating that the scores are valid and there's no thread drop.

Analysis​

To be honest, I've never expected my 3700X to be this good at overclocking, though when benchmarked at 4.6 GHz 1.3375 V, Ryzen Master showed a temperature of 75 ℃. It probably explains why it can't go any higher in regard to very high heat density on the small die chip.
We've got this chart of the most important stats.

And we've got a graph according to this chart.

Notes: This is not a linear equation. It is growing faster and faster. The Maths analysis indicates that this is very close and accurate and the results are valid.
V = 0.1306*e^(0.5062*F), R^2 = 0.9995 ≈ 1

Assume that there's no problems with heat sink, we can predict this 3700X to be overclocked in a trend like this:
4.65 GHz - 1.3747 V
4.70 GHz - 1.4099 V
4.75 GHz - 1.4460 V
4.80 GHz - 1.4831 V
4.85 GHz - 1.5211 V
4.90 GHz - 1.5601 V
4.95 GHz - 1.6001 V
5.00 GHz - 1.6411 V
...... (Any volunteer who doesn't fear explosion? lol)

Besides, my 3700X could pass the stress test in CTR at 4.05 GHz 1.0375 V (1034mV).
That's all the test is about. I hope it could give you some ideas. Thx.

This article has already been published by myself on www.bilibili.com, with all copyright reserved.

 

[WHDS]

I've got a 2035 r5 3600 and it does 4.525 GHz at 1.35v in every benchmark I've thrown at it, the later model chips seem to have significantly better silicon quality as I've seen many people struggle to get even 4.1 all core with the earlier chips.

Nice article btw!

 

[Sabotaged_Enigma]

I've got a 2035 r5 3600 and it does 4.525 GHz at 1.35v in every benchmark I've thrown at it, the later model chips seem to have significantly better silicon quality as I've seen many people struggle to get even 4.1 all core with the earlier chips.

Nice article btw!

Thanks a lot! I've used TPU website for a long time, but I'm still a green hand in the TPU forum. Look, I've posted this article in the GPU section lol... Is there a way to change this article into the correct section?

 

[tabascosauz]

My my, that's a golden chip if I've ever seen one. Treasure it! If the 'stress testing' is through CTR AVX-light, that's pretty damn good even for a late production chip.

My 3700X did 4.0 @ 1.194Vmin. Batch 1924PGT.
My 4650G does 4.0 @ 1.206Vmin. Batch 2037SUS.

I will say though, probably not a good idea to stress past 1.35V on CTR AVX without a big AIO at the very minimum. My 4650G used to do 4.3 @ 1.35V - just a week of furiously heavy R23 benching trying to get a 4.4 score definitely abused it a bit and now it struggles to manage even 4.1 @ 1.3V.

Awesome writeup!

 

[Sabotaged_Enigma]

My my, that's a golden chip if I've ever seen one. Treasure it! If the 'stress testing' is through CTR AVX-light, that's pretty damn good even for a late production chip.

My 3700X did 4.0 @ 1.194Vmin. Batch 1924PGT.
My 4650G does 4.0 @ 1.206Vmin. Batch 2037SUS.

I will say though, probably not a good idea to stress past 1.35V on CTR AVX without a big AIO at the very minimum. My 4650G used to do 4.3 @ 1.35V - just a week of furiously heavy R23 benching trying to get a 4.4 score definitely abused it a bit and now it struggles to manage even 4.1 @ 1.3V.

Awesome writeup!

Thanks! Yeah, CTR did give the result of Golden Sample, but I am not fully convinced with that and the CPPC tags... CPPC tags in my view can be a reference rather than something to fully rely on when overclocking lol.

 

[freeagent]

Sweet sample youve got there, clocks a little nicer than my 3600XT

 

[Sabotaged_Enigma]

Sweet sample youve got there, clocks a little nicer than my 3600XT

Well, so what's the point of Zen 2 Refresh then...? I thought that 8-core Zen 2 die chips of excellent quality are made into 3800XT and also 3600XT... And now it turns out some XT models still are of bad silicon quality...???

 

[Sabotaged_Enigma]

Thanks a lot! I've used TPU website for a long time, but I'm still a green hand in the TPU forum. Look, I've posted this article in the GPU section lol... Is there a way to change this article into the correct section?

Ah, nice. This article has been moved to where it should've been! Good job, TPU~

 

[Toothless]

That's a nice sample. Clocks just a bit better than mine, however I wouldn't be pushing it past your 1.275v line. Too many samples have degraded or failed past the 1.3v line.

Unless you're wanting to upgrade in two or so years (data is still being collected) I would run er' at 4.5ghz.

 

[Sabotaged_Enigma]

That's a nice sample. Clocks just a bit better than mine, however I wouldn't be pushing it past your 1.275v line. Too many samples have degraded or failed past the 1.3v line.

Unless you're wanting to upgrade in two or so years (data is still being collected) I would run er' at 4.5ghz.

Yeah, I guess I'll be using this processor for at least one year, and I'm waiting for the time to buy a 5900X lol.

 

[QuietBob]

Thanks for sharing your detailed findings! I got my 3300X to 4.5 at the same voltage as you. It isn't Prime-stable at these settings, but otherwise holding up just fine - no restarts, BSODs or WHEAs. Also, mine wasn't stable at 4.6, even with 1.4v.
I would also advise against keeping your CPU at over 1.3v for all core loads. The small gains just aren't worth it. Besides higher temperatures, there's an increased risk of degrading your chip. 4.5 GHz all core seems to be the sensible limit for Zen 2. Seeing how voltage requirements grow at every step, there's a reason why these chips top out at about 4.2 on stock.

 

[tabascosauz]

Thanks! Yeah, CTR did give the result of Golden Sample, but I am not fully convinced with that and the CPPC tags... CPPC tags in my view can be a reference rather than something to fully rely on when overclocking lol.

Well, no per-core OC for Zen 2, so CPPC really doesn't do a whole lot for overclocking haha. Speaking of which, what does the CPPC order look like for your 3700X?

CPPC list is just fused in hardware like Turbo Boost 3.0. The Windows unfortunately doesn't always listen to CPPC and in some applications like CPU-Z always limits ST loads to Core 0. Core 0 on my 3700X was #8 lol and could only manage 4.28GHz, so my ST scores were always low there (520-525). Same for my 4650G, #1 is Core 5.

The high end Zen 3 parts are binned better now. CCD1 is guaranteed to have the better cores, less chance of ending up with a dud Core 0. I finally lucked out with my 5900X, Core 0 is #2 and Core 1 is #1.

 

[Sabotaged_Enigma]

Well, no per-core OC for Zen 2, so CPPC really doesn't do a whole lot for overclocking haha. Speaking of which, what does the CPPC order look like for your 3700X?

CPPC list is just fused in hardware like Turbo Boost 3.0. The Windows unfortunately doesn't always listen to CPPC and in some applications like CPU-Z always limits ST loads to Core 0. Core 0 on my 3700X was #8 lol and could only manage 4.28GHz, so my ST scores were always low there (520-525). Same for my 4650G, #1 is Core 5.

The high end Zen 3 parts are binned better now. CCD1 is guaranteed to have the better cores, less chance of ending up with a dud Core 0. I finally lucked out with my 5900X, Core 0 is #2 and Core 1 is #1.

From Core 1 to 8 my CPPC tags are: 134, 126, 122, 130, 145, 141, 145, 137; CTR Efficiency: 3.92. My CCX2 has better binning than CCX1... HWiNFO indicates that all cores are able to boost to 4.4 GHz which obviously surpasses stock specs. And of course it is PBO, so it comes at the cost of high voltage like 1.5 V lol. I wonder if PBO2 still has the chance to be available to Zen 2.

 

[Sabotaged_Enigma]

Yeah, of course, one sample does not stand for all. But it could give you ideas on how voltage would increase on Zen 2 μarch and 7 nm TSMC process node. You could try to increase or decrease based on this equation and your CPU condition. Thx.

 

[Sabotaged_Enigma]

Thanks for sharing your detailed findings! I got my 3300X to 4.5 at the same voltage as you. It isn't Prime-stable at these settings, but otherwise holding up just fine - no restarts, BSODs or WHEAs. Also, mine wasn't stable at 4.6, even with 1.4v.
I would also advise against keeping your CPU at over 1.3v for all core loads. The small gains just aren't worth it. Besides higher temperatures, there's an increased risk of degrading your chip. 4.5 GHz all core seems to be the sensible limit for Zen 2. Seeing how voltage requirements grow at every step, there's a reason why these chips top out at about 4.2 on stock.

Sorry for replying today...
That's true. Now I've restored everything to default. But seeing PBO could give 1.45 V and up to 1.5 V still worries me... But there's no PBO2 available for Zen 2...