$100 to anyone that can fix my boosting problem

[neurotix]

Hello everyone,

 

I am offering $100 and rep on the site for anyone who can help me fix my boosting problerms with my motherboard and CPU.

 

CPU is a 9950x, previous was a 7900X3D. Motherboard is a ROG Strix X670E-E Gaming Wifi, as well as a Gigabyte Aorus X670 Elite AX. Memory is G.skill TridentZ5 Intel XMP 7200MHz running at 7600MHz cas 36 (is stable in an hour of GSAT, as well as 4 hours of GSAT I did once).

 

Synopsis:

 

Across two different boards, two CPUs, I have this strange issue where my CPU will not boost under any single threaded load and it is massively underperforming with multithread as well.

 

Basically, with PBO set to 90C, 10X scalar, and +200 boost overiride, I will see the chip boosting to 5600-5700MHz in hwinfo64 on the desktop when I am not doing anything. However, putting it under any single thread or lightly threaded load like 3dmark11 or CPU-Z bench, it only does 5450MHz under that load which is the default boost for the 9950x. The 7900X3D also didnt boost on its second CCD, I would also see it hitting 5650 in hwinfo64 on the desktop but under any actual load it would drop to 5150MHz.

 

This problem with PBO is the same if I do it in Asus or Gigabyte's Tweaker menus, or in the AMD overclocking section. It has persisted across multiple motherboards, chips and Win10/Win11 installs. I cant find a way to fix it. Help please. I've finally gotten fed up enough with it that I am desperately asking for help. Thanks anyone in advance who tries.

Edited September 17 by neurotix

[Alex]

I wouldn't worry about 150mhz personally.

 

Have you run everything stock, JEDEC spec RAM, etc. and validated single core boost performance uplift? Likely the +200mhz is not stable enough to sustain boost under load despite passing tests. Modern CPU overclocking is closer to GPU Boost than traditional overclocking... there are many flags to why it won't boost, and we can't control all of them.

 

Would also do a fresh WinOS install on a new/spare SSD to validate. (Not over existing install on existing SSD). Ideally if you can avoid Windows...

[Slaughtahouse]

What type of benchmarks are you running?

 

What is your performance at stock?

 

You should try a single core load, an all core load, and a multi core / blend load (game).

 

According to Anandtech (rip) 7000 X3D and 9000 series have a “Pro-visionary PPM” driver to consolidate loads to one CCD where all cores aren’t needed, like in games, to avoid cross CCD latency. That could explain why you’re not seeing the second CCD boost, depending on how you’re benching. https://www.anandtech.com/show/21524/the-amd-ryzen-9-9950x-and-ryzen-9-9900x-review

 

 

 

The official boost clock is “up to 5.7ghz” and AMD always states it’s for bursty, single core loads. You’re not always guaranteed to sustain max boost clock on one core. I would avoid PBO or curve tuner and bench to understand how your CPU performs at stock first with your system / cooling. If it doesn’t align with reviews, then I would deep dive.

 

Example: TechPowerUp always has a nice graph near the end of their reviews, highlighting how boost clocks drop as core/thread count increases. https://www.techpowerup.com/review/amd-ryzen-9-9950x/26.html

 

 

 

Once you know the in’s and out’s of stock performance, I think you’ll be better prepared to fine tweak PBO. 
 

I’m only posting because I agree with the user above, I don’t believe you’re losing massive performance. You may lose some, but it seems negligible.

[UltraMega]

Just a note, the most recent bios update cuts CCD latency in half for the 9000 series. Seems to have fixed a glitch. Make sure you have that, tho I'm sure it's not related to your boosting issue. 

[neurotix]

16 hours ago, UltraMega said:

Just a note, the most recent bios update cuts CCD latency in half for the 9000 series. Seems to have fixed a glitch. Make sure you have that, tho I'm sure it's not related to your boosting issue. 

 

Last bios for my board. The one before it fixed sinkclose. I am on the latest BIOS but I don't think it has what you're talking about.

 

Everyone else: I've already tried everything recommended. I appreciate the attempt to help though. And seeing that graph helps a lot, this is basically expected behavior I take it?

 

With the 9950x, it's the first CCD. When I first got AM5 around April of last year, with the 7900X3D in Windows it WAS boosting the second CCD to 5550,5650,5500 etc. with power plan set to performance. I was able to benchmark 3dmark benches and get much higher scores even with the small difference in clock speed. Unfortunately, it stopped boosting after a month. I'm an enthusiast benchmarker, https://hwbot.org/user/neurotix so the 9950x doing 5450 versus 5650-5700 is the difference between a 66k and 70k 3dmark11 score and that really matters and affects the points you get on the site. That is the impetus for me being bothered so much over a small difference.

 

[damric]

try scaler at x1.

 

 

[Slaughtahouse]

6 hours ago, neurotix said:

 

Last bios for my board. The one before it fixed sinkclose. I am on the latest BIOS but I don't think it has what you're talking about.

 

Everyone else: I've already tried everything recommended. I appreciate the attempt to help though. And seeing that graph helps a lot, this is basically expected behavior I take it?

 

With the 9950x, it's the first CCD. When I first got AM5 around April of last year, with the 7900X3D in Windows it WAS boosting the second CCD to 5550,5650,5500 etc. with power plan set to performance. I was able to benchmark 3dmark benches and get much higher scores even with the small difference in clock speed. Unfortunately, it stopped boosting after a month. I'm an enthusiast benchmarker, https://hwbot.org/user/neurotix so the 9950x doing 5450 versus 5650-5700 is the difference between a 66k and 70k 3dmark11 score and that really matters and affects the points you get on the site. That is the impetus for me being bothered so much over a small difference.

 

 

RE: Graph, Yes, it’s the expected behaviour at stock.

 

It gives a good approximation of how X cpu should preform under various load conditions.

 

Example 1: Using software that is primarily single threaded? Expect an average around 5.7 GHz on one to two cores.

 

Example 2: Running an all core, 32 threaded render? You’re going to be near or below 5.3 GHz.

 

Example 3: Playing a game that stresses multiple cores? You’re likely going to see bouncing 5.4 to 5.5 GHz on some cores. Since the vast majority of games don’t effectively stress a full 16 core chip.

 

In other words, it’s a baseline. I highly recommend you develop your own at stock. That way, as you adjust levers (variables like Scalar) in PBO, you can see the impact to your clock speeds under different benchmarks. If you don’t want to develop your own, you can reference theirs but it won’t be 100%, 1:1.

 

I’m not a good overclocker. I usually don’t push chips nearly as far as others do on enthusiast forums or probably as far as you have. I’ve just learned through trial and error, especially on Zen, it’s best to understand how your own chip behaves at stock before tweaking.

 

RE: CCD not boosting, does this occur outside of 3DMark, like an all core CB R23 bench? If so, something appears fundamentally wrong. If all cores are active and boosting around 5.2 - 5.3, that’s seems normal. At least against TechPowerUps data.

[Alex]

Your 7900X3D boosted booth CCDs the same at first? Interesting... my 7950X3D has never boosted the first CCD above 5250mhz (X3D CCD).

 

This is a 7 day hwinfo average:

             name                                           |    current   |      min       |       max     |     avg

Edited September 19 by Alex

[neurotix]

On 19/09/2024 at 07:45, Alex said:

Your 7900X3D boosted booth CCDs the same at first? Interesting... my 7950X3D has never boosted the first CCD above 5250mhz (X3D CCD).

 

This is a 7 day hwinfo average:

             name                                           |    current   |      min       |       max     |     avg

 

My 7900X3D never boosted the first CCD past 5150MHz, which is it's max limit. In Balanced power plan (which you need to use for V-Cache to work) it would hit and more or less maintain that 5150 in games.

 

The second CCD at first would hit 5650, 5550, 5500 etc in high performance power plan in benchmarks and games, but stopped doing so after month and would only hit 5200MHz max, it seemed like the limit for the first CCD was being applied to the second.

 

Also, I'm not leaving you out Slaughtahouse. Thank you for your advice. I will take a look at it more thoroughly sometime soon.