It checks out as authentic

[oxrufiioxo]

Good to hear.

 

[Darkholm]

Tested. It's good and authentic.

You had us in the first half, I'm not gonna lie
Good to hear.

 

[Zach_01]

Now time to update system spec status?

 

[Courier 6]

good to know

 

[Zach_01]

I think AMD should keep an eye on this kind of crap

What do you mean exactly?
Like AMD should police every market for fake CPUs?

 

[A Computer Guy]

Now time to update system spec status?

The 5950x King isn't quite dethroned just yet. I have some testing to do (see list below) to figure out how I'm going to rearrange things and I'd like to mess around a bit with PBO and CO and memory tweaking to see if there are any settings I'd like to apply or just leave it at stock. I'm intending this to become my primary work PC and replace one of my 5950x Taichi systems that will in turn retire my 3950x home server that will remain as an emergency board in spare in storage. Since I can't find an AM5 board at the right price that will support x8/x8/x4/x1 PCIe slot configuration I'm gong to have to split up machine duties between the new 7950x Live Mixer and the remaining 5950x Taichi.

New System Specs so far...

• AMD Ryzen 9 7950x
• ASRock B650 Live Mixer
• 64GB Crucial (2 x 32GB) DDR5-5600 ECC UDIMM 2Rx8 CL46 (MTC20C2085S1EC56BR)
• Noctua NH-D9L

Quick Benches

Currently running at stock but RAM per UEFI defaults ram at DDR5-5200
MCLK = 2600
FCLK = 1733
UCLK = 2600

• CPU-Z
  • Underperforming a bit at multi-thread 15,640 vs. bench reference 15,728
  • Parity in single thread 766 vs. bench reference 767
• Cinebench R23
  • Underperforming in multi-thread 35,455 vs. TPU reference 37,523
  • Underperforming in single thread 1,979 vs. TPU reference 2051

To Do

• Learn how to tune Zen4 and DDR5
• ECC appears to be recognized by OS and HWiNFO64. Try to get OS to log error corrections with overclock.
• Need to test USBC with current USBC switch as primary (powers switch) or secondary
• Determine if the 7950x iGPU adequately supports my VM's. If so I need an HDMI to DP cable for dual monitor setup.
  • Previously on my existing systems I tried an Nvidia Quadro K600 but it was terrible and allowed some kind of system interrupts to disrupt or slow down the VM periodically to the point it couldn't process audio/video briefly and/or was distorted periodically. The RX-5500XT, RX-5700, RX-6600XT, and RX-6700XT had no issues with VM's and they were smooth as butter. I have an RX-560-B and Nvidia GTX 1060 3GB but never tried them in this role.
• Try Linux in this role to host VM's. If suitable with iGPU or GPU then start farewell to Windows tour using this new PC.
• If iGPU isn't adequate then install GPU into 2nd x16 slot (is x4 from CPU)
• Transfer Dual Optanes to primary x16 slot (bifurcation support confirmed x4x4x4x4 config supported)
• Not sure if my existing AM4 water blocks are compatible (Velocity or Velocity2) - Might switch to Heatkiller WB for AM5 if I put system into my 011D.

( edit )

Fixing UEFI to read RAM config and adjusting to matching DDR5 module had a bit of multi-thread improvement.

MCLK = 2800
FCLK = 2000
UCLK = 2800

• Cinebench R23
  • in multi-thread 37,107 vs. TPU reference 37,523
  • in single thread 1,975 vs. TPU reference 2051

More pictures...

 

[tabascosauz]

The 5950x King isn't quite dethroned just yet. I have some testing to do (see list below) to figure out how I'm going to rearrange things and I'd like to mess around a bit with PBO and CO and memory tweaking to see if there are any settings I'd like to apply or just leave it at stock. I'm intending this to become my primary work PC and replace one of my 5950x Taichi systems that will in turn retire my 3950x home server that will remain as an emergency board in spare in storage. Since I can't find an AM5 board at the right price that will support x8/x8/x4/x1 PCIe slot configuration I'm gong to have to split up machine duties between the new 7950x Live Mixer and the remaining 5950x Taichi.

New System Specs so far...

• AMD Ryzen 9 7950x
• ASRock B650 Live Mixer
• 64GB Crucial (2 x 32GB) DDR5-5600 ECC UDIMM 2Rx8 CL46 (MTC20C2085S1EC56BR)
• Noctua NH-D9L

Quick Benches

Currently running at stock but RAM per UEFI defaults ram at DDR5-5200
MCLK = 2600
FCLK = 1733
UCLK = 2600

• CPU-Z
  • Underperforming a bit at multi-thread 15,640 vs. bench reference 15,728
  • Parity in single thread 766 vs. bench reference 767
• Cinebench R23
  • Underperforming in multi-thread 35,455 vs. TPU reference 37,523
  • Underperforming in single thread 1,979 vs. TPU reference 2051

To Do

• Learn how to tune Zen4 and DDR5
• ECC appears to be recognized by OS and HWiNFO64. Try to get OS to log error corrections with overclock.
• Need to test USBC with current USBC switch as primary (powers switch) or secondary
• Determine if the 7950x iGPU adequately supports my VM's. If so I need an HDMI to DP cable for dual monitor setup.
  • Previously on my existing systems I tried an Nvidia Quadro K600 but it was terrible and allowed some kind of system interrupts to disrupt or slow down the VM periodically to the point it couldn't process audio/video briefly and/or was distorted periodically. The RX-5500XT, RX-5700, RX-6600XT, and RX-6700XT had no issues with VM's and they were smooth as butter. I have an RX-560-B and Nvidia GTX 1060 3GB but never tried them in this role.
• Try Linux in this role to host VM's. If suitable with iGPU or GPU then start farewell to Windows tour using this new PC.
• If iGPU isn't adequate then install GPU into 2nd x16 slot (is x4 from CPU)
• Transfer Dual Optanes to primary x16 slot (bifurcation support confirmed x4x4x4x4 config supported)
• Not sure if my existing AM4 water blocks are compatible (Velocity or Velocity2) - Might switch to Heatkiller WB for AM5 if I put system into my 011D.

D9L for a 7950X? Good god

Scores don't seem too shabby at all considering the cooling. And ST is never truly comparable across samples for reasons I'm sure you already know.

EK was sending out compatibility kits for older AM4 blocks to upgrade to AM5 due to the screws not being compatible, but given where EK is now.............

 

[A Computer Guy]

D9L for a 7950X? Good god

Scores don't seem too shabby at all considering the cooling. And ST is never truly comparable across samples for reasons I'm sure you already know.

EK was sending out compatibility kits for older AM4 blocks to upgrade to AM5 due to the screws not being compatible, but given where EK is now.............

It was cheaper than any of the D15 variants. I used Thermal Grizzly (pink stuff) including pasting the fingers of the CPU to maximize thermal contact and used the offset mounting kit that was included.

 

[AusWolf]

What do you mean exactly?
Like AMD should police every market for fake CPUs?

I think what was meant is that AMD should announce these small updates to avoid situations like OP's dilemma.

It was cheaper than any of the D15 variants. I used Thermal Grizzly (pink stuff) including pasting the fingers of the CPU to maximize thermal contact and used the offset mounting kit that was included.

It's cheaper for a reason. You're probably seeing a slightly reduced MT score because you're hitting a thermal wall.

 

[nomdeplume]

Was holding out for plot twist:

CPU is troll knockoff that outperforms on all hardware fronts and has better firmware implementation.

 

[tabascosauz]

It was cheaper than any of the D15 variants. I used Thermal Grizzly (pink stuff) including pasting the fingers of the CPU to maximize thermal contact and used the offset mounting kit that was included.

I sure would hope it's much cheaper haha, D9L and D15 are very far apart in performance

Raphael's iGPU has some unique functionality in that it works as hybrid graphics; even if one of the monitors is connected to the iGPU and not dGPU, if you move your game window over to that monitor you still get the dGPU running the game. I'm not sure what happens outside of 3d loads though. While still getting the idle power benefits, if using iGPU on your particular setup happens to meet a specific VRAM clock breakpoint.

I stopped using the iGPU because it's either higher UCLK or FCLK that wreaks unavoidable havoc on iGPU output and causes it to flicker from time to time. But it's handy to have for sure.

 

[AusWolf]

Raphael's iGPU has some unique functionality in that it works as hybrid graphics; even if one of the monitors is connected to the iGPU and not dGPU, if you move your game window over to that monitor you still get the dGPU running the game. I'm not sure what happens outside of 3d loads though. While still getting the idle power benefits, if using iGPU on your particular setup happens to meet a specific VRAM clock breakpoint.

Main monitor stuff still works on the dGPU. That's why I had to disable hardware acceleration in Chrome if I wanted my 7800 XT's fans to stay off (they still turned on from time to time, though) while I still had it.

 

[Courier 6]

What do you mean exactly?
Like AMD should police every market for fake CPUs?

no, nevermind

 

[A Computer Guy]

I sure would hope it's much cheaper haha, D9L and D15 are very far apart in performance

Raphael's iGPU has some unique functionality in that it works as hybrid graphics; even if one of the monitors is connected to the iGPU and not dGPU, if you move your game window over to that monitor you still get the dGPU running the game. I'm not sure what happens outside of 3d loads though. While still getting the idle power benefits, if using iGPU on your particular setup happens to meet a specific VRAM clock breakpoint.

I noticed that same behavior with the 4750G which used Vega 8 with Windows and Linux.

I stopped using the iGPU because it's either higher UCLK or FCLK that wreaks unavoidable havoc on iGPU output and causes it to flicker from time to time. But it's handy to have for sure.

When I was testing the 4750G over the HDMI Cinebench would crash the driver output briefly until the test run completed. Perhaps it's related to same reasons you indicated. DP was 100% stable but HDMI was giving me issues like flickering and display output loss with workloads like Cinebench. I was hoping 7950x RDNA2 iGPU would behave a bit better. I haven't tried the HDMI output yet.

I sure would hope it's much cheaper haha, D9L and D15 are very far apart in performance

Indeed it was cheaper and I needed to get 2 so I can also replace my dying original H100i that I mostly use for testing but has been pulling full time duty since the only other thing I had on hand was a few Wraith Stealth and a NH-L9a cooler. I wanted to try NH-D9L before I possibly get invested in NH-D15 (original) which is a bit more reasonable in price compared to the new NH-D15 series. My experience with NH-C14S & 5950x was good despite also being somewhat lackluster in performance specs. I'm also open to considering alternatives although I'm concerned about weight so NH-D9L addresses that concern by also being significantly more compact.

 

[tabascosauz]

I noticed that same behavior with the 4750G which used Vega 8 with Windows and Linux.

When I was testing the 4750G over the HDMI Cinebench would crash the driver output briefly until the test run completed. Perhaps it's related to same reasons you indicated. DP was 100% stable but HDMI was giving me issues like flickering and display output loss with workloads like Cinebench. I was hoping 7950x RDNA2 iGPU would behave a bit better. I haven't tried the HDMI output yet.


Indeed it was cheaper and I needed to get 2 so I can also replace my dying original H100i that I mostly use for testing but has been pulling full time duty since the only other thing I had on hand was a few Wraith Stealth and a NH-L9a cooler. I wanted to try NH-D9L before I possibly get invested in NH-D15 (original) which is a bit more reasonable in price compared to the new NH-D15 series. My experience with NH-C14S & 5950x was good despite also being somewhat lackluster in performance specs. I'm also open to considering alternatives although I'm concerned about weight so NH-D9L addresses that concern by also being significantly more compact.
View attachment 363826

The NSPR numbers are more or less on the mark. D9L is a downgrade from U9S, designed mainly for server rack/SFF 115mm height compatibility. Performance at that point gets awfully close to larger low profile coolers like L12S.

I didn't have that behaviour with Vega 7/8, but the 7000 iGPU behaviour at higher UCLK/FCLK is pretty well documented. Fortunately at 6000, I wouldn't say that it's very intrusive, for a second monitor. Maybe DP would have better luck, maybe not.

 

[A Computer Guy]

Just some more benchmarks at stock (with RAM adjusted for DDR5-5600 and a CPU temp limit of 85c) for those who are interested.
Despite the small cooler and slower RAM it scores average in SSE and AVX.

Memory reads are average with writes just a bit ahead of average but nowhere near the top performer in either memory reads or writes.

 

[dirtyferret]

Amazon shipped by Amazon is the surest warranty there is.

even if it's shipped by the merchant you can still start an A to Z claim if the merchant sucks. As much as I don't care for Amazon (work with them every day) they always side with the customer.

 

[tabascosauz]

Just some more benchmarks at stock (with RAM adjusted for DDR5-5600 and a CPU temp limit of 85c) for those who are interested.
Despite the small cooler and slower RAM it scores average in SSE and AVX.

No idea what those benchmark scores are, but that's the beauty of a 16-core, you can limit it as much as you want and it will still handily outperform a 12-core and I wouldn't expect D9L to have trouble with ST

As it dynamically throttles all the time to the temp limit by design, I think you might have to take a look at wattage and clocks in HWInfo for better context. Seems to be a big spread in stock performance for a lot of 7000 hardware.

Being an ECC kit and Micron, there's probably not an incredible amount of headroom on either freq or timings, can probably just leave it where it is

 

[A Computer Guy]

Getting a rough idea of how my VMWare Windows Guest performance will improve.

Being an ECC kit and Micron, there's probably not an incredible amount of headroom on either freq or timings, can probably just leave it where it is

Without any changes in voltage I was able to bump up the speed a notch for some minor improvements. I'll probably just leave it at DDR5-6000 unless someone can spot some easy things to try for a more optimal memory configuration.

Performance Test 10.2 (Ya I know it's old and might not be a very good benchmark tool. Just using it to see comparative differences.)

 

[Sarajiel]

Without any changes in voltage I was able to bump up the speed a notch for some minor improvements. I'll probably just leave it at DDR5-6000 unless someone can spot some easy things to try for a more optimal memory configuration.

As mentioned in another thread, I’m not familiar with DDR5 Micron ICs, but you could always try to tighten tRFC a bit and increase the tREFI a lot, depending on the temperature of your system. However, I'm not sure if ECC error reporting will still accurately work then.

Personally, I've never overclocked any ECC memory, mainly because in the past it was only supposed to work at the exact speed the sticks were specified for. Maybe I'm just getting old, or it's different with DDR5, since Kingston happily sells overclockable reg. ECC kits these days, but I would definitely test not just for stability, but also if ECC still works after tightening timings or changing clocks.

CPU temp limit of 85c

If you are going for improved energy efficiency, you can probably reduce the PPT of the 7950X down into the 160W to 200W range without loosing much performance, or even go full eco-mode, aka the 142W PPT preset like the mid-range Ryzen 7000s. I locked both my 7950X's to 85°C and 200W without noticing much of a performance loss in real world applications, but they run a decent bit quieter without the fans ramping up for those annoying load spikes above 215W. I'm using AIOs with a quiet fan curve for them, though.

 

[A Computer Guy]

As mentioned in another thread, I’m not familiar with DDR5 Micron ICs, but you could always try to tighten tRFC a bit and increase the tREFI a lot, depending on the temperature of your system. However, I'm not sure if ECC error reporting will still accurately work then.

Personally, I've never overclocked any ECC memory, mainly because in the past it was only supposed to work at the exact speed the sticks were specified for. Maybe I'm just getting old, or it's different with DDR5, since Kingston happily sells overclockable reg. ECC kits these days, but I would definitely test not just for stability, but also if ECC still works after tightening timings or changing clocks.

The last ones I really overclocked for daily use was a set of Crucial DDR4-2667 DIMMS that ran happily at DDR4-3200 with a bit of a voltage bump. The real 3200 was a bit more expensive at the time so I opted for cheaper 2667 and did some experimentation.

If you are going for improved energy efficiency, you can probably reduce the PPT of the 7950X down into the 160W to 200W range without loosing much performance, or even go full eco-mode, aka the 142W PPT preset like the mid-range Ryzen 7000s. I locked both my 7950X's to 85°C and 200W without noticing much of a performance loss in real world applications, but they run a decent bit quieter without the fans ramping up for those annoying load spikes above 215W. I'm using AIOs with a quiet fan curve for them, though.

Right now I have that NH-D9L cooler so I might restrict the power level a bit. Running stock I just finished a Linux test zipping one of my VM's for 45 mins at 5GHz peaking 96c (230-ish watts) while listening to some random youtube 80's classic mix and browsing for how to get temp status. Just as just YT starting playing a-ha "Take On Me" I stopped the zipping because it was taking too long and the dumb compress dialog can't or won't show progress, then shortly my wifi card failed on the cooldown cycle. Just then I was feeling nostalgic for Windows, 7zip dialog, and HWiNFO64 because it all just works. The switch is painful.

I'm starting another test. It seems 7zip is only pulling 200-ish watts (total system power) @80c. Adding YT Foreigner "I want to know what love is" pushes it to 230w 96c but at least I can see my zip progress now. RAM is pushing 54c. I think my VRM temp is around 60c but no specific sensor named VRM so I can't be sure.
( edit )
Test 80% complete

It seems I'm right in the ballpark for frequency and temperatures with this little air cooler.

With the default voltages set from my RAM after quite a bit of testing my infinity fabric speed tops out at 2166 (2200 no post) and RAM at 6200MT/s (6400MT/s will reboot after a short time).

Keeping MCLK=UCLK and FCLK 2:3 ratio I land at the following configuration without changing any timings.
2800-2000-2800 => 3100-2066-3100

OCCT's benchmark shows a bit of improvement

Reads
1,538.65 => 1,655.71
Writes
1,473.81 => 1,588.77

Comparing Thaiphoon Burner output it seems UEFI/BIOS memory training might be tightening up timings a bit?

I took out my old cheat sheet to see what the differences in timings look like in nanoseconds.

I reworked my spreadsheet a bit and I was able to squeeze in a nearly universal improvement in timings just by raising frequency and after making some manual corrections where UEFI was on AUTO. I should be able to use this as a guide to see how much I might be able to trim the timings at different speeds under 6200 at the current voltages I'm using and then take a look at the overall latency and choose which one has the best balance in latency and bandwidth. Well that's the theory anyway.