[xeome]

Can anyone run the "Latency / Bandwidth benchmark" on OCCT?
I'm questioning my speeds, can only find 1 screenshot which mentioned about 400-500+GiB/s in L3 Cache read on a 7800x3d as well.

Meanwhile mine can't even break 200 GiB/s so i'm questioning if something is wrong.
AIDA64 does provide me about +- 680 on the benchmark but latency is quite high there.

 

[f00zz]

looks about right, i had browsers open playing music etc but it's almost the exact same numbers so i'd say yours is good. Mine's not overclocked but RAM set @ 6000

 

[bigcryp]

Was able to get 6400CL30 stable with nitro (1 - 2 - 0 8x 8x).
Next up is upping VDD and seeing what I can do to the primary timings. (Let me know if you see anything obvious )

 

[bigcryp]

Was able to get 6400CL30 stable with nitro (1 - 2 - 0 8x 8x).
Next up is upping VDD and seeing what I can do to the primary timings. (Let me know if you see anything obvious )

View attachment 2720660 View attachment 2720661

Nitro 1 - 2 - 0 instead of 1 - 2 - 1 netted a good reduction in latency but has made it a lot more difficult to lower timings.

 

[Zero989]

Was able to get 6400CL30 stable with nitro (1 - 2 - 0 8x 8x).
Next up is upping VDD and seeing what I can do to the primary timings. (Let me know if you see anything obvious )

View attachment 2720660 View attachment 2720661

Isn't auto 8x and 8x anyway as you enabled Nitro

 

[bigcryp]

Isn't auto 8x and 8x anyway as you enabled Nitro

Probably, but in my MSI bios it shows 'Auto' and I'd rather set it to something of which I know the value then on 'Auto'. Just as a precaution.

 

[Zero989]

Probably, but in my MSI bios it shows 'Auto' and I'd rather set it to something of which I know the value then on 'Auto'. Just as a precaution.

If 2 and 4 dont work, then it can only be 8 unless Auto is 16

 

[bigcryp]

If 2 and 4 dont work, then it can only be 8 unless Auto is 16

Not sure, but results improved when I set nitro to on. Now that you said this I might try and see the difference between the different burst settings.

 

[annikalang92]

can someone give me the TLDR for an easy and stable overclock?

PBO. Tjmax = 85°C and Curve Optimizer – 20mV?


also one question:

Profile	Avg Clock	Eff. Clock	Max Temp	Avg Temp	Score
Cinebench Defaults	4682 MHz	4624 MHz	82.9 °C	80.3 °C	1053
Cinebench –20 CO +200 MHz	4815 MHz	4705 MHz	82.8 °C	80.3 °C	1097
Cinebench –20 CO	4825 MHz	4805 MHz	82.8 °C	81.3 °C	1098

Why is the score and effective clock with a –20 all-core Curve Optimizer the same as –20 all-core Curve Optimizer plus a +200 MHz max CPU boost-clock override? Should the latter score higher?

 

[Dekatris]

can someone give me the TLDR for an easy and stable overclock?

PBO. Tjmax = 85°C and Curve Optimizer – 20mV?


also one question:

Profile	Avg Clock	Eff. Clock	Max Temp	Avg Temp	Score
Cinebench Defaults	4682 MHz	4624 MHz	82.9 °C	80.3 °C	1053
Cinebench –20 CO +200 MHz	4815 MHz	4705 MHz	82.8 °C	80.3 °C	1097
Cinebench –20 CO	4825 MHz	4805 MHz	82.8 °C	81.3 °C	1098

Why is the score and effective clock with a –20 all-core Curve Optimizer the same as –20 all-core Curve Optimizer plus a +200 MHz max CPU boost-clock override? Should the latter score higher?

There isn't really a TL;DR because every CPU is different and you have to find the limits for the one you received.

The "easiest and most stable" overclock is just enabling PBO and letting the CPU find it's own limits with the default voltage/frequency curves. The next step is finding how low you can set the all-core CO while remaining stable in something like CoreCycler with one its Ryzen configs that use y-cruncher. The step after that is finding per-core values, which CoreCycler can kind of do automatically for 7000 series - you have to set them pretty low to start, then let it work its way up until its stable. If you're looking for easy, I wouldn't go much further than that.

As for your score with +200 vs without, that +200 just moves the boost frequency limit, which you aren't hitting, so it is irrelevant.

 

[annikalang92]

There isn't really a TL;DR because every CPU is different and you have to find the limits for the one you received.

The "easiest and most stable" overclock is just enabling PBO and letting the CPU find it's own limits with the default voltage/frequency curves. The next step is finding how low you can set the all-core CO while remaining stable in something like CoreCycler with one its Ryzen configs that use y-cruncher. The step after that is finding per-core values, which CoreCycler can kind of do automatically for 7000 series - you have to set them pretty low to start, then let it work its way up until its stable. If you're looking for easy, I wouldn't go much further than that.

As for your score with +200 vs without, that +200 just moves the boost frequency limit, which you aren't hitting, so it is irrelevant.

youre the GOAT thank you!

 

[Kimir]

Can confirm, CoreCycler with y-cruncher is pretty good for that:

 

[Dekatris]

Can confirm, CoreCycler with y-cruncher is pretty good for that:
View attachment 2722386

Jesus... -18 as your lowest CO on a perf 1 core on CCD1... My 7950X3D can only do -1 on one of them, which means it's basically running stock, and that's after RMAing the first one I got because it couldn't even pass AIDA SHA3 stock.

 

[Kimir]

Jesus... -18 as your lowest CO on a perf 1 core on CCD1... My 7950X3D can only do -1 on one of them, which means it's basically running stock, and that's after RMAing the first one I got because it couldn't even pass AIDA SHA3 stock.

My 7950X3D was like your (and I had two RMAed before that, 1st one died, 2nd wasn't passing occt stock), I had two cores I couldn't even do -1 or barely. That 9950X3D is definitively better in that way.

 

[Blackowl]

Dumping My Bios Settings on
CPU : Ryzen 7950X3D
GPU : RX 6800XT
RAM : M-DIE (V1) G.Skill 6000mhz 30CL
MB : ASUS ProArt X670E-CREATOR
Here for reference in case anyone looking for it.

These settings aiming toward all aspect of the machine, Performance, Power Consumption, Heat and also Security.

If you are after Maximum Performance and Lowest possible Latency this is NOT for you, Some setting might have performance as cost. The same with anyone what want to run their system at maximum efficient and lowest power draw, this setting is no where close to Eco mode and enhance all power saving features

Yes, This latency can achieve with SVM Enable.

Dickey Lee like this result (old song reference)

 

[MarkPost]

I just opened a new thread to compare the latest CPU gens perfomance, hope it will be a good place to have a reliable database about it.

Also, it will be a good way to maintain the specific CPU threads as cleaner as possible.

Here is the link Perfomance comparison between the latest CPU gens | Overclock.net

 

[MamdOoO]

Hello, new x3d owner her :


Has anyone else seen this issue with a new build?

System:

• Ryzen 7 7800X3D
• Asus TUF Gaming B650-E WiFi
• Radeon 7900XT
• Corsair Vengeance 6000 MHz RAM

The problem:
When I leave the system idle on the desktop for a longer time, Adrenalin eventually shows very low clocks (around 0.5–0.8 GHz), and at some point the CPU just sticks to a single low clock speed.

HWInfo and Windows then show that only 3 cores are actually active, with the clocks stuck at around 3.8–4.2 GHz according to Task Manager. In Cinebench R23 I can’t reach full CPU utilization anymore (max ~86%), the score drops significantly, and not all cores ramp up.

A reboot fixes the issue immediately.

When everything works normally, I see a stable 5 GHz boost in Cinebench, and even after an hour of load temps stay below 75°C.

It almost feels like the CPU goes into some kind of deep sleep state after being idle too long and then never fully wakes up.

Power plan is set to Balanced.

Has anyone experienced this before or knows what could cause it?

 

[Blackowl]

Hello, new x3d owner her :


Has anyone else seen this issue with a new build?

System:

• Ryzen 7 7800X3D
• Asus TUF Gaming B650-E WiFi
• Radeon 7900XT
• Corsair Vengeance 6000 MHz RAM

The problem:
When I leave the system idle on the desktop for a longer time, Adrenalin eventually shows very low clocks (around 0.5–0.8 GHz), and at some point the CPU just sticks to a single low clock speed.

HWInfo and Windows then show that only 3 cores are actually active, with the clocks stuck at around 3.8–4.2 GHz according to Task Manager. In Cinebench R23 I can’t reach full CPU utilization anymore (max ~86%), the score drops significantly, and not all cores ramp up.

A reboot fixes the issue immediately.

When everything works normally, I see a stable 5 GHz boost in Cinebench, and even after an hour of load temps stay below 75°C.

It almost feels like the CPU goes into some kind of deep sleep state after being idle too long and then never fully wakes up.

Power plan is set to Balanced.

Has anyone experienced this before or knows what could cause it?

Use Hwinfo and observe the C state.

if it's C-state change the APCI setting I show above.