[Agent-A01]

Have to disagree here, I found running 8000Mhz vs 6400Mhz in games with ray tracing always gave more FPS and games with zero ray tracing was a wash, but on my system 8000Mhz always was best, hence why I only use 8000Mhz for gaming now.

You'll have to provide some data to prove that.
How would there be any tangible benefit when in single CCD CPU the FCLK link is the limiting factor. FCLK bandwidth is less than memory bandwidth.

With dual CCD CPUs there's 2x FLCK links so memory bandwidth can be beneficial in certain circumstances but majority of games are going to be pushed to a single CCD anyways.

The best overall performance possible for gaming is 2200+ FCLK + DDR5 6600. Chips that won the IMC lottery can do 6800.

Now if you've compared tight 8000 vs a plain EXPO 6400 setup then that could explain why it's always "best". But plenty of personal testing have been posted here as well as reviewer's posting where 1:2 mode has an overwhelming majority of tests being worse than 1:1 mode.

I'm sure there may be an outlier or two out there but for 99% of games it makes no sense to sacrifice latency for more bandwidth on a single CCD CPU.

 

[BlueShark]

You'll have to provide some data to prove that.
How would there be any tangible benefit when in single CCD CPU the FCLK link is the limiting factor. FCLK bandwidth is less than memory bandwidth.

With dual CCD CPUs there's 2x FLCK links so memory bandwidth can be beneficial in certain circumstances but majority of games are going to be pushed to a single CCD anyways.

The best overall performance possible for gaming is 2200+ FCLK + DDR5 6600. Chips that won the IMC lottery can do 6800.

Now if you've compared tight 8000 vs a plain EXPO 6400 setup then that could explain why it's always "best". But plenty of personal testing have been posted here as well as reviewer's posting where 1:2 mode has an overwhelming majority of tests being worse than 1:1 mode.

I'm sure there may be an outlier or two out there but for 99% of games it makes no sense to sacrifice latency for more bandwidth on a single CCD CPU.

My testing was a 7950X3D dual CCD CPU, if you search my posts I did some benchmarking and 8000Mhz came on top or equal to 6400Mhz, now 6600Mhz requires a lot of voltages to be daily stable, just not worth it.

In games like cyberpunk and spiderman basically ray tracing titles always use the frequency cores if only one core like cyberpunk and gives a nice up lift.

Reviewers let’s be honest do not tighten their timings and just use EXPO at 8000, I have used tight 6400 and 8000 timings and again like I said they are equal or 8000 wins.

When I not busy as I have a lot of new games including Avatar which as a lot of ray tracing I will compare, and I will upload them here.

 

[J7SC]

My testing was a 7950X3D dual CCD CPU, if you search my posts I did some benchmarking and 8000Mhz came on top or equal to 6400Mhz, now 6600Mhz requires a lot of voltages to be daily stable, just not worth it.

In games like cyberpunk and spiderman basically ray tracing titles always use the frequency cores if only one core like cyberpunk and gives a nice up lift.

Reviewers let’s be honest do not tighten their timings and just use EXPO at 8000, I have used tight 6400 and 8000 timings and again like I said they are equal or 8000 wins.

When I not busy as I have a lot of new games including Avatar which as a lot of ray tracing I will compare, and I will upload them here.

...yeah, this came up before a while back, either here or in the AMD DDR5 OC thread. I also found some gains in Cyberpunk 2077 w/all ray tracing features on/maxed, and in MSFS 2020 (both at 4K, w/4090) for 2:1 when comparing CL30 6400 vs CL36 8000 on my 7950X3D. It also made a difference if r_BAR was active / forced. That said, from what I recall, the difference re. RAM setting wasn't earth-shattering in those two apps.

On a related note, I have forced CCD0 only vs. 'unrestricted' via Task Manager, and both CP 2077 and MSFS 2020 showed different core usage patterns via HWInfo running in the background. Then there is Windows 11 Pro and its 3000+ Threads and 100,000 Handles per task manager. So background / 'other' tasks play a role even apart from game usage. May be this would help...

Spoiler

 

[BlueShark]

...yeah, this came up before a while back, either here or in the AMD DDR5 OC thread. I also found some gains in Cyberpunk 2077 w/all ray tracing features on/maxed, and in MSFS 2020 (both at 4K, w/4090) for 2:1 when comparing CL30 6400 vs CL36 8000 on my 7950X3D. It also made a difference if r_BAR was active / forced. That said, from what I recall, the difference re. RAM setting wasn't earth-shattering in those two apps.

On a related note, I have forced CCD0 only vs. 'unrestricted' via Task Manager, and both CP 2077 and MSFS 2020 showed different core usage patterns via HWInfo running in the background. Then there is Windows 11 Pro and its 3000+ Threads and 100,000 Handles per task manager. So background / 'other' tasks play a role even apart from game usage. May be this would help...

Spoiler

I remember your testing and like you say it was not earth shattering, a few frames here and there, but it was never worser, but testing on a single CCD CPU to honest my testing as been limited, but 2 CCD CPU’s if you can do 8000Mhz then I would do that for gaming as there nothing to lose.

 

[GRABibus]

My testing was a 7950X3D dual CCD CPU, if you search my posts I did some benchmarking and 8000Mhz came on top or equal to 6400Mhz, now 6600Mhz requires a lot of voltages to be daily stable, just not worth it.

In games like cyberpunk and spiderman basically ray tracing titles always use the frequency cores if only one core like cyberpunk and gives a nice up lift.

Reviewers let’s be honest do not tighten their timings and just use EXPO at 8000, I have used tight 6400 and 8000 timings and again like I said they are equal or 8000 wins.

When I not busy as I have a lot of new games including Avatar which as a lot of ray tracing I will compare, and I will upload them here.

How can the frequency CCD be used in a game (you mention here with Ray tracing).
I thought as long as the game is seen as game by Game bar, only CCD0 is used.

 

[Agent-A01]

My testing was a 7950X3D dual CCD CPU, if you search my posts I did some benchmarking and 8000Mhz came on top or equal to 6400Mhz, now 6600Mhz requires a lot of voltages to be daily stable, just not worth it.

In games like cyberpunk and spiderman basically ray tracing titles always use the frequency cores if only one core like cyberpunk and gives a nice up lift.

Reviewers let’s be honest do not tighten their timings and just use EXPO at 8000, I have used tight 6400 and 8000 timings and again like I said they are equal or 8000 wins.

When I not busy as I have a lot of new games including Avatar which as a lot of ray tracing I will compare, and I will upload them here.

I don't mean general reviewers, there are some that understand the FLCK speed, UCLK ratio etc etc.

As for cyberpunk, if your setup is using frequency cores then no wonder. 7800X3D would be faster. Cache CCD > Frequency CCD in that case.

7950x by default has pretty bad scheduling and I would suggest manual intervention.

 

[LtMatt]

I don't mean general reviewers, there are some that understand the FLCK speed, UCLK ratio etc etc.

As for cyberpunk, if your setup is using frequency cores then no wonder. 7800X3D would be faster. Cache CCD > Frequency CCD in that case.

7950x by default has pretty bad scheduling and I would suggest manual intervention.

View attachment 2654315

What software is that?

...yeah, this came up before a while back, either here or in the AMD DDR5 OC thread. I also found some gains in Cyberpunk 2077 w/all ray tracing features on/maxed, and in MSFS 2020 (both at 4K, w/4090) for 2:1 when comparing CL30 6400 vs CL36 8000 on my 7950X3D. It also made a difference if r_BAR was active / forced. That said, from what I recall, the difference re. RAM setting wasn't earth-shattering in those two apps.

On a related note, I have forced CCD0 only vs. 'unrestricted' via Task Manager, and both CP 2077 and MSFS 2020 showed different core usage patterns via HWInfo running in the background. Then there is Windows 11 Pro and its 3000+ Threads and 100,000 Handles per task manager. So background / 'other' tasks play a role even apart from game usage. May be this would help...

Spoiler

How are you forcing rebar exactly?

 

[GRABibus]

How are you forcing rebar exactly?

With nvidia inspector.
I do it for all my games.

 

[LtMatt]

With nvidia inspector.
I do it for all my games.

Can you share a screenshot? Because when I check inspector it seems like its already on for Cyberpunk for example.

 

[GRABibus]

Can you share a screenshot? Because when I check inspector it seems like its already on for Cyberpunk for example.

I am not at home.

if rebar is already enabled in inspector for your game, then it is supported natively.

‘if, not, just enable it.

 

[LtMatt]

I am not at home.

if rebar is already enabled in inspector for your game, then it is supported natively.

‘if, not, just enable it.

So it's not really forced then? Just want to make sure I'm not missing out of anything in terms of extra performance.

 

[GRABibus]

So it's not really forced then? Just want to make sure I'm not missing out of anything in terms of extra performance.

Yes, theorically you can disable a native rebar supported game via inspector

 

[BlueShark]

don't mean general reviewers, there are some that understand the FLCK speed, UCLK ratio etc etc.

As for cyberpunk, if your setup is using frequency cores then no wonder. 7800X3D would be faster. Cache CCD > Frequency CCD in that case.

7950x by default has pretty bad scheduling and I would suggest manual intervention.

I think most on here know about FLCK speed, UCLK ratio, but like most things in OC rules do not apply.

Yes, my basis is for dual CCD’s but not tested a 7950x non 3D CPU, but what testing have you done and what is your CPU.

 

[BlueShark]

How can the frequency CCD be used in a game (you mention here with Ray tracing).
I thought as long as the game is seen as game by Game bar, only CCD0 is used.

No, if the driver sees an overflow of sorts that the game needs more resources it will allow a single or couple of frequency cores to join in the gaming frenzy.

Plus, there are games like cyberpunk that as an AMD optimization that seems to use all 3D cores and one frequency core, so that means it as more memory bandwidth available for it to use.

 

[just20sent]

Hi boys, returning back from 13900ks to a 7800x3d, i forgot what settings to change in bios beyond PBO and EXPO! Thanks for your help!

 

[J7SC]

Can you share a screenshot? Because when I check inspector it seems like its already on for Cyberpunk for example.

So it's not really forced then? Just want to make sure I'm not missing out of anything in terms of extra performance.

...here you go; see blue-frame box and also note top left (_GLOBAL_DRIVER_PROFILE...)

 

[Ivan B.]

Ran the CPU and Memory BENCHMARK for the newest OCCT 13.0.0.b2
*CPU config in signature.
Memory latency 67.4ns reflects 10ns slower than Aida64: 57.4ns - Wonder if this is what others see..
View attachment 2654250 View attachment 2654251 View attachment 2654252
View attachment 2654254 View attachment 2654255

My setup is very similar. I wonder what speed the ddr5 8000 has.

 

[Kimir]

After a week of waiting, I emailed caseking to know the progress of my RMA of the 7950X3D that failed the OCCT core cycle test...
I was disappointed in their reply yo say the least.

We have received your item and regret to inform you that the fault you have indicated cannot be reproduced despite intensive, multiple checks. The goods are free of defects. We therefore reject your claim for supplementary performance within the scope of liability for material defects (§ 439 BGB).

We were able to benchmark your CPU using various stress tests. Among other things, OCCT was used for 2 hours and Cinebench and PRIME were run successfully for 2 hours. There were no error messages during the entire test period.


If you wish, we can forward your goods to the manufacturer for warranty testing.

This examination is expected to take between 4 and 6 weeks. Please note that in this case we only act as an intermediary for the inspection. We can neither speed up the result nor influence it in any other way.

There is no legal entitlement to this service from us or to testing by the manufacturer. If the inspection reveals that there is a defect and that it is a manufacturing defect, we will of course accept this as proof of a claim for rectification, provided that the deadlines for this have been met.

We must also draw your attention to the fact that repair costs may be incurred by the manufacturer. If the manufacturer estimates costs, we will inform you immediately by e-mail so that you can decide whether a repair is worthwhile in this case.



Please note that we cannot enforce a premature return of the goods from the manufacturer. If you agree to the above conditions, we will be happy to forward your goods to the manufacturer. Please give us your written consent within 7 days in response.

If you have any questions, please do not hesitate to contact us.

I'm not gonna shop there anymore, if they can't even replicate the issue by simply following the OCCT settings to do so.

 

[GRABibus]

After a week of waiting, I emailed caseking to know the progress of my RMA of the 7950X3D that failed the OCCT core cycle test...
I was disappointed in their reply yo say the least.

I'm not gonna shop there anymore, if they can't even replicate the issue by simply following the OCCT settings to do so.

You couldn’t apply the « 14 days » rétractation rule ? Which means full refund whatever the cpu is ok or not.
I did it with my ****ty sample from a French reseller.
I tested it, I was no satisfied.
I sent it back and was fully refunded today.

 

[Kimir]

You couldn’t apply the « 14 days » rétractation rule ? Which means full refund whatever the cpu is ok or not.
I did it with my ****ty sample from a French reseller.
I tested it, I was no satisfied.
I sent it back and was fully refunded today.

I can't, I bought it in December, was replaced once already after a sudden death. I'm messaging AMD, with the OCCT screen, I'll see what they have to say about it.
Such a headache...