• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Zen 5 rumours

 
That will be interesting to see, will that mean the end of core parking?

On another note

X670E Gen5 SSD issues


Asus response doesn't fill me with hope that they have improved their customer support issues.
I hope this is OK on X870 as I have a 4TB T700 Gen 5 and a 8TB SN850X Gen 4 bought for my new build
 
^That's quite shocking from Asus, basically go away and chase it up with the drive manufacturer when it's a motherboard issue that seems to be at fault. Then again why am I not surprised by their response after their customer service went down the pan years ago? Running a gen 5 drive at gen 4 speeds seems to be a cure and you won't notice the difference in speed outside of a benchmark anyway. Sucks that people paid the extra cash for a gen 5 drive only to have to run it at gen 4 speeds though. They could have saved a fair chunk of cash buying a gen 4 drive instead.
 
The 9900x3d will be ideal for me if it has cache on both ccds. Really is what I have been waiting for to replace my 5900x. When are these expected to release?
 
Last edited:
Not sure what dual 3d cache ccds will do for gaming when ccd to ccd has lantancy and clocks reduced compared to the none cache ccd so reduced performance with most tasks that 3d cache makes no difference

Let's see what these new features are
 
Last edited:

Such a boring platform due to the severe bottleneck of the IOD. Oh well.
 
Last edited:
The 9900x3d will be ideal for me if it has cache on both ccds. Really is what I have been waiting for to replace my 5900x. When are these expected to release?
Yeah if they sort out the inter-CCD latency this would fall right in to my view - I'm pretty happy with my 5950x at the moment but a small jump on the 7950x3D and without the worry about thread allocation would get my attention (although tbh it'll have to be a decent uplift from 5950x to get my attention - 7950x3D at ~20-30% for threaded loads wasn't quite there)
 
I was under the impression that AMD/Windows had figured out how to get games to stay on a single CCD back when I was gaming on my 5950X.

However, when I heard about the dual-CCD Vcache chips needing game bar, it sounded like they had introduced another hurdle in that picking "a" CCD and sticking too it was no longer good enough. Now they had to pick a *specific* CCD for each specific workload and also remember to stick to it.

Dual Vcache chips simplifies things back to where just picking "a" CCD and sticking to it will work.

I like this idea.
 

Such a boring platform due to the severe bottleneck of the IOD. Oh well.
8000C38 what the hell, and here's me running 8000C32 as a daily.

What FCLK are they running with 8000?

Handy tip for non X3D dual CCD bois, see below.

My timings, Windows 10. Cyberpunk 2077, high preset nothing else changed.


SMT On Core Parking Enabled


SMT On Core Parking Off


SMT Off Core Parking On


SMT Off Core Parking Off


SMT Off Core Parking Off + Max CPU fan speed


If you are using a 9900/9950X, I would recommend that if you want the absolute best performance, use the High Performance Windows Power profile. This will use a bit more power, as games can use 'all cores', whilst still preferring the higher performance cores. When you use the Balanced profile, games will mostly stick to CCD0 and will therefore use a bit less CPU package power. However, if CPU utilization is high enough, cores on the second CCD can still be used so one or more may wake up. However, this still leaves a bit of performance on the table vs using the High Performance power profile which basically disables Core Parking. And yeah, Cyberpunk prefers SMT off as you can see above.

If you are wondering how this compares to a tightly tuned 7800X3D running 8000Mhz C32 with 2200 FCLK see below.


3% faster than a tuned 9950X in this title. And disabling SMT with the 7800X3D hurts performance due to only having 8 cores.
 
Last edited:

Such a boring platform due to the severe bottleneck of the IOD. Oh well.

Yet it still gives the best gaming performance, and minimum effort needed to setup, some prefer plug n play with minimum effort

Looking forward how arrow lake and new x3d compare in gaming
 
Last edited:
8000C38 what the hell, and here's me running 8000C32 as a daily.

What FCLK are they running with 8000?

Handy tip for non X3D dual CCD bois, see below.

My timings, Windows 10. Cyberpunk 2077, high preset nothing else changed.


SMT On Core Parking Enabled


SMT On Core Parking Off


SMT Off Core Parking On


SMT Off Core Parking Off


SMT Off Core Parking Off + Max CPU fan speed


If you are using a 9900/9950X, I would recommend that if you want the absolute best performance, use the High Performance Windows Power profile. This will use a bit more power, as games can use 'all cores', whilst still preferring the higher performance cores. When you use the Balanced profile, games will mostly stick to CCD0 and will therefore use a bit less CPU package power. However, if CPU utilization is high enough, cores on the second CCD can still be used so one or more may wake up. However, this still leaves a bit of performance on the table vs using the High Performance power profile which basically disables Core Parking. And yeah, Cyberpunk prefers SMT off as you can see above.

If you are wondering how this compares to a tightly tuned 7800X3D running 8000Mhz C32 with 2200 FCLK see below.


3% faster than a tuned 9950X in this title. And disabling SMT with the 7800X3D hurts performance due to only having 8 cores.


I'm going to have so many questions for you when I start my 9950X build as RAM overclocking is not something I have tried before. I had already bookmarked your post showing your timings from earlier.
I'm still struggling with what RAM to get, I was waiting for the GSkill 64GB kit @ CL28-36-36-96, I'm hoping it will appear with X870 next week, but if it doesn't then I need a plan B, ideally Id like 64GB, but I could drop down to 48GB, and maybe go for some teamgroup Xtreme. Thing is that techpowerup article has me thinking, I could go for 8000 RAM to get over the 2:1 latency but the silicon lottery may prevent that if I get a bad IMC, or I could go for the 6400 kit and try to go for CL28 timings instead. Im really not sure what way to go.
Im sticking your screenshots into One Note to look at when I get back later
 
I'm going to have so many questions for you when I start my 9950X build as RAM overclocking is not something I have tried before. I had already bookmarked your post showing your timings from earlier.
I'm still struggling with what RAM to get, I was waiting for the GSkill 64GB kit @ CL28-36-36-96, I'm hoping it will appear with X870 next week, but if it doesn't then I need a plan B, ideally Id like 64GB, but I could drop down to 48GB, and maybe go for some teamgroup Xtreme. Thing is that techpowerup article has me thinking, I could go for 8000 RAM to get over the 2:1 latency but the silicon lottery may prevent that if I get a bad IMC, or I could go for the 6400 kit and try to go for CL28 timings instead. Im really not sure what way to go.
Im sticking your screenshots into One Note to look at when I get back later
Yeah feel free to ping me if needed dude.

If you want to maximise memory frequency, you'll need either a 32GB A Die kit (F5-8000J3848H16GX2-TZ5NR) for 8000Mhz or a 48GB M Die kit (F5-8400J4052G24GX2-TZ5RK) for 8200/8400Mhz.

Can't go wrong with GSkill kits either, so stick with those.

With a bit of luck you should be able to run 2200 FCLK like me, helps provide extra bandwidth at these higher memory frequencies. I've tried half a dozen 9950X, and only one of them didn't like FCLK at 2200 so odds should be in your favour.
 
Yeah feel free to ping me if needed dude.

If you want to maximise memory frequency, you'll need either a 32GB A Die kit (F5-8000J3848H16GX2-TZ5NR) for 8000Mhz or a 48GB M Die kit (F5-8400J4052G24GX2-TZ5RK) for 8200/8400Mhz.

Can't go wrong with GSkill kits either, so stick with those.

With a bit of luck you should be able to run 2200 FCLK like me, helps provide extra bandwidth at these higher memory frequencies. I've tried half a dozen 9950X, and only one of them didn't like FCLK at 2200 so odds should be in your favour.
Thanks for that - if the CL28 Gskill doesn't come out in time I will go for the gskill trident z5 or the 8400 kit from teamgroup from somewhere else as OCUK don't do it. I was leaning towards the teamgroup as plan B as their heatsinks review very well.

I will pay close attention to the Die numbers, that is useful info.

Only another 700 or so to spend then (RAM + X870) and Im done :eek:
 
I'm going to have so many questions for you when I start my 9950X build as RAM overclocking is not something I have tried before. I had already bookmarked your post showing your timings from earlier.
I'm still struggling with what RAM to get, I was waiting for the GSkill 64GB kit @ CL28-36-36-96, I'm hoping it will appear with X870 next week, but if it doesn't then I need a plan B, ideally Id like 64GB, but I could drop down to 48GB, and maybe go for some teamgroup Xtreme. Thing is that techpowerup article has me thinking, I could go for 8000 RAM to get over the 2:1 latency but the silicon lottery may prevent that if I get a bad IMC, or I could go for the 6400 kit and try to go for CL28 timings instead. Im really not sure what way to go.
Im sticking your screenshots into One Note to look at when I get back later

The Gskill Royals have been available since late August, but only in America, they are pretty expensive and you probably have a long delivery wait and pay a bit of import tax. I decided not to bother and got the CL30 Neo instead (30% cheaper) - they were very easy to overclock to 1.4v CL28 with no stability issues. That said benchmark difference was negligible so I question whether the Royals are actually worth it unless they have substantially better chips.
 
Last edited:
The Gskill Royals have been available since late August, but only in America, they are pretty expensive and you probably have a long delivery wait and pay a bit of import tax. I decided not to bother and got the CL30 Neo instead (30% cheaper) - they were very easy to overclock to 1.4v CL28 with no stability issues. That said benchmark difference was negligible so I question whether the Royals are actually worth it unless they have substantially better chips.


This guy seemed pretty impressed with them in the benchies, though it would have been good to see him go up against 8000MT ram to see if that made a difference.


Thanks for the heads up for the GSkill though, looks like I better start researching plan B
 
I was under the impression that AMD/Windows had figured out how to get games to stay on a single CCD back when I was gaming on my 5950X.

However, when I heard about the dual-CCD Vcache chips needing game bar, it sounded like they had introduced another hurdle in that picking "a" CCD and sticking too it was no longer good enough. Now they had to pick a *specific* CCD for each specific workload and also remember to stick to it.

Dual Vcache chips simplifies things back to where just picking "a" CCD and sticking to it will work.

I like this idea.

Dual vcache solves one problem but it only works if inter ccd latency is low and both CCDs run at the same clocks. Currently on ryzen 9000 models the two CCDs do not run at the same clocks and remains to be seen if x3d is the same or not.

There are various issues with keeping gaming threads on the right ccd. A recent example is that with a 9950x and 9900x, AMD says you gotta keep windows in balanced power profile mode; but if you're a VR gamer and you try to play a VR game on Steam, Steam automatically puts windows into High performance power mode, causing the game to be on the wrong CCD
 
Last edited:
Back
Top Bottom