AMD Ryzen 9 9950X3D and 9900X3D to Feature 3D V-cache on Both CCD Chiplets

[ValenOne]

It is quite the contrary to what you say. Scheduling will now be even more important to the point it becomes SUPER-DUPER-MEGA-EXTRA-important with cache on both CCDs. For this dual cache setup to work correctly, games/apps (via the scheduler) always need to request the cached data from the "correct" cache on the "correct" CCD or else you will suffer latencies from hell if/when data needs to be fetched from the cache across the CCDs because e.g. Core 3 requests data that was previously stored to the cache by Core 14 on the other CCD. Can't have a scenario like that. Ever.

So, both the scheduler and the CPU always need to "know" exactly "who" (which core) cached something (what) and where it was cached to avoid the dreaded inter-CCD and inter-cache latencies. This is definitely going to be a challenge and very complex on the level of correct scheduling and correct CCD assignment etc.

AMD does not exactly have the best track record when it comes to these scheduling and core assignment shenanigans so I would be quite surprised if they get this to work flawlessly out of the gate.
Personally, I have avoided multi CCD CPUs like the plague due to the Xbox GameBar and 'GameMode On' requirements (I have a PC and not a console, you muppets). It will be interesting to see if the GameBar requirement will be dropped now(?) since core parking will no longer be required.

We'll have to wait and see how well this is gonna work in practice. I would expect some growing pains, to say the least...

The current Window 11 build is already aware of highest clocked CPU cores, hence only one good CDD is needed.

The problem with 7950X3D is non-X3D CCD has the higher clockspeed against lower clocked X3D CCD which is against standard Window 11's thread scheduler behaviour.

9900X3D/9950X3D should have one higher clocked CCD with X3D and lower clocked CCD with or without X3D.

 

[InVasMani]

I remember when 128MB of VRAM was like a big deal.

 

[LabRat 891]

Compelling, even for us 5#00X3D AM4ever'ers.

 

[igormp]

TPU tested Zen5 without SMT proved otherwise

We found the Missing Performance: Zen 5 Tested with SMT Disabled

Reviews of AMD’s Zen 5 processors this week surprised many, with lower-than-expected results. After some investigation, we discovered that turning off Simultaneous Multithreading (SMT) can yield notable performance gains, particularly in gaming. This article presents our findings, including...

www.techpowerup.com

Seems like only stuff that's really single threaded (hence making the whole point of SMT moot) got some benefits.
For most other stuff it barely made a difference (like it only being 2% better for games), while most other stuff actually regressed. On average it ended up regressing.

And another thing, if you only care about single threaded stuff, would you even go for a 9950x3D to begin with?

 

[Caring1]

Jayz2cents predicted this a few days ago in a recent video.

 

[ACE76]

Finally, now the people will see that the 3D cache on both dies is useless and will stop crying for this

Yeah people have no clue. Gaming performance won't go up with this...but specialized workloads that benefit from large cache and memory will get a boost. They're probably doing this to make it easier to use with the Windows scheduler which has been problematic until recently.

I guess you’ve never heard of thread scheduling and how hit or miss it is. This solves that problem. Even better if clock speeds can also be higher.

The scheduling issue has been resolved with current AMD drivers. I personally never had an issue since day 1 but it's definitely been addressed. There's even a system app called 3D cache running at all times and you can see it in task manager under system processes.

AMD needs to lower prices on the new 9000 series CPU's, period. They are just not value oriented enough and are way too expensive for what they bring to the table compared to AMD's own 7000 series and the Intel 13k series. Intel is offering 14 cores for as cheap as $300, while AMD are stuck with 6 cores at those prices, its absurd.

The 9600x is $280 for 6 cores and barely 3-4% improvement in gaming over the 7600x, around 9% improvement in multithreading, while Intel's 14600 offers 14 cores for $260. Its around 15% faster over the 9600x in multithreaded apps and offers similar gaming performance.

Thing is the 9600x should cost $230 at most, with the 7600x starting at $200 and the 7600 vanilla at $180. The 9700x trades blows with the 14600k in terms of gaming, but falls short in multithreading, while being $70 dollars more.

Intel e-cores are about the most useless thing ever conceived of in computing...and sadly a lot of people have fallen for it thinking they're getting some crazy superior CPU with 14-20 cores.

No, it just that everyone wonders what the next Intel GPU will be offering. No one really cares about the AI capabilities or the plus 5% performance from the CPU. Everyone wants to know if Battlemage will be a small jump or a huge jump for Intel. It will also be an indication if AMD will have a strong competition in x86 handheld consoles.

Intel's GPU troubles have little to do with the hardware. It's their driver's that has them many generations behind AMD and Nvidia in the GPU segment. Hopefully they've improved a bit with Battlemage but in all honesty they're still several years away from being able to compete well.

 

[nguyen]

And another thing, if you only care about single threaded stuff, would you even go for a 9950x3D to begin with?

Well when the 9950X3D is the fastest gaming CPU and it's awesome in productivity too then why not.

Basically AMD is fixing the current flaw of 7950X3D where sometimes the performance are just too erratic and that is unbecoming for an ultra high end product

 

[ACE76]

Well when the 9950X3D is the fastest gaming CPU and it's awesome in productivity too then why not.

Basically AMD is fixing the current flaw of 7950X3D where sometimes the performance are just too erratic and that is unbecoming for an ultra high end product

I have the 7950x3D and there were definitely issues earlier on but AMD addressed them with bios updates and new drivers... today it works flawlessly...it'll be interesting to see how thec9950x3D performs. I'm guessing it won't be as good as the 7950x3D because it'll have a 300-400mhz lower clock speed unless they've been able to address that. It'll probably be amazing at certain workloads that benefit from large cache and memory. To be able to compete with other flagship CPUs, it'll need a higher clock speed but that's been a problem for the cache because it's very sensitive to voltage and high clock speeds..I'm definitely looking forward to how it performs though. If it's clocked at even 5.5ghz, it'll be a massive win.

 

[nguyen]

I have the 7950x3D and there were definitely issues earlier on but AMD addressed them with bios updates and new drivers... today it works flawlessly...it'll be interesting to see how thec9950x3D performs. I'm guessing it won't be as good as the 7950x3D because it'll have a 300-400mhz lower clock speed unless they've been able to address that. It'll probably be amazing at certain workloads that benefit from large cache and memory. To be able to compete with other flagship CPUs, it'll need a higher clock speed but that's been a problem for the cache because it's very sensitive to voltage and high clock speeds..I'm definitely looking forward to how it performs though. If it's clocked at even 5.5ghz, it'll be a massive win.

How can you tell if you are getting the highest perf out of your 7950X3D in every game, without having some reference baseline?

YTbers like HUB and Der8auer said just recently that the 7950X3D sometimes just give lower FPS than it did before for some reason, HUB said after using it for 6 months Windows just give lower FPS. These guys know their 7950X3D are not up to snuff because they test hardware all the time, for average users that is not possible.

 

[RootinTootinPootin]

YTbers like HUB and Der8auer said just recently that the 7950X3D sometimes just give lower FPS than it did before for some reason

the huge dependency on the software as well for the driver to correct the scheduling of "Games" to the 3D Cores is kinda inefficient and inconsistent, IMHO, they just need to trash this idea on the next iteration.

 

[Minus Infinity]

I have the 7950x3D and there were definitely issues earlier on but AMD addressed them with bios updates and new drivers... today it works flawlessly...it'll be interesting to see how thec9950x3D performs. I'm guessing it won't be as good as the 7950x3D because it'll have a 300-400mhz lower clock speed unless they've been able to address that. It'll probably be amazing at certain workloads that benefit from large cache and memory. To be able to compete with other flagship CPUs, it'll need a higher clock speed but that's been a problem for the cache because it's very sensitive to voltage and high clock speeds..I'm definitely looking forward to how it performs though. If it's clocked at even 5.5ghz, it'll be a massive win.

Not a snowball's chance in hell 9950X3D clocks lower than 7950X3D, quite the opposite for all the X3D parts this gen. AMD has been dropping hints all over the place.

 

[ACE76]

How can you tell if you are getting the highest perf out of your 7950X3D in every game, without having some reference baseline?

YTbers like HUB and Der8auer said just recently that the 7950X3D sometimes just give lower FPS than it did before for some reason, HUB said after using it for 6 months Windows just give lower FPS. These guys know their 7950X3D are not up to snuff because they test hardware all the time, for average users that is not possible.

Earlier on, there were definitely issues. Scheduler issues, etc...but they've fixed it. I haven't had anything but the max performance in all games now. JaysTwoCents did a video on this a few weeks back highlighting a BIOS option. I think some motherboards *might still cause issues but my Asus x670e Extreme hasn't had any with the latest BIOS, chipset drivers and CPU drivers. You can even check task manager to make sure AMD's vcache app is running. If it is, you won't have any problems in games.

 

[igormp]

Intel e-cores are about the most useless thing ever conceived of in computing...

e-cores are great for multi-threading and efficiency. It's a tried and true concept in other areas that's finally being brought to the x86 world.
Heck, even their all e-cores xeon looks amazing and provides awesome value.

Now, if all you care is about gaming, then yeah, it's pretty useless.

Well when the 9950X3D is the fastest gaming CPU

Why would it be faster than a 9800x3D?

it's awesome in productivity too then why not.

Not that awesome after you disable SMT. Unless you're ok with toggling back on everytime, then I guess it's fair.

 

[Darmok N Jalad]

e-cores are great for multi-threading and efficiency. It's a tried and true concept in other areas that's finally being brought to the x86 world.
Heck, even their all e-cores xeon looks amazing and provides awesome value.

Now, if all you care is about gaming, then yeah, it's pretty useless.


Why would it be faster than a 9800x3D?

Not that awesome after you disable SMT. Unless you're ok with toggling back on everytime, then I guess it's fair.

I have an Adler lake cpu in my work laptop, and I’m not sold on the E-core concept. There are too many instances where my PC crawls at basic office workloads due to scheduling issues. Part of that is Intel specced an i7 to be 2P+8E, and often times MT tasks go the actual P cores, and not even use HT. I think the 4+4 concept is way better.

 

[dyonoctis]

Intel e-cores are about the most useless thing ever conceived of in computing...and sadly a lot of people have fallen for it thinking they're getting some crazy superior CPU with 14-20 cores.

So it's fine when Apple does it, as soon as it's intel it's a bad tech ?

 

[Darmok N Jalad]

So it's fine when Apple does it, as soon as it's intel it's a bad tech ?

No, it's not just an Intel thing. Microsoft and Windows clearly struggle with using the right core(s) for the job. It's one thing to load all of them like in Cinebench or Handbrake, but what about an app that doesn't assign 100% workloads across all cores? That's when we see the struggle, and it's a real struggle as far as my own personal experience is concerned. This is what I experience quite often on my 2P+8E i7 work machine--Excel grinding away on calculations. Note the affinity for real P-cores only. I'd be better off with just 4C/8T.

 

[A Computer Guy]

No, it's not just an Intel thing. Microsoft and Windows clearly struggle with using the right core(s) for the job. It's one thing to load all of them like in Cinebench or Handbrake, but what about an app that doesn't assign 100% workloads across all cores? That's when we see the struggle, and it's a real struggle as far as my own personal experience is concerned. This is what I experience quite often on my 2P+8E i7 work machine--Excel grinding away on calculations. Note the affinity for real P-cores only. I'd be better off with just 4C/8T.
View attachment 365329

I notice something interesting too. On my 5950x a 7zip 32 thread workload goes all core but 7950x it's not so and CCD2 isn't working to it's maximum capability unless I set 7zip to do a 64 thread workload on a 32 thread cpu. Not sure if something is wrong there but it seemed odd.

 

[dyonoctis]

No, it's not just an Intel thing. Microsoft and Windows clearly struggle with using the right core(s) for the job. It's one thing to load all of them like in Cinebench or Handbrake, but what about an app that doesn't assign 100% workloads across all cores? That's when we see the struggle, and it's a real struggle as far as my own personal experience is concerned. This is what I experience quite often on my 2P+8E i7 work machine--Excel grinding away on calculations. Note the affinity for real P-cores only. I'd be better off with just 4C/8T.
View attachment 365329

Indeed I can see how it's a problem, but It seems that Intel tweaked the behavior of the e-cores from Lunar Lake and onwards, on mobile a task will move to the P-core only if the e-cores can't handle it, which seems to be different from what I'm observing right now where it's very background focused.

I don't exactly know how the thread scheduler judges if a task is too heavy or not though, but recent Lunar lake benchmarks seems to suggest that it's working pretty well

 

[igormp]

I have an Adler lake cpu in my work laptop, and I’m not sold on the E-core concept. There are too many instances where my PC crawls at basic office workloads due to scheduling issues. Part of that is Intel specced an i7 to be 2P+8E, and often times MT tasks go the actual P cores, and not even use HT. I think the 4+4 concept is way better.

Eh, I'd have expected such things to now be an issue on windows anymore due to that hw director. Seems like the windows' scheduler is still a pain even with that.

 

[R0H1T]

How do you know it's not the thread director that's the issue?

 

[JWNoctis]

No, it's not just an Intel thing. Microsoft and Windows clearly struggle with using the right core(s) for the job. It's one thing to load all of them like in Cinebench or Handbrake, but what about an app that doesn't assign 100% workloads across all cores? That's when we see the struggle, and it's a real struggle as far as my own personal experience is concerned. This is what I experience quite often on my 2P+8E i7 work machine--Excel grinding away on calculations. Note the affinity for real P-cores only. I'd be better off with just 4C/8T.
View attachment 365329

I think Intel might have seen the same thing you had. Lunar Cove Lake is 4C4T+4c4t, and kicked everything else off the core ringbus so that only the P-cores stayed. It would be close to Tiger Lake, minus the hyperthreading and plus the architectural improvements.

Including the E-cores in a lot of workloads with such bus architecture would probably slow things down even further, if it ends up causing problems in the P-cores' cache hierarchy.

 

[goose280672]

What happened to the 8xxx series? Why the jump from 7 to 9??

What do you mean, 8000 series was released a while ago, have a look at any PC parts store.

 

[duraz0rz]

Is possible because more cache space.

I don't know if you're serious or not, but that's not how cache hits and misses work.

 

[mkppo]

Earlier on, there were definitely issues. Scheduler issues, etc...but they've fixed it. I haven't had anything but the max performance in all games now. JaysTwoCents did a video on this a few weeks back highlighting a BIOS option. I think some motherboards *might still cause issues but my Asus x670e Extreme hasn't had any with the latest BIOS, chipset drivers and CPU drivers. You can even check task manager to make sure AMD's vcache app is running. If it is, you won't have any problems in games.

Yeah It's silly how people go on and on about 7950x3d scheduler issues when a) it was mostly during launch where a couple of games had issues b) most issues are fixed by now c) worst case you can use project lasso for that really random game that seems to be having performance issues d) disable second CCD if you want something faster than 7800x3d

I have both and haven't found a single game where it doesn't match the 7800x3d but I only play a handful of games.

 

[chrcoluk]

Oh boy, is 9000 revived with this?

No, it's not just an Intel thing. Microsoft and Windows clearly struggle with using the right core(s) for the job. It's one thing to load all of them like in Cinebench or Handbrake, but what about an app that doesn't assign 100% workloads across all cores? That's when we see the struggle, and it's a real struggle as far as my own personal experience is concerned. This is what I experience quite often on my 2P+8E i7 work machine--Excel grinding away on calculations. Note the affinity for real P-cores only. I'd be better off with just 4C/8T.
View attachment 365329

It overloads preferred cores, I down clocked my 2 fastest cores last week to solve the scheduling mess that is Windows. Simply couldnt find a solution in all of the hidden scheduler tunables. A circa 2% faster clock speed Windows is like ok I am going to put all the threads on this baby.