AMD Ryzen 9 9950X3D and 9900X3D to Come with Clock Speeds Resembling non-X3D SKUs

[Hankieroseman]

I think I'll be in that line too for 2025. Assuming my X670E board supports it and gets a BIOS for it like 9800X3D.

 

[kapone32]

I think I'll be in that line too for 2025. Assuming my X670E board supports it and gets a BIOS for it like 9800X3D.

It's AMD the BIOS is probably already ready. The BIOS for 9000 was already released.

 

[evernessince]

But I guess even if scheduling was 100%, fixed there will still be people claiming "if only AMD had cache on both CCD's it would be even faster!"

Clearly there is a benefit to multiple stacks because AMD does have multiple stacks in it's server products. Maybe singular consumer workloads won't see much of a benefit but this is a 16 core CPU, there's a decent chance the user may be running multiple heavy programs at once. A scenario that multiple stacks of cache is more likely to benefit.

 

[izy]

I would still prefer a single CCD 10 / 12 cores 3D rather than 12 / 16 cores dual CCD.

 

[Hankieroseman]

It's AMD the BIOS is probably already ready. The BIOS for 9000 was already released.

I'm up to 2506 but I see a new 2604. Too bad cause this board is RMA.

 

[mama]

In my view the only way a 9900X3D makes sense is if both CCDs sport 3D vcache.

 

[AusWolf]

So then the rumours about 9000X3D having 3D cache on both CCDs are false, I suppose.

 

[dave2150]

The big question is will both CCDs have the stacked V-cache? I think the answer is no, because AMD still haven't solved how to stop them thrashing.

There is no big question - it's already been confirmed in multiple leaks that only one CCD has 3D cache.

 

[JoeTheDestroyer]

Not sure who's going to buy this rumor when we already know it's not true for the 9800x3d vs 9700x.

 

[Frick]

So no word on a 9600X3D? That's the one I'm interested in...

 

[Raiden85]

I really hope they can do a 32 core 64 thread CPU on AM5 if they can do 16 cores per CCD while maintaining high clocks and still use X3D cache. Even if its on one CCD with that ammount of cores that would be great.

Intel are already behind, this would bury them for quite a while.

 

[_roman_]

For 32 core cpus you should look in the threadripper platform

 

[Princess Garnet]

Having the potential to replace two computers with one is very nice. A person might buy a 7800X3D for their main rig and a 7950X for their secondary. A double X3D cache 9950X3D could stand in for two systems.

The x950X3D already does that, no? I'm not saying a CPU that bridges those uses shouldn't exist so this seems like a strange statement to make (likewise, the other two statements sort of respond to angles I never made).

What I was saying was that I imagine a second CCD with v-cache would return a very slim (often nil) level of extra performance in gaming, but that it would cost much more, and that I figure this might be the reason AMD hasn't done it. Nothing more, nothing less.

 

[AusWolf]

The x950X3D already does that, no? I'm not saying a CPU that bridges those uses shouldn't exist so this seems like a strange statement to make (likewise, the other two statements sort of respond to angles I never made).

What I was saying was that I imagine a second CCD with v-cache would return a very slim (often nil) level of extra performance in gaming, but that it would cost much more, and that I figure this might be the reason AMD hasn't done it. Nothing more, nothing less.

AMD's reasoning with costs is solid. My reasoning is that I don't want to rely on software scheduling to do stuff on my CPU. Therefore, I'll never buy a desktop CPU with heterogenous cores or CCDs.

 

[Grebber01]

They should really act and come up with a more powerful processor to replace the 8700G in IGP

 

[_roman_]

What I was saying was that I imagine a second CCD with v-cache would return a very slim (often nil) level of extra performance in gaming, but that it would cost much more, and that I figure this might be the reason AMD hasn't done it. Nothing more, nothing less.

You rely on the kernel scheduler to manage the tasks. This is a big overhead. For intel processors most likely not an issue because they have a lot of E-fake cores which can not do heavy calculations. Still a lot of waste of resources to manage such. I doubt the code is working. I doubt the code is mature for such mixed core processors, regardless if its with or without the cache or with or without p and e cores.

 

[Godrilla]

And here I thought there would be something different from Zen4 x3d linkup. Hey at least we have better memory support now and some oc capabilities to look forward too. Scheduler is the rate limiting step of being an ideal cpu outside of a dual cached ccd cpu.

 

[evernessince]

The x950X3D already does that, no? I'm not saying a CPU that bridges those uses shouldn't exist so this seems like a strange statement to make (likewise, the other two statements sort of respond to angles I never made).

What I was saying was that I imagine a second CCD with v-cache would return a very slim (often nil) level of extra performance in gaming, but that it would cost much more, and that I figure this might be the reason AMD hasn't done it. Nothing more, nothing less.

No the x950X3D don't because the crappy windows scheduler creates all sorts of issues in games for the dual ccd chips. The difference in TPU's charts doesn't show you that often windows will put the game on the wrong CCD which will cause the game to run 20-30% slower.

A 2nd X3D cache would fix that issue. The average in benchmarks would hardly improve but the consistency would astronomically improve.

As a result, a 9950X3D with only one X3D die is dead in the water for gaming. Absolutely would not recommend it over getting a 9800X3D for a gaming rig and a separate AMD or Arrow Lake rig. It's silly.

 

[kapone32]

No the x950X3D don't because the crappy windows scheduler creates all sorts of issues in games for the dual ccd chips. The difference in TPU's charts doesn't show you that often windows will put the game on the wrong CCD which will cause the game to run 20-30% slower.

A 2nd X3D cache would fix that issue. The average in benchmarks would hardly improve but the consistency would astronomically improve.

As a result, a 9950X3D with only one X3D die is dead in the water for gaming. Absolutely would not recommend it over getting a 9800X3D for a gaming rig and a separate AMD or Arrow Lake rig. It's silly.

Interesting. TPU should test City Skylines 2 at 1440P with 600000+ population and associated services. In that Game I see 100% CPU usage. Space Marine 2 also uses more than 60% of my CPU. BTW exactly how long is 1 nanosecond?

 

[dragontamer5788]

Interesting. TPU should test City Skylines 2 at 1440P with 600000+ population and associated services. In that Game I see 100% CPU usage. Space Marine 2 also uses more than 60% of my CPU. BTW exactly how long is 1 nanosecond?

100% CPU usage is weird when your CPUs are all different.

Remember that CPU usage is Windows measuring how much a core is active. It's not necessarily saying it's an efficient or useful amount of work.

In an extreme case: 100% usage on an E-cores which spends all of its time waiting on RAM might be more overall efficient than a 100% usage on a P-Cores that's stuck on a spinlock (thread synchronization instruction). Both are 100% usage.

It's more complex as different instructions now have different speeds on the same computer but different cores (x3d for AMD or p vs e core for Intel)

 

[kapone32]

100% CPU usage is weird when your CPUs are all different.

Remember that CPU usage is Windows measuring how much a core is active. It's not necessarily saying it's an efficient or useful amount of work.

In an extreme case: 100% usage on an E-cores which spends all of its time waiting on RAM might be more overall efficient than a 100% usage on a P-Cores that's stuck on a spinlock (thread synchronization instruction). Both are 100% usage.

It's more complex as different instructions now have different speeds on the same computer but different cores (x3d for AMD or p vs e core for Intel)

I use the Overlay in AMD software. It also has frame stutter and 1% lows.

 

[Princess Garnet]

No the x950X3D don't because the crappy windows scheduler creates all sorts of issues in games for the dual ccd chips. The difference in TPU's charts doesn't show you that often windows will put the game on the wrong CCD which will cause the game to run 20-30% slower.

A 2nd X3D cache would fix that issue. The average in benchmarks would hardly improve but the consistency would astronomically improve.

As a result, a 9950X3D with only one X3D die is dead in the water for gaming. Absolutely would not recommend it over getting a 9800X3D for a gaming rig and a separate AMD or Arrow Lake rig. It's silly.

Scheduling issues aren't the only thing working against the Ryzen 9 X3Ds to keep them from reaching the performance of the x800X3D. The other factor is cross CCD latency (the massive dropoff that the x900X3D in particular, a six core per CCD chip, sometimes shows could be taken as a sign that this is leading to a large loss of performance too), and adding v-cache to both CCDs doesn't address that at all. Improving the scheduler (which is probably far easier said than done) or increasing the numbers of cores per CCD does.

In any case, like I said, I was just explaining why I think the CPUs are probably like this, and at least for now (maybe it changes in the future). I think you're misunderstanding me as saying there would be no benefit at all to having v-cache on both CCDs when that's not what I'm saying. I just don't agree with the notion that they are entirely useless without it nor that adding v-cache to both CCDs fixes them as much as you think. This is why I suggested higher core counts per CCD, or as others have stated, scheduling improvements, as viable things to look forward to.