News Ryzen 9 9950X3D purportedly surfaces with 5.65 GHz boost clocks — leaked specs inlcude 96MB 3D V-Cache on a single CCD and 170W TDP

Given that the regular dual CCD Zen 5 parts require the same software to be installed as the dual CCD Zen 4 X3D parts nothing is going to save the scheduling issues. Bouncing between CCDs is still awful so it would make sense that they're sticking to a single CCD for the 3D V-cache given the lack of advantage for it being on both.

The biggest deal with these parts is that the higher clocks should make them a less compromised heavy load part while being great for gaming. I think the 9950X3D will probably end up being the best all around CPU on the market until next generation architectures arrive.
 
And surely that's the only config that makes any real sense because it avoids all the Windoze oh-dear-which-CCD-should-I-run-this-thread-on nonsense.
I was just discussing this aspect with a colleague. My best guess is that including 3D vcache on both would cut into TR/workstation territory more so now that the thermal issue has apparently been solved by bottom-die attachment. Placing the burden on Microsoft may not be too bad, as they should have a good head start since 79xxX3D support is fairly mature.
 
Will higher clocks though be enough for the 9900X3D not to suck this time? 7900X3D was poor for productivity and weaker at gaming not only due to lower clocks but some games use more than 6 cores and having to access non v-cache ccd hurt badly. 9900X3D will have much better clocks so shouod be great for productivity but will still only have 6 cores.

I suspect until Zen 6 launches where it is rumoured ccd's grow to 12 or 16 cores, x900X3D series will remain a worse choice for gaming. So 10900X3D with just 1 ccd and 12 cores would be a beast

I have no interest in 9950X3D, already 9950X is over $1000 in Australia at 99% of stores, so X3D version will hit $1200+.
 
I have no interest in 9950X3D, already 9950X is over $1000 in Australia at 99% of stores, so X3D version will hit $1200+.
Prices will come down eventually, even if it's after the 10k series (??) comes out.

I'm happy with my 5950X, but when it's time to upgrade, the 9950X3D is looking appealing.

After prices go down...
 
Interesting if true. Both CCDs getting cache this time around was widely rumored.
No reputable source leaked any such rumor. Based on every other product AMD has released/cancelled up to this point, there was zero chance this was going to happen. Anyone who thought it would happen does not understand how the technology works.
What will Intel's answer be? And if they don't have any answer what kind of future does Intel have?
They don't need one. The market for people who want to game on their workstation is not a market worth Intel's time. Intel needs to focus on markets that actually matter and can generate sufficient revenue and profits to offset what their foundry business is doing right now.
 
Given that the regular dual CCD Zen 5 parts require the same software to be installed as the dual CCD Zen 4 X3D parts nothing is going to save the scheduling issues. Bouncing between CCDs is still awful so it would make sense that they're sticking to a single CCD for the 3D V-cache given the lack of advantage for it being on both.

The biggest deal with these parts is that the higher clocks should make them a less compromised heavy load part while being great for gaming. I think the 9950X3D will probably end up being the best all around CPU on the market until next generation architectures arrive.
My understanding is the scheduling is still basically non-existant... that whenever a game launch is detected, all cores on the non-x3d ccd are parked. To me that says every time a game is in flight half the cpu is disabled. That's a total non-starter to my way of thinking. The main reason I would opt for a 16 core part is so I can play a game while I'm waiting on status bars. I know I'm not a usual use case, but I may have up to a dozen VMs running at any time when I launch a game.

Am I misunderstanding how the scheduling works with these? Is it unreasonable to expect something as smooth as the way threads slide around P and E cores on an Intel platform? X3d has been around for many years at this point.
 
meh, this release reflects a bit of laurel resting. we've been at a 100Mhz bus speed (yes I understand it's quad pumped) for some time.... and the cache on only one CCD set again is disappointing.
 
My understanding is the scheduling is still basically non-existant... that whenever a game launch is detected, all cores on the non-x3d ccd are parked. To me that says every time a game is in flight half the cpu is disabled. That's a total non-starter to my way of thinking. The main reason I would opt for a 16 core part is so I can play a game while I'm waiting on status bars. I know I'm not a usual use case, but I may have up to a dozen VMs running at any time when I launch a game.

Am I misunderstanding how the scheduling works with these? Is it unreasonable to expect something as smooth as the way threads slide around P and E cores on an Intel platform? X3d has been around for many years at this point.
AMD's software solution should act no differently than Process Lasso does. I believe it requires the game bar to work which leads me to believe that's how they detect games. So while it should prevent a game from leaving a single CCD it shouldn't have any impact on the usage of the other. AFAIK only their turbo mode locks out parts of the CPU, but that also requires a reboot to enable/disable as it's really a one button hardware solution.
 
@spongiemaster
Doesn't AMD use their 3D vcache in server/workstation parts? I had thought AMD was making serious in-roads into the server market as well w/their threadrippers and EPYC parts?
Yes. The massively parallel workloads that can be generated in the enterprise market are not as latency sensitive like gaming is. It's the same reason there are super computer with 100's of thousands of GPU's, but the gaming market has abandoned even using 2 GPU's at the same time.
 
@spongiemaster
So Intel is losing ground in both the server/workstation market AND the gaming market to AMD?
Yes, AMD has been chipping away for the past few years in most markets. They're still not close to Intel in market share. Intel commands over 75% of the x86 market for computers.

gWQLWQPZKQcgydNJJGeiy9-1200-80.png.webp
 
  • Like
Reactions: 80251
AMD's software solution should act no differently than Process Lasso does. I believe it requires the game bar to work which leads me to believe that's how they detect games. So while it should prevent a game from leaving a single CCD it shouldn't have any impact on the usage of the other. AFAIK only their turbo mode locks out parts of the CPU, but that also requires a reboot to enable/disable as it's really a one button hardware solution.
I did some (re)research to verify, and from what I can tell, cores 16-32 are parked when a game launch is detected and no threads can be scheduled on those cores. In other words, it's significantly dumber than process lasso, and it's required on all dual ccd, 9000 series chips even if they aren't x3d.
 
What will Intel's answer be? And if they don't have any answer what kind of future does Intel have?
We have seen intel's answer for years now, they increase the cache bit by bit instead of all at once like amd does, that way they don't harm the sale of their normal cpus, since all of them are the same, while amd has a real issue with selling their non x3d cpus because of how much better the x3d ones are for the mainstream users.

https://www.google.com/search?q=amd+zen+5+not+selling
 
  • Like
Reactions: 80251
@TerryLaze but the vast majority of CPU sales aren't for mainstream users right?

@spongiemaster
I can't believe Intel has that kind of market lead over AMD considering their CPUs don't have the performance edge anywhere anymore. I seem to remember AMD taking the perf. crown before way back w/their server opteron parts (back when Intel was still selling the Pentium D), then Intel wrested it back w/their core series of CPUs.
 
@TerryLaze but the vast majority of CPU sales aren't for mainstream users right?

@spongiemaster
I can't believe Intel has that kind of market lead over AMD considering their CPUs don't have the performance edge anywhere anymore. I seem to remember AMD taking the perf. crown before way back w/their server opteron parts (back when Intel was still selling the Pentium D), then Intel wrested it back w/their core series of CPUs.
The 3 largest OEM's, Lenovo, HP and Dell, ship around 13 million PC's a month (40 million\quarter). Almost all of those are Intel based.


What you see on enthusiast websites like this one are articles about Mindfactory sales:

https://www.tomshardware.com/pc-com...ultra-200s-sales-stagnate-after-just-one-week

ROFLCOPTER! The 7800X3D outsold all Intel CPU's combined! Intel is dead. Why do they even bother anymore?

AMD sold 730 CPU's in a week to Intel's 40. That's about 3000 AMD CPU's sold by Mindfactory in a month. Meanwhile, Intel sold 13 million CPU's to the top three OEM's over that same time span. Mindfactory sales numbers are completely worthless as a barometer for global sales data.
 
Last edited:
  • Like
Reactions: 80251
Thanks for the info. I mistakenly assumed a parked core wouldn't get threads scheduled, but that video clearly shows a parked core waking up to accept new work.
I understand that and this is what happens when AMD isn't clear about how it works (I had to check several places to verify the actual behavior). While parking is just a CPU state it is possible to lock that in and AMD doesn't refer to anything beyond core parking.