• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

AMD officially announces 9950X3D2, dual 3D V-Cache CCD 16-core/32-thread AM5 chip

Wow, did not expect Corridor Digital to be in there.

Now, the real question, what's the cost?
 
Wonder if I would even see/feel a difference on my system based on use - I am sure benchmarks would see something but how much I wonder?
 
Random thought - is there a direct(ish) connection between the two caches? To help when a thread jumps cores?
 
Elf_Boy said:
Random thought - is there a direct(ish) connection between the two caches? To help when a thread jumps cores?
Click to expand...
My suspicion would be no, so that could be a problem. For zen 7 i believe the solution is sea of wires that is supposed to dramatically reduce that effect.
 
Twisted Kidney said:
It will be interesting to see what this gets you performance-wise. With most of the scheduler problems ironed out from the 9950X3D I have to wonder what it will do.
Click to expand...

Yeah, I am curious too. I wonder how it handles the two sets of caches, and the need to sometimes access data that is in the opposite cache.
 
I don't know about AMD's supply on the 9950X3D, but if the could "chunk it", I would. And make the 9950X3D2 that replacement. When you the one "guessing" at how to win, like Intel, sure, release a confusing array of products and see what sticks. But, I think (and I'm guessing) a release of a much more expensive "top tier" consumer X3D2 part that will exist alongside a slightly slower X3D part... IMHO, I'd want to avoid that story.

Edit: with regards to "how to handle", I doubt it's much different from dual socket scenarios. So, the benefit will far outweigh anything else and at least schedulers can look at it from a dual socket point of view. Which means we can use the CPU instead of doing "forced parking" to get uniformity and throw away half the capability.
 
Some info:

Ryzen 9 9950X3D2 Dual Edition

16-core, 32-thread
base clock of 4.3 GHz and a boost clock of 5.6 GHz (100 MHz lower than the Ryzen 9 9950X3D)

Previous Ryzen 9000X3D chips had 3D V-Cache on only one CCD, this one has vertically stacked cache on both core complexes for 208MB total cache. Compared to the 16C/32T 9950X3D with “only” 128MB.


TDP of 200 watts (compared to 170W for 9950X3D). So that’s a noticable power increase for same core count at slightly lower clock speed.

Announced release date is April 22, no price given yet.
 
Zarathustra[H] said:
Yeah, I am curious too. I wonder how it handles the two sets of caches, and the need to sometimes access data that is in the opposite cache.
Click to expand...
Probably no different than how the current processors handle it. Just more lag time if there's a need to transfer large amounts of data, so software optimized for it would do its best to prevent such transfers from happening as much as possible, which should already be happening. I doubt AMD made the Infinity Fabric any faster just for the 9950X3D2. They'll probably have most of the inter-CCD tweaks reserved for Zen 6.
cjcox said:
I don't know about AMD's supply on the 9950X3D, but if the could "chunk it", I would. And make the 9950X3D2 that replacement. When you the one "guessing" at how to win, like Intel, sure, release a confusing array of products and see what sticks. But, I think (and I'm guessing) a release of a much more expensive "top tier" consumer X3D2 part that will exist alongside a slightly slower X3D part... IMHO, I'd want to avoid that story.

Edit: with regards to "how to handle", I doubt it's much different from dual socket scenarios. So, the benefit will far outweigh anything else and at least schedulers can look at it from a dual socket point of view. Which means we can use the CPU instead of doing "forced parking" to get uniformity and throw away half the capability.
Click to expand...
Price. 3D cache is expensive, and I expect the X3D2 to be at least $150 more expensive than the X3D. I wouldn't be surprised if AMD puts it at the $1000 mark just to have a halo product there, then lower the price later.

As for how much faster the X3D2 will be- it will be primarily limited to workstation applications that scale beyond 8 cores and can utilize the large amounts of cache. If the workstation application is 20% faster on the 9800X3D as compared to the 9700X, then the 9950X3D2 should be 20% faster than the 9950X and about 10% faster than the 9950X3D, with the amount of gain being highly dependent on the workload itself (using these numbers as an example, I don't know which and how much workstation applications benefit from V-cache). For games, it won't be any quicker, at least not noticeably. In a sense, the 9950X3D2 will be the new HEDT chip, but still on a consumer platform with limited PCI-E lanes.
westrock2000 said:
Some info:

Ryzen 9 9950X3D2 Dual Edition

16-core, 32-thread
base clock of 4.3 GHz and a boost clock of 5.6 GHz (100 MHz lower than the Ryzen 9 9950X3D)

Previous Ryzen 9000X3D chips had 3D V-Cache on only one CCD, this one has vertically stacked cache on both core complexes for 208MB total cache. Compared to the 16C/32T 9950X3D with “only” 128MB.


TDP of 200 watts (compared to 170W for 9950X3D). So that’s a noticable power increase for same core count at slightly lower clock speed.

Announced release date is April 22, no price given yet.
Click to expand...
I highly doubt that power consumption is actually higher and it's more that it's simply harder to get rid of the heat due to the v-cache. The 9950X3D generally draws about the same amount of power as the 9950X at the same clock speeds and I would expect the 9950X3D2 to be similar.
 
Tsumi said:
Probably no different than how the current processors handle it. Just more lag time if there's a need to transfer large amounts of data, so software optimized for it would do its best to prevent such transfers from happening as much as possible, which should already be happening. I doubt AMD made the Infinity Fabric any faster just for the 9950X3D2. They'll probably have most of the inter-CCD tweaks reserved for Zen 6.

Price. 3D cache is expensive, and I expect the X3D2 to be at least $150 more expensive than the X3D. I wouldn't be surprised if AMD puts it at the $1000 mark just to have a halo product there, then lower the price later.

As for how much faster the X3D2 will be- it will be primarily limited to workstation applications that scale beyond 8 cores and can utilize the large amounts of cache. If the workstation application is 20% faster on the 9800X3D as compared to the 9700X, then the 9950X3D2 should be 20% faster than the 9950X and about 10% faster than the 9950X3D, with the amount of gain being highly dependent on the workload itself (using these numbers as an example, I don't know which and how much workstation applications benefit from V-cache). For games, it won't be any quicker, at least not noticeably. In a sense, the 9950X3D2 will be the new HEDT chip, but still on a consumer platform with limited PCI-E lanes.

I highly doubt that power consumption is actually higher and it's more that it's simply harder to get rid of the heat due to the v-cache. The 9950X3D generally draws about the same amount of power as the 9950X at the same clock speeds and I would expect the 9950X3D2 to be similar.
Click to expand...
I'm trying to convince myself that I don't need this... I'm running a 7700x, I need to clean my loop anyways, and I do use Adobe CC apps and do video encoding. Multiple excuses in favor of getting one. I just feel like this beast is gonna be the eye-watering type of expensive. Like $800-1000, ugh.
 
I expect it to cost a decent chunk o change for sure. My AM5 setup has a 7700 and I'm holding off to see if AMD releases a 12 core or more singe CCD part before retiring the socket. If not I suppose something like this will top it off some day.
 
Tsumi said:
As for how much faster the X3D2 will be- it will be primarily limited to workstation applications that scale beyond 8 cores and can utilize the large amounts of cache.
Click to expand...
C compilers say "hi". Technically, they're more or less single-threaded, but make and more modern build systems are capable of running as many copies as you want, and 28 threads on an i7-14700K are absurdly fast at compiling large projects. (I have my Snapdragon laptop as a point of comparison and you can tell.)

Any language's compiler with a similar model, like Java, will probably work the same way, too.

This isn't really to disagree with you so much as to point out that there's more things that can take advantage of it. I compiled a 23-file microcontroller project in a about a second (885ms, including "make clean" first).
 
  • Like
Reactions: kac77
like this
1_rick said:
C compilers say "hi". Technically, they're more or less single-threaded, but make and more modern build systems are capable of running as many copies as you want, and 28 threads on an i7-14700K are absurdly fast at compiling large projects. (I have my Snapdragon laptop as a point of comparison and you can tell.)

Any language's compiler with a similar model, like Java, will probably work the same way, too.

This isn't really to disagree with you so much as to point out that there's more things that can take advantage of it. I compiled a 23-file microcontroller project in a about a second (885ms, including "make clean" first).
Click to expand...

I mean if you're already compiling large projects in a second, do you need faster? You can't be doing it often enough to benefit that much.

This product is just, what, a 9950X3D, but they also increased the cache on the other CCD, too...? The specific use for that very specific improvement has to be insanely niche.
 
StoleMyOwnCar said:
This product is just, what, a 9950X3D, but they also increased the cache on the other CCD, too...? The specific use for that very specific improvement has to be insanely niche.
Click to expand...
Well, it's not for me, anyway: I don't want to buy another CPU and motherboard, I was just saying it's another area where it would probably be useful, although as LukeTbk pointed out, clock rate's probably more important for this than cache.
 
Reviews before my wallet comes out of the dungeon, but I am excited to see that this is happening.

Either I get one of these new 9950x3d2's or grab a 9950x3d. I'm in need of more cores, and RAM.
Bump my 7800x3d to the 9800x3d and stuff a 9950xd variant in my pain box.
 
Somewhat related. A third of the twitter comments on this announcement were asking where the hell int8 fsr4 is for rdna2/rdna3. Maybe someday amd will figure it out.
 
sc5mu93 said:
Somewhat related. A third of the twitter comments on this announcement were asking where the hell int8 fsr4 is for rdna2/rdna3. Maybe someday amd will figure it out.
Click to expand...
I thought zen 6 (Prometheus) simd was going to include new types and a new matrix engine? You're talking about RDNA, I know but im thinking AMD won't have just one approach here.
 
1_rick said:
Well, it's not for me, anyway: I don't want to buy another CPU and motherboard, I was just saying it's another area where it would probably be useful, although as LukeTbk pointed out, clock rate's probably more important for this than cache.
Click to expand...
I guess it depends on how big your project is, too. For linking large projects like firefox or linux, it could help a lot?
 
erek said:
$1000 for this chip,

“Retailers are listing AMD’s Ryzen 9 9950X3D2 at roughly $1,000, hinting at premium pricing for the dual‑CCD 3D V‑Cache Zen 5 flagship.”

https://x.com/HotHardware/status/2041154097483600329
Click to expand...
Very very very tiny upgrade. Competitive, but not amazing performance against Intel. Huge power draw. Astronomical price. AMD really really needs to stop.... and figure things out.

In short, they could have taken my advice and ditched the 9950X3D, and kept that price with this as the total replacement.

Hoping the consumers clobber this one with zero sales. Wake up AMD. Weird.
 
cjcox said:
Very very very tiny upgrade. Competitive, but not amazing performance against Intel. Huge power draw. Astronomical price. AMD really really needs to stop.... and figure things out.

In short, they could have taken my advice and ditched the 9950X3D, and kept that price with this as the total replacement.

Hoping the consumers clobber this one with zero sales. Wake up AMD. Weird.
Click to expand...
Why? People said they wanted it. They'll probably make a few hundred and see how those sell, and unless they sell out, we won't see another X3D2.
 
  • Like
Reactions: erek
like this
1_rick said:
Why? People said they wanted it. They'll probably make a few hundred and see how those sell, and unless they sell out, we won't see another X3D2.
Click to expand...
People "wanted it" because they want to use the cores they paid for. So, it was like fixing a mistake. Which is why I say make this the new normal for multi-CCD with 3D cache from now on. (and keep the price)
 
cjcox said:
People "wanted it" because they want to use the cores they paid for. So, it was like fixing a mistake. Which is why I say make this the new normal for multi-CCD with 3D cache from now on. (and keep the price)
Click to expand...
That makes no sense. They are able to use the cores they paid for. It requires some software shenanigans to make it work optimally but it still works. V-cache on both CCDs is an expensive addition that won't benefit gaming and will only benefit a few specific type of tasks. Giving people a cheaper alternative for various use cases is not a bad thing. It's not like the X3D is a dual v-cache chip with one of the v-cache disabled, which would be the only way an X3D2 part replacing an X3D part at the same price would make sense. AMD has generally been pretty consistent- adding v-cache is a $100-$150 premium over no v-cache. With the 9950X3D retailing around $700, $1000 is completely within expectations for the halo 9950X3D2 launch price, with a high probability of it dropping to the $850 range a few months after launch.
 
Tsumi said:
That makes no sense. They are able to use the cores they paid for. It requires some software shenanigans to make it work optimally but it still works. V-cache on both CCDs is an expensive addition that won't benefit gaming and will only benefit a few specific type of tasks. Giving people a cheaper alternative for various use cases is not a bad thing. It's not like the X3D is a dual v-cache chip with one of the v-cache disabled, which would be the only way an X3D2 part replacing an X3D part at the same price would make sense. AMD has generally been pretty consistent- adding v-cache is a $100-$150 premium over no v-cache. With the 9950X3D retailing around $700, $1000 is completely within expectations for the halo 9950X3D2 launch price, with a high probability of it dropping to the $850 range a few months after launch.
Click to expand...
I was taking that from the "wanted it" perspective. If "wanted it" means, as implied, they wanted to not have to do workarounds to make efficient use of their high end CPU by parking/limiting capacity, etc.. Sure, there is the other perspective, that is, that (we'll say) very few users with multi-CCD CPUs were "ok" with the limitations, and then, sure, for them it may very well "make no sense". I was arguing from the "they wanted it" perspective.

I would never say this "makes no sense".
 
cjcox said:
I was taking that from the "wanted it" perspective. If "wanted it" means, as implied, they wanted to not have to do workarounds to make efficient use of their high end CPU by parking/limiting capacity, etc.. Sure, there is the other perspective, that is, that (we'll say) very few users with multi-CCD CPUs were "ok" with the limitations, and then, sure, for them it may very well "make no sense". I was arguing from the "they wanted it" perspective.

I would never say this "makes no sense".
Click to expand...
Core parking and limiting capacity was only ever really needed to prevent stuttering in games, which don't really use more than 8 cores anyways in the vast majority of them. Software workarounds are generally good enough now that such measures aren't needed most of the time. "They wanted it"..... did they really? Or were they under the illusion an X3D2 part would magically solve all the stuttering problems, which I doubt it will without the aforementioned software workarounds.

I will absolutely stick by the "makes no sense" for AMD to sell an X3D2 part for the same price as an X3D part. Should they have launched an X3D2 part from the beginning? Sure, you can make the argument that it "makes no sense" for that to not be an offering from the beginning. But AMD would be stupid to not offer an X3D and an X3D2 part at different prices.
 
Segmentation for segmentations sake now it feels like - this should absolutely replace the 9950x3d in the lineup IMO
 
kram182 said:
Segmentation for segmentations sake now it feels like
Click to expand...

I don't think it's that.

AMD has been maintaining a fairly quick release cadence in recent years. There was only 1.5 years between the 7800X3D and 9800X3D for example. Problem is that it's already been about that much time since the 9800X3D was released, but Zen 6 is delayed until 2027. So the 9850X3D egg hatched instead. Same story with the 9950X3D2 really. It's just filling a "shiny new thing" gap until Zen 6. IMO, if you've already waited this long, and what you already have isn't completely terrible, then just wait until Zen 6.
 
  • Like
Reactions: kac77
like this
I still wanna see where it shines for specific use cases and where it does should be absolutely marvelous. Ashes of the singularity is gonna get smashed.:rolleyes:
 
learners permit said:
I still wanna see where it shines for specific use cases and where it does should be absolutely marvelous. Ashes of the singularity is gonna get smashed.:rolleyes:
Click to expand...
My guess is it will be basically what the 9950x3D should have been... stupidly fast in every situation, and zero core parking needed.
 
I keep imagining that massive cache of both CCD's loaded with shaders ready for the render pipeline in games.
 
  • Like
Reactions: erek
like this
Having a 9950X3D now, I'll just wait for next generation and the rumored 12-core CCDs, but part of me wishes this was out when I built my system a year ago. Then again, you mostly want games on a single CCD, so I have a feeling the other CCD would rarely get used. I actually like having the non-X3D CCD for tasks that do not require the large cache outside gaming as it is slightly faster in clock speeds.
 
$899 price announced in US.

Promise 5-10% performance increase in professional programs, nothing really on games.

Makes basically no sense to spend $~200-25@ more than 9950x3d. If some suckers start selling their 9950x3ds on the cheap to upgrade the. Highly recommend hopping on it. I picked one up and it’s an absolute beast that runs surprisingly efficient for the amount of performance.
 
As a 7950X3D user, I skipped the generational uplift on the 9950X3D figuring it wasn't worth the price, all while hearing the rumors for ages about a chip with symmetrical 3D cache. Even if the hardware itself has been good, I've never been a fan of the kludge necessary to deal with the asymmetric CCD "one with extra cache, one with higher speed" AMD CPUs. Without using an on-die governor like Intel did when the P vs E chip situation arose (and even that took a WIndows Thread Director update) , it has leaned very heavily on the software side of things which especially early on was often more of a "Leave it up to the Windows Game Bar; if it flags something as a game, then prefer the cache cores CCD first!". The software side has had a bit of a rocky road with the asymmetric configuration as well as with latency between both CCDs (apparently as linked in this thread, the Zen5 chips had it worse than Zen4 until recently?), and I often thought that perhaps AMD would come up with a more comprehensive software/driver package with not just AGESA (I'm still waiting for OpenSIL! isn't this year- or perhaps next - supposed to be the one it debuts?) updates on the firmware side but a profiler of sorts that included both many default usage situations for given titles and the ability for users to create their own ; not unlike how SLI/CrossfireX worked back in the day. Unfortunately, nothing of the sort came to pass. Luckly, Linux's scheduler was already a bit more compatible and/or seemed to work better in these situations than Windows' did especially early on.

For the high end chips with multiple CCDs (with previous fixes to the latency between them), symmetrical 3D cache means no longer needing to park cores, depend on Windows Game Bar, or otherwise beat the scheduler into compliance with hopefully minimal side effects, microstutter, and overhead. That alone seems like a good reason for this to replace the original 9950X3D and/or simply become the next iteration of what "X3D" should be on high end chips, but the additional cache is likely to have benefits in both gaming and other workloads that benefit from the increased cache and it spread equally across multiple CCDs. This is a time for AMD to insulate themselves against Intel making a "return to form" with their upcoming architecture, by keeping X3D2 chips relatively inexpensive upgrades and eventually to move beyond that cache layout as they move into future generations.

yinzerniner said:
$899 price announced in US.

Promise 5-10% performance increase in professional programs, nothing really on games.

Makes basically no sense to spend $~200-25@ more than 9950x3d. If some suckers start selling their 9950x3ds on the cheap to upgrade the. Highly recommend hopping on it. I picked one up and it’s an absolute beast that runs surprisingly efficient for the amount of performance.
Click to expand...
Higher price than I'd like. The professional utility uplift makes sense; wonder if it will be even larger with certain software. I'm curious to see how things evolve with gaming down the line, particularly those that can make use of the extra cache. I can understand there not being much major gaming difference at least at first. There are relatively few games that both make use of the extra cache AND can take advantage of more than 8c/16t, but I am curious to see how it works on some of those that do, as well as more subtle benefits from the symmetrical core arrangements.

I can understand how those already on a 9000 series CPU will be hard pressed to see the value in an ugprade unless something really comes out of the woodwork, but those with 7000 series and compatible boards or those looking to build something new on AM5 now, it may be a good option. I hope that AMD starts discounting them a bit to make them more alluring immediately, while ensuring that future platforms (both Ryzen and Threadripper) offer a symmetrical cache option.
 
RanceJustice said:
...For the high end chips with multiple CCDs (with previous fixes to the latency between them), symmetrical 3D cache means no longer needing to park cores, depend on Windows Game Bar, or otherwise beat the scheduler into compliance with hopefully minimal side effects, microstutter, and overhead....
Click to expand...
I'm going to bet that you are going to be disappointed.

I very well could be wrong, yet I fully expect the second CCD to be parked for most games due to the latency penalty involved if it wasn't - same as the 9950X3D. With the few games that don't park the second CCD, I'd expect marginal improvements over the 9950X3D.
 
You must log in or register to reply here.
Back
Top