• 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

In my opinion, there will always be a place for the asymmetric v-cache design becauae there are people who don't use workstation software that benefit from v-cache on their gaming computers, so they would rather save the money. For games that do utilize more than 8 cores, I bet they will benefit far more from additional cores on a single CCD than v-cache on the second CCD.
 
Personally, I, like most, believe the asymmetric design was a bad design caused by technological limitation of the times. If all you need is a 9800X3D, get it and save some money. Again, because the performance difference is tiny, I'd nuke the "mistake" designs and "look" like a winner for doing the "right thing" when the technological limitation was overcome. But... that's me. Of course, as mentioned, this could strictly be temporal "posturing" knowing that the "mistake" is fixed with 10000 series.
 
This is too little too late. Zen 6 is supposed to have 12 cores per CCU and is less than a year out. Finally releasing a "flagship" product this late in the cycle is just a money grab. The 7950x3d should have been this part from the get go and here we are 3+ years later still waiting for it to drop.
 
So what? A bunch of people asked for it, there's nothing wrong with them deciding to give their customers what they wanted.

We asked for it 3 years ago, I'm not going to bother upgrading my AM5 release 7950X at this point in 9000 series lifecycle.
 
We asked for it 3 years ago, I'm not going to bother upgrading my AM5 release 7950X at this point in 9000 series lifecycle.
The target market is the kind of person who upgrades their 4090 to a 5090, probably.
 
We asked for it 3 years ago, I'm not going to bother upgrading my AM5 release 7950X at this point in 9000 series lifecycle.
If you're gaming, you'll see a pretty major difference. Otherwise, probably not worth it.
 
I might bite but I want to see test results first. I built a 7800X3D/X670 in 2024 and I would jump on this if worth it this year. I would definitely swap the motherboard for a X870.
 
Personally, I, like most, believe the asymmetric design was a bad design caused by technological limitation of the times. If all you need is a 9800X3D, get it and save some money. Again, because the performance difference is tiny, I'd nuke the "mistake" designs and "look" like a winner for doing the "right thing" when the technological limitation was overcome. But... that's me. Of course, as mentioned, this could strictly be temporal "posturing" knowing that the "mistake" is fixed with 10000 series.
There never was any technological limitation in putting the v-cache in both CCDs. For the 7xxx, there was an incentive to NOT do so because both CCDs would be limited in clock speed compared to the non v-cache version, so a hypothetical 7950X3D2 would lose out to the 7950X and 7950X3D by 10-20% in the majority of workstation applications and not be any faster in gaming. AMD knew they would get shit on if they released a 7950X3D2 because it would have been an objectively worse processor than the 7950X3D for 99% of the market. The 9xxx series solves this problem due to the rearranged v-cache, allowing the X3D CCDs to clock almost as high as the standard CCDs. There will be minimal performance penalties on workstation applications that aren't v-cache sensitive while providing a minor boost in applications that are. Now that there no longer is a supply shortage of X3D CCDs, they may as well use the extra chips to make a halo product purely for milking some extra money out of enthusiasts and keeping the attention on AMD. There was no extra engineering or technological progress made in the creation of the 9950X3D2.
I might bite but I want to see test results first. I built a 7800X3D/X670 in 2024 and I would jump on this if worth it this year. I would definitely swap the motherboard for a X870.
Unless you need GPU PCI-E 5.0 and/or Thunderbolt 4, X870 is a downgrade from X670.
 
The target market is the kind of person who upgrades their 4090 to a 5090, probably.

Which in the normal upgrade cycle would have had it released as the first 9000 series 3d part like nVidia does with their flagship GPUs. Instead it's likely the last before they release 6 core parts at the very end of its lifecycle. But here we are a full year after the release of the first 9000 series 3d parts. I definitely would have upgraded had it been released in March of 2025 when it should have been.

Unless you need GPU PCI-E 5.0 and/or Thunderbolt 4, X870 is a downgrade from X670.

Having an X670E ProArt, there aren't really any X870E boards to even upgrade to. Even Asus downgraded the ProArt in this cycle and it doesn't have proper Thunderbolt implementation.
 
Which in the normal upgrade cycle would have had it released as the first 9000 series 3d part like nVidia does with their flagship GPUs. Instead it's likely the last before they release 6 core parts at the very end of its lifecycle. But here we are a full year after the release of the first 9000 series 3d parts. I definitely would have upgraded had it been released in March of 2025 when it should have been.



Having an X670E ProArt, there aren't really any X870E boards to even upgrade to. Even Asus downgraded the ProArt in this cycle and it doesn't have proper Thunderbolt implementation.
Slightly different logistically as nVidia is trying to sell as many top tier chips at a premium as possible before the "defective" chips are released and potentially put downward price pressure on the top chips. Every 9950X3D2 sold is one less 9950X3D or 9800X3D that could be sold, and that matters during the initial launch and supply shortage that seems to occur with every launch these days.
 
Slightly different logistically as nVidia is trying to sell as many top tier chips at a premium as possible before the "defective" chips are released and potentially put downward price pressure on the top chips. Every 9950X3D2 sold is one less 9950X3D or 9800X3D that could be sold, and that matters during the initial launch and supply shortage that seems to occur with every launch these days.

Right, so even though we've been asking for it since the launch of the 7000 series 3d CPUs, they purposely didn't release it to sell neutered CPUs before finally releasing the dual CCD version. It's a pure money grab late in the 9000 series cycle instead of actually trying to release the best CPU from the get go.
 
Right, so even though we've been asking for it since the launch of the 7000 series 3d CPUs, they purposely didn't release it to sell neutered CPUs before finally releasing the dual CCD version. It's a pure money grab late in the 9000 series cycle instead of actually trying to release the best CPU from the get go.
Neutered is giving it the wrong representation; there was nothing neutered about the 9950X3D. The 9950X3D2 is physically a different hardware configuration. You're not wrong in calling it a cash grab, but this is nothing like disabling cores and then selling an unlocked version later.

Also, as I said previously, a hypothetical 7950X3D2 would have been an objectively worse processor for 99% of the market than the 7950X and 7950X3D while being more expensive than both. AMD would have shot themselves in the foot releasing a 7950X3D2, despite what some naysayers on here think.
 
Neutered is giving it the wrong representation; there was nothing neutered about the 9950X3D. The 9950X3D2 is physically a different hardware configuration. You're not wrong in calling it a cash grab, but this is nothing like disabling cores and then selling an unlocked version later.

Any mixing of the unlike CCDs is neutering the CPU. They're introducing scheduling complexity that doesn't need to exist. We saw the same issues with the launch of Intel 12th gen and AMD said, let's try it too just with cache differences! They had a chance to do it right from 9000 3d series launch and they didn't.

Also, as I said previously, a hypothetical 7950X3D2 would have been an objectively worse processor for 99% of the market than the 7950X and 7950X3D while being more expensive than both. AMD would have shot themselves in the foot releasing a 7950X3D2, despite what some naysayers on here think.

For those of us that want like for like cores across the CPU, even the lower clock speeds associated with the 7000 series 3d CCDs would have been a welcome upgrade along with the lower TDP.
 
Any mixing of the unlike CCDs is neutering the CPU. They're introducing scheduling complexity that doesn't need to exist. We saw the same issues with the launch of Intel 12th gen and AMD said, let's try it too just with cache differences! They had a chance to do it right from 9000 3d series launch and they didn't.



For those of us that want like for like cores across the CPU, even the lower clock speeds associated with the 7000 series 3d CCDs would have been a welcome upgrade along with the lower TDP.
So basically, you're mad at AMD for making smart business choices. A 7950X3D2 would have painted a giant bullseye on AMD's back for haters to jump on- your market wouldn't have been worth the negative publicity.

Lower TDP is a laughable excuse. You could have manually clocked the non v-cache CCD lower yourself, there wasn't a need to have v-cache for that purpose.
 
So basically, you're mad at AMD for making smart business choices. A 7950X3D2 would have painted a giant bullseye on AMD's back for haters to jump on- your market wouldn't have been worth the negative publicity.

No, I'm not mad about anything. I'm disappointed AMD chose to release an inferior product with scheduler issues which is why I still haven't upgraded from my launch 7950x. If their "smart" business decision worked for their bottom line, more power to them. But it has prevented me from doing my normal upgrade cycle. It didn't make sense to upgrade to the 9950x with 3d CPUs in the pipeline and then they released the same asymetric CCDs with the 9000 series instead of correcting their original blunder.

Lower TDP is a laughable excuse. You could have manually clocked the non v-cache CCD lower yourself, there wasn't a need to have v-cache for that purpose.

Yeah, it's great having to go in and manually set the PBO limits every time you have a bios update just to get the processor to run at the proper TDP when AMD can set them properly from the get go like they've done with every subsequent AM5 release.
 
No, I'm not mad about anything. I'm disappointed AMD chose to release an inferior product with scheduler issues which is why I still haven't upgraded from my launch 7950x. If their "smart" business decision worked for their bottom line, more power to them. But it has prevented me from doing my normal upgrade cycle. It didn't make sense to upgrade to the 9950x with 3d CPUs in the pipeline and then they released the same asymetric CCDs with the 9000 series instead of correcting their original blunder.



Yeah, it's great having to go in and manually set the PBO limits every time you have a bios update just to get the processor to run at the proper TDP when AMD can set them properly from the get go like they've done with every subsequent AM5 release.
You keep calling it a blunder and all these other negative connotations. It's not when it's the commercially successful strategy. It might not work for you and others like you, just like nVidia's $2000+ flagship GPUs are a major disappointment to those that enjoyed the $600 flagships of the past, but it doesn't matter how you feel. You feeling disappointed does not equal a mistake on AMD's part. The 9xxxX3D launch was successful and it took a few months for supply to match demand. A 9950X3D2 would have hindered supply.

Updating bios? Manually tweaking every time? How often are you updating your bios that this would actually be a major time sink? Saving profiles and recalling saved profiles across updates have been a feature for years now as well. You're literally making a mountain out of a molehill here.
 
You keep calling it a blunder and all these other negative connotations. It's not when it's the commercially successful strategy. It might not work for you and others like you, just like nVidia's $2000+ flagship GPUs are a major disappointment to those that enjoyed the $600 flagships of the past, but it doesn't matter how you feel. You feeling disappointed does not equal a mistake on AMD's part. The 9xxxX3D launch was successful and it took a few months for supply to match demand. A 9950X3D2 would have hindered supply.

Getting two disparate CCDs to function properly across all the apps that may be running on a system hasn't exactly been a walk in the park for users. Sure the launch was successful, that's because the 9800x3d was released 4 months prior to the 9950x3d as opposed to the 7950x3d being released 2 months before the 7800x3d. AMD saw that getting the 8 core single CCD part out first was the way to go and demand for the async pairing of CCDs wasn't nearly the same. They had a full 4 months to build up supply of a dual x3d CCD version and still went with an upgraded 7950x3d part.


Updating bios? Manually tweaking every time? How often are you updating your bios that this would actually be a major time sink? Saving profiles and recalling saved profiles across updates have been a feature for years now as well. You're literally making a mountain out of a molehill here.

I don't know about other boards, but there have been 35 bios releases since the initial that shipped with my X670E ProArt. I prefer staying relatively up to date with the non-beta versions so that when a critical one drops I'm not making a huge jump.

https://www.asus.com/us/motherboard...esk_bios?model2Name=ProArt-X670E-CREATOR-WIFI
 
Seems to be the norm lately. Hype up something and it turns out to be a dud.

No Ryzen 9950X3D2 for TechPowerUp, Gamers Nexus, or ComputerBase

by W1zzard Today, 12:06 Discuss (2 Comments)
If you've been refreshing our front page today looking for the Ryzen 9 9950X3D2 "Dual Edition" review, I don't have good news for you: we won't be publishing a review, because we had no access to a review sample. AMD typically reaches out to us, offering to be part of the reviews, but not this time. So, after waiting for a few days, I reached out to them, because I would have loved to test this really interesting SKU, but I was told no samples were available for TechPowerUp.

We're not alone. As VideoCardz noted in their review roundup, Gamers Nexus reacted strongly after being denied a sample, ComputerBase, one of the top publications, was also denied, just like many others that you know for their deep, methodical testing—exactly the kind of reviews that dig into cache behavior, inter-CCD latency, power scaling, and per-game CCD parking quirks, which on a part like this are arguably the whole story.“


View: https://youtu.be/p6O4LCmah98?si=SEHvrXOot0ffd1GI
 
https://www.phoronix.com/review/amd-ryzen-9950x3d2-linux

AMD Ryzen 9 9950X3D2 Benchmarks: The Best Desktop Performance For Linux Developers, Creators​

Written by Michael Larabel in Processors on 21 April 2026 at 09:00 AM EDT. Page 1 of 11. 23 Comments.

Today we can finally share performance benchmarks of the long-rumored AMD Ryzen 9 9950X3D2 Dual Edition processor. This new halo product for the Ryzen 9000 series desktop line-up offers captivating performance for developers frequently compiling code, creators, technical computing workloads for students or hobbyists or those not able to afford a Threadripper / EPYC type workstation, or similar heavy computing use. With the 16 cores / 32 threads and both CCDs having 3D V-Cache, the AMD Ryzen 9 9950X3D2 offers leading performance among current generation desktop processors.
Last month AMD announced the Ryzen 9 9950X3D2 that had been rumored in online communities for months for having both CCDs with 3D V-Cache compared to just one of the CCDs with the Ryzen 9 9950X3D. The Ryzen 9 9950X3D2 Dual Edition goes on sale tomorrow and is expected to retail for around $899 USD for this AM5 desktop CPU with a total cache size of 208MB.

In my testing the past two weeks with the AMD Ryzen 9 9950X3D2, it's been working out very well under Linux without any problems to note. With the current AM5 platform support very mature under Linux for a while now and the main Ryzen 9 9950X3D2 simply being more 3D V-Cache on both CCDs, there isn't much to worry about from the Linux support perspective if you are running a modern Linux distribution. One little caveat/pointer is that the AMD 3D V-Cache Optimizer Driver is still loading on the 9950X3D2 processor. That is the driver for CPUs with having just 3D V-Cache on some CCDs to select your frequency vs. cache preference for tasks under Linux. I've confirmed with AMD this driver has no actual affect on the Ryzen 9 9950X3D2 for any cache vs. frequency preference since both CCDs are the same.
 
  • Like
Reactions: erek
like this

No Ryzen 9950X3D2 for TechPowerUp, Gamers Nexus, or ComputerBase

by W1zzard Today, 12:06 Discuss (2 Comments)
If you've been refreshing our front page today looking for the Ryzen 9 9950X3D2 "Dual Edition" review, I don't have good news for you: we won't be publishing a review, because we had no access to a review sample. AMD typically reaches out to us, offering to be part of the reviews, but not this time. So, after waiting for a few days, I reached out to them, because I would have loved to test this really interesting SKU, but I was told no samples were available for TechPowerUp.

We're not alone. As VideoCardz noted in their review roundup, Gamers Nexus reacted strongly after being denied a sample, ComputerBase, one of the top publications, was also denied, just like many others that you know for their deep, methodical testing—exactly the kind of reviews that dig into cache behavior, inter-CCD latency, power scaling, and per-game CCD parking quirks, which on a part like this are arguably the whole story.“


View: https://youtu.be/p6O4LCmah98?si=SEHvrXOot0ffd1GI

They may not realize it but this is a back handed insult to the reviewers that DID get a sample. If tough reviewers were denied samples you could assume the reviewers that DID get samples are soft reviewers at best and AMD shills at worst.
 
They may not realize it but this is a back handed insult to the reviewers that DID get a sample. If tough reviewers were denied samples you could assume the reviewers that DID get samples are soft reviewers at best and AMD shills at worst.
I think it's more simple than that - they were looking for minimum viable coverage for a launch with the aim to avoid the inevitable collective yawn, much like the reception of the Intel KS chips. With Zen 6 around the corner, this will probably end up being a low volume niche part...
 
I think it's more simple than that - they were looking for minimum viable coverage for a launch with the aim to avoid the inevitable collective yawn, much like the reception of the Intel KS chips. With Zen 6 around the corner, this will probably end up being a low volume niche part...
Do you have any opinion on which sites got samples and the ones that did not?
 
Seems to be the norm lately. Hype up something and it turns out to be a dud.
maybe the 9950X3D2 is good at this,

"AMD 3D V-Cache Turns Ryzen Into a Surprise RAG AI Weapon, With An 88% Boost Over Non-X3D CPUs


AMD's 3D V-Cache CPUs deliver a huge boost versus the Non-X3D part in AI benchmarks, showcasing why they are best suited for RAG pipelines.

AMD 3D V-Cache Vs Non 3D V-Cache CPU Benchmarks in AI Showcase a Massive Uplift For RAG Pipelines​

We know that there are two ways to do AI: the first is LLM, which is currently the most popular model. LLMs are AI models that have been pre-trained on a large set of data and feature various parameter sizes. But LLMs' limitations can be seen when it needs to generate responses on data it wasn't trained on."

https://wccftech.com/amd-3d-v-cache...se-rag-ai-weapon-88-percent-boost-vs-non-x3d/
 
Last edited:
maybe the 9950X3D2 is good at this,

"AMD 3D V-Cache Turns Ryzen Into a Surprise RAG AI Weapon, With An 88% Boost Over Non-X3D CPUs


AMD's 3D V-Cache CPUs deliver a huge boost versus the Non-X3D part in AI benchmarks, showcasing why they are best suited for RAG pipelines.

AMD 3D V-Cache Vs Non 3D V-Cache CPU Benchmarks in AI Showcase a Massive Uplift For RAG Pipelines​

We know that there are two ways to do AI: the first is LLM, which is currently the most popular model. LLMs are AI models that have been pre-trained on a large set of data and feature various parameter sizes. But LLMs' limitations can be seen when it needs to generate responses on data it wasn't trained on."

https://wccftech.com/amd-3d-v-cache...se-rag-ai-weapon-88-percent-boost-vs-non-x3d/
That tracks...but how many people are running a RAG box with a good gpu and a consumer class cpu?

(admittedly i am literally doing this, but I am running 14b bf16 reasoning models on dual 3090s with nvlink and that's a rare combo to have)
 
Last edited:
I wonder if maybe some soft/firmware level issue still persist, maybe with the task scheduler and AMD wants to fix it before anyone finds it. The whole we dont have problems we dont admit to Corporate BS.

All kinda silly no matter the actual reasons, in the end, if they want these sites will just buy some silicone to test and get a random lottery ticket instead of a hand chosen review sample.

We will see what we see in the end, I suppose.
 
That tracks...but how many people are running a RAG box with a good gpu and a consumer class cpu?

(admittedly i am literally doing this, but I am running 14b bf16 reasoning models on dual 3090s with nvlink and that's a rare combo to have)
What's that for anyway?
 
Last edited:
Do you have any opinion on which sites got samples and the ones that did not?

I think, for the most part, the emphasis on the selected sites were those that were not primarily gaming sites. My site, TPU and GN, for example, have a heavy emphasis on gaming and did not get sampled. We didn't even get the launch memo until we asked about it. Yet, you see Phoronix on the list who will certainly not be doing much gaming - they'll be running workloads that actually show a tangible benefit of the additional cache. Any site that focuses on these chips from a pure gaming perspective will be disappointed to find that there's not really any lift to be had, just as AMD publicly stated long before they launched these.
 
HUB got one and are gaming heavy, derBauer can tend to be, looking at the number of flags in the small media list on videocard if it is representative, some world/language distribution of a limited amount of unit made available could have been a criteria, if PR teams are by region and get each X small amounts because AMD was not that interested in pushing it, as it is not a mainstream product for a mainstream audience that much, will look useless and overpriced for 99% of even new desktop users, 99.9% for PC users.
 
Last edited:
I think people thought this would blow every current AMD CPU out of the water and this is not the case.
 
HUB got one and are gaming heavy, derBauer can tend to be, looking at the number of flags in the media, some world/language distribution of a limited amount of unit made available could have been a criteria.
Perhaps pick one or three per major language/region....
 
AI workloads, FML, all we need is another thing to drive up the cost of DRAM.. LOL
 
What's that for anyway?
RAG?

Retrieval augmented generation.

If you have a smaller thinking model then the data it needs isn't parametrically embedded in the model weights.

It will try to generate an answer to a prompt it lacks information about.

RAG basically embeds information in something like a vectorDB of information and a small retrieval model runs on the cpu to get relevant context and feeds that to the model with your prompt so it can quickly see the relevant information and give better answers.

My case i can comfortably fit a 14b 16 bit model on my 2x 3090s vram and that leaves enough context for the retrieved data to go with the prompt and give better answers.

The RAG approach is memory bandwidth sensitive so adding more cache reduces how often the cpu needs to hit the ram.
 
AI workloads, FML, all we need is another thing to drive up the cost of DRAM.. LOL
RAG is ideal for smaller models. So this could help someone running local AI inference get better performance will less vram.

I don't see this fundamentally impacting the overall memory shortfall here, but it could help people playing with smaller model's locally.
 
RAG?

Retrieval augmented generation.

If you have a smaller thinking model then the data it needs isn't parametrically embedded in the model weights.

It will try to generate an answer to a prompt it lacks information about.

RAG basically embeds information in something like a vectorDB of information and a small retrieval model runs on the cpu to get relevant context and feeds that to the model with your prompt so it can quickly see the relevant information and give better answers.

My case i can comfortably fit a 14b 16 bit model on my 2x 3090s vram and that leaves enough context for the retrieved data to go with the prompt and give better answers.

The RAG approach is memory bandwidth sensitive so adding more cache reduces how often the cpu needs to hit the ram.
Right but....I meant; what do you actually use it for?
 
HUB got one and are gaming heavy, derBauer can tend to be, looking at the number of flags in the small media list on videocard if it is representative, some world/language distribution of a limited amount of unit made available could have been a criteria, if PR teams are by region and get each X small amounts because AMD was not that interested in pushing it, as it is not a mainstream product for a mainstream audience that much, will look useless and overpriced for 99% of even new desktop users, 99.9% for PC users.
sth got one,

https://www.servethehome.com/amd-ryzen-9-9950x3d2-dual-edition-review-going-a-bit-higher/
 
Back
Top