Thursday, March 26th 2026

AMD Announces Ryzen 9 9950X3D2 Dual Edition Desktop Processor with 192 MB L3 Cache

AMD today launched the Ryzen 9 9950X3D2 Dual Edition Socket AM5 desktop processor. This processor is positioned above the Ryzen 9 9950X3D, and is the much awaited implementation of 3D V-Cache memory on both CCDs (CPU complex dies) of the processor. The 16-core/32-thread processor is based on the Zen 5 microarchitecture, and features a total of 192 MB of shared L3 cache—that's 32 MB of on-die L3 cache plus 64 MB of 3D V-Cache that's contiguous with it, across both CCDs. This enables the maximum possible gaming performance, as worker threads of the game can be located on either of the two CCDs and benefit from large last-level cache. The base clock is 4.30 GHz, with a boost clock of 5.60 GHz—100 MHz lower boost than the 9950X3D, and the same clocks as the 9850X3D—no word if these are the same for each CCD.

The AMD Ryzen 9 9950X3D2 Dual Edition is positioned to be a 16-core Zen 5 Flagship part, and to ensure the chip has not just the most cache, but also the highest possible clock speeds in the series, and retain its overclocking capabilities, AMD expanded the TDP of the processor to 200 W—the highest ever for an AMD Ryzen part. AMD is offering gaming performance rivaling its latest Ryzen 7 9850X3D part, but with significantly improved productivity performance in applications with large cached data-sets, such as software development, AI training and inferencing, and video production. The company's first-party productivity benchmarks show the 9950X3D2 post significant single-digit percentage gains over the 9950X3D. AMD did not announce price, but said that the 9950X3D2 Dual Edition will be available from April 22, 2026.
Known specifications of the Ryzen 9 9950X3D2 Dual Edition follow.

Add your own comment

149 Comments on AMD Announces Ryzen 9 9950X3D2 Dual Edition Desktop Processor with 192 MB L3 Cache

#76
Sol_Badguy
ZendouA 14900K does not have a TDP of 125W. That is steady state after boost. When boosted it goes to 253W. Which enforces my inquiry towards that if 125W goes to 253W for intel, then 200W goes to what for AMD.
PBP is 125 W and MTP 253 W. But it goes over, not truly capped like the 9950X.
tpucdn.com/review/amd-ryzen-9-9950x/images/power-applications-compare-vs-14900k.png
tpucdn.com/review/amd-ryzen-9-9950x/images/power-applications-compare-vs-pbo.png
Direct comparison as the scale is the same.
PBO max for 9950X is more than 200 W, so the 9950X3D2 will not go as high but it will be halfway, I think it will tango with the 14900K in many cases.
ZendouIf it helps for arguments sake, lets say the air cooler would be a NHD15G2. I would hope that is sufficient to cool it at load and keep it sub 90, but I don't know.
If it can draw the heat quick enough perhaps, but if the thermal density is high coupled with a thick IHS that can be problematic.
Dr. DroYeah, but realistically both Zen 6 and Nova Lake are coming out within a year. Since I'm on Raptor Lake, I need a new motherboard either way.
I would say that this RAM crisis hurts AMD less than Intel.
Why?
Well the AM5 people have DDR5 already so for them it's just a CPU swap. Because they're not going to change sides even if Nova Lake wins, Zen6 will still be a significant upgrade over Zen5.

The people on other platforms however will choose the arch that wins, and they have to buy everything (except RAM for the ones that have DDR5), basically Raptor Lake (not all as some have DDR4) and Arrow Lake people.

And that's the thing, Intel will be pushing CUDIMM for Nova Lake in order to get the best results and maximize stability for high capacity and speed (the whole semi-HEDT, productivity angle).
Expensive shit that isn't really needed for Zen6, unless of course AMD is optimizing for that as well, but the ones already on AM5 don't have CUDIMM and they're not going to go after it in these market conditions.

So without super fast CUDIMM Nova Lake will likely be held back, some Arrow Lake owners have CUDIMM kits that can be repurposed but Raptor Lake owners don't.

That's the vibe I'm getting, if Nova Lake wins CUDIMM will play a part in the outcome, so those who want a beast of a platform will have to pay crazy money, if not they're getting only 90%.
In this regard I believe Zen6 might have a slight advantage because it doesn't need all of the stars to align in order to prevail.

For this aspect above AMD is not as pressured to delay, however the older MLID rumors were that AMD would delay in order to make sure it can reach the crazy clocks it wants for Zen6 (if not with N2P then with N2X).
Posted on Reply
#77
Dr. Dro
Sol_BadguyI would say that this RAM crisis hurts AMD less than Intel.
Why?
Well the AM5 people have DDR5 already so for them it's just a CPU swap. Because they're not going to change sides even if Nova Lake wins, Zen6 will still be a significant upgrade over Zen5.
Doesn't hurt neither AMD nor Intel, it sure hurts my wallet tho! :p

It'll suck building a new rig and having my Alder Lake-era kit in that build, but RAM is just like power supply... probably the most boring money you can spend on a build. The consolation prize is that RAM was cheap. Well, was. At least it's Hynix A and I know it can run at least 7600 ;)
Posted on Reply
#78
Rulumi
I thought this was cancelled at this point so interesting they are actually releasing it, announcement date seems specifically made to coincide with the Core Ultra 200 Plus releases date today. As others expressed though, I also think this release is kind of a bit too late at this point with Nova Lake and Zen 6 most likely coming next year alongside the very likely premiun pricing for this by AMD.
ZendouSo this is a non sequitur. Not sure what a 14900k has to do with if a 9950X3D2 can be Air cooled. The hedge at the end of we will see when reviews are out does not make it less absurd of a statement.

Great to know about your feelings towards AIOs and cores.

A 14900K does not have a TDP of 125W. That is steady state after boost. When boosted it goes to 253W. Which enforces my inquiry towards that if 125W goes to 253W for intel, then 200W goes to what for AMD.

If it helps for arguments sake, lets say the air cooler would be a NHD15G2. I would hope that is sufficient to cool it at load and keep it sub 90, but I don't know.
Intel and AMD boost behaviour is a bit different, PL2 in Intel works towards when there has been time and thermal space to go for PL2 for only an specific amount of time, default Tau Boost in K models is 56 seconds and 28 seconds for all other models, PL1 can be reached back in a lower time than setted up and "throttling" to PL1 quicker if a certain temperature threshold is reached.

AMD handles boost behaviour in a different and somewhat more agressive way, but in values what AMD uses are TDP and PPT, PPT being higher than TDP. In AMD the setted up PPT is the actual allowed maximun power to draw to the CPU socket, how much of it is used will vary in some other settings like TDC, EDC and the workload itself. But unlike Intel I think there's not a Tau Boost time window equivalent so as long as there's headroom for what the boost algorythm seeks and the workload requires and other settings and temperatures allow for the headroom, it may be able to sustain higher without an specified time window maximum limit to go back to the base power setting different like how Intel does by default.

The 9950X3D in specific is TDP: 170W and PPT: 230W. www.techpowerup.com/cpu-specs/ryzen-9-9950x3d.c3993

For most AM5 CPUs AMD uses the calculation of PPT being x1.35 of the TDP value. So if AMD keeps consistent with it on this 9950X3D2 Dual Edition that lists TDP as 200W, PPT will be 270W unless AMD specifically decides PPT to be a slightly lower calculation in this case.
Sol_BadguyAnd that's the thing, Intel will be pushing CUDIMM for Nova Lake in order to get the best results and maximize stability for high capacity and speed (the whole semi-HEDT, productivity angle).
Expensive shit that isn't really needed for Zen6, unless of course AMD is optimizing for that as well, but the ones already on AM5 don't have CUDIMM and they're not going to go after it in these market conditions.

So without super fast CUDIMM Nova Lake will likely be held back, some Arrow Lake owners have CUDIMM kits that can be repurposed but Raptor Lake owners don't.

That's the vibe I'm getting, if Nova Lake wins CUDIMM will play a part in the outcome, so those who want a beast of a platform will have to pay crazy money, if not they're getting only 90%.
In this regard I believe Zen6 might have a slight advantage because it doesn't need all of the stars to align in order to prevail.

For this aspect above AMD is not as pressured to delay, however the older MLID rumors were that AMD would delay in order to make sure it can reach the crazy clocks it wants for Zen6 (if not with N2P then with N2X).
Nova Lake compared to Arrow Lake will have FRED, AVX10.2 and APX. Zen 6 seems like only will have FRED, but even if APX can be good in general for the future just with recompiled binaries, I expect it to take some time to get widespread adoption at firsst.

With AVX10.2 full backwards compatibility with AVX-512 with all the umbral instructions included in AVX10.1 is possible though, so that's a part in which AMD won't have an advantage in anymore. In both cases I think they actually can be interesting and will be a far more interesting showing than Arrow Lake and Zen 5 against their predecessors.
Posted on Reply
#79
LastDudeALive
Sol_BadguyAnd that's the thing, Intel will be pushing CUDIMM for Nova Lake in order to get the best results and maximize stability for high capacity and speed (the whole semi-HEDT, productivity angle).
Expensive shit that isn't really needed for Zen6, unless of course AMD is optimizing for that as well, but the ones already on AM5 don't have CUDIMM and they're not going to go after it in these market conditions.

So without super fast CUDIMM Nova Lake will likely be held back, some Arrow Lake owners have CUDIMM kits that can be repurposed but Raptor Lake owners don't.
It's exactly the opposite, actually. AMD CPU's would really benefit from more memory bandwidth, it's just the chiplet design prevents it. The Infinity Fabric isn't stable enough to keep up with faster RAM, and so you have to drop down to a 1:2 setup to maintains stability, which hurts performance and you end up with barely any improvement. Of course the CPUs would do better with more memory bandwidth, but AMD's design is not capable of supporting it.

Intel isn't held back by the lack of CUDIMM (see the 270K matching the 9950X with equal DDR5-6000 RAM), but CUDIMM is a definite advantage they have over AMD. So of course they're be pushing it.

For ordinary users without CUDIMMs, Zen 6 and Nova Lake will probably be roughly equal, unless one side really screwed up the architecture. Both will have models with super-sized cache (X3D vs bLLC). But CUDIMMs will be a definite advantage for enthusiasts.
Posted on Reply
#80
Sol_Badguy
Dr. DroThe consolation prize is that RAM was cheap. Well, was. At least it's Hynix A and I know it can run at least 7600 ;)
Only 32 GB though! For the big boy 385K it's just a snack!
A Computer GuyI had suggested 9950X3DD but somehow Karen in marketing thought it to be inappropriate. :shadedshu:
Intel's got an Erin :laugh::
Posted on Reply
#81
Cheeseball
Not a Potato
Ah yes, the true replacement for my 7950X3D.

If its anything like the 9950X3D, it should have a total PPT of up to 250W (sensor reading from HWiNFO) if the default TDP is 200W. A 9950X (170W TDP) that I've tested in late 2024 goes up to 230W.
Posted on Reply
#82
Sol_Badguy
LastDudeALiveAMD CPU's would really benefit from more memory bandwidth, it's just the chiplet design prevents it. The Infinity Fabric isn't stable enough to keep up with faster RAM, and so you have to drop down to a 1:2 setup to maintains stability, which hurts performance and you end up with barely any improvement. Of course the CPUs would do better with more memory bandwidth, but AMD's design is not capable of supporting it.
True but all Zen6 needs to do is raise the 1:1 sweetspot to 6400, maybe 6800 (fingers crossed). Granted most users have 6000 max so yeah... bummer.
LastDudeALiveIntel isn't held back by the lack of CUDIMM (see the 270K matching the 9950X with equal DDR5-6000 RAM)
Arrow Lake gets by alright.
But Nova Lake still has tiles like Arrow Lake, and worse the flagship has two compute tiles so now the P-cores are separated, there is probably a way to fix the inter-tile latency without throwing super speed CUDIMM at it but the RAM will definitely play a crucial (pun intended) part.
Posted on Reply
#83
Panther_Seraphin
Dr. DroI only question why did it take so long. I've wanted this CPU for years and now that it's here, I pretty much don't want it anymore. It's at the tail end of Zen 5's life as the flagship architecture, we're 6 months at best from Nova Lake and Zen 6.
Zen 5 is going to be around a long time. Mid 2027 is when I bet we will see a Zen6 part on the shelf.

My guess as to why it took so long is 4 fold:
  • 98%+ of people will see 0 benefit over a 98xxX3D or 9950X3D depending on use cases.
  • 2 X3D dies is a LOT more expenisve in terms of time/resources vs normal dies. I am guessing the failure rate of these on the line just didnt make sense from a time/cost perspective as you introduce more and more failure points
  • No real competition from Intel to even consider it. 9950X/X3D covers the productivity market, 98xxX3D covers the Gamers and if your buying these to seriously make money there are far better EPYC options for those use cases.
  • Possibly Cannabalisng EPYC sales of the X parts where things like Core Licencing comes into play. Lot of SMBs/Cloud providers may be able to run things like SQL instances bare metal on these platforms for a fraction of the cost vs an EPYC platform.
Posted on Reply
#85
Zendou
Sol_BadguyI would say that this RAM crisis hurts AMD less than Intel.
Why?
Well the AM5 people have DDR5 already so for them it's just a CPU swap. Because they're not going to change sides even if Nova Lake wins, Zen6 will still be a significant upgrade over Zen5.
If Nova Lake is a significant improvement over a 9800X3D, then I would switch. The still ever present USB issues with Win 11 on top of other annoying quirks of the chipset is just a motherboard in addition to the CPU.
Posted on Reply
#86
AusWolf
200 W TDP... That's a 270 W power limit... On AMD! What has the world become!? :laugh:
Posted on Reply
#87
xorbe
We can go deeper ... X3D2 X2X
Posted on Reply
#88
A Computer Guy
AusWolf200 W TDP... That's a 270 W power limit... On AMD! What has the world become!? :laugh:
Pretty tame considering Threadripper Zen2/Zen3 280W TDP or Zen4/Zen5 380W TDP.
This new CPU could be amazing manually capping it at a lower PPT from 270W.
It may be interesting to see how to scales. I'm a bit tired of the newer more efficient CPU's being more power hungry in the race for minimal gains.
Posted on Reply
#89
AusWolf
A Computer GuyPretty tame considering Threadripper Zen2/Zen3 280W TDP or Zen4/Zen5 380W TDP.
This new CPU could be amazing manually capping it at a lower PPT from 270W. It may be interesting to see how to scales.
I would definitely do that if I ever bought one (I probably won't because Zen 6 is a lot more tempting with its 12-core CCD). I've got no means to dissipate so much heat.
Posted on Reply
#90
A Computer Guy
AusWolfI would definitely do that if I ever bought one (I probably won't because Zen 6 is a lot more tempting with its 12-core CCD). I've got no means to dissipate so much heat.
It's is definitely tempting to wait for that. Personally I don't really want to put much more investment into AM5 due to my dissatisfaction with it's platform limitations but it's CPU performance offerings have been pretty solid without taking the BIG.little approach as Intel. Pricing doesn't make Ryzen the catch all deal it once was now AMD appears to be sitting in a favorable and comfortable position while Intel "catch's up".

With Intel using TSMC now I think they are gong to leverage that and bring back some healthy competition but with AI industry knock on effects limiting the supply chain it may be a rough few years with both AMD and Intel bringing new CPU's to market potentially with PC sales slowing down while costs are so high. The fact AMD has at least one more CPU generation for AM5 is pretty favorable to AMD at this time as people might be more reluctant to upgrade their entire system.
Posted on Reply
#91
TumbleGeorge
What interesting times. Before, halo products (for enthusiasts) were called ThreadripperX and LGA 2066 Xeon. Today, that's what they call mainstream flagships... Somewhat as a halo product today, I only see Strix Halo and only in the maximum configuration with 128GB of memory and that's a light weight halo, in the sense of mobile platforms or small boxes.
Posted on Reply
#92
usiname
evernessinceFalse

"AMD is offering gaming performance rivaling its latest Ryzen 7 9850X3D part,"

The 9850X3D is 7.1% faster than the 9950X3D, thus it's a 7.1% gain.

www.techpowerup.com/review/amd-ryzen-7-9850x3d/26.html
I literally said this "in the best case scenario will match the single CCD X3D parts"
Its not the miracle the people expected, it was obvious that adding cache to the second CCD, won't make it faster than the single CCD, just like 9700X and 9950X are within 1-2% difference. Now will be the same with 3D cache on the both CCDs vs single one.
Posted on Reply
#93
danc
Prima.VeraWhat does this means, exactly?
Means license to priced it exorbitantly
Posted on Reply
#94
AusWolf
usinameI literally said this "in the best case scenario will match the single CCD X3D parts"
Its not the miracle the people expected, it was obvious that adding cache to the second CCD, won't make it faster than the single CCD, just like 9700X and 9950X are within 1-2% difference. Now will be the same with 3D cache on the both CCDs vs single one.
That's what I'm thinking, too. Ryzen is held back by inter-CCD latency, and not a lack of cache or raw speed.

Edit: That's why I swapped back to my trusty little 7800X3D after playing with a 7950X3D for a couple of months (which is also a great CPU, but it dies when a game uses more than 8 cores to load something...).
Posted on Reply
#95
AleXXX666
ZendouMy curiosity is how hot this thing is going to run. Can air even tame it...
exactly...
OnasiThat’s cool
THAT COULDN'T BE "COOL" LOL:D:D:D
bonehead123if they add a few more letters/numbers to the model number, they can harken back to days of
NO. If they "add more letters/numbers" to the model number, they will look same stupid as phones naming currently. "Ryzen AI PRO MAX ULTRA DUAL" SE (SE means Scam Edition)
wNotyarDThe X3D2 suffix should suffice to convey that. Why the "Dual Edition" then?
the whole X3D makes me LAUGH as "XD" smiley-emoji thing, LMFAO:rolleyes::D:D:D:D:D
Posted on Reply
#96
AVATARAT
usinameI literally said this "in the best case scenario will match the single CCD X3D parts"
Its not the miracle the people expected, it was obvious that adding cache to the second CCD, won't make it faster than the single CCD, just like 9700X and 9950X are within 1-2% difference. Now will be the same with 3D cache on the both CCDs vs single one.
That’s the interesting part: will the difference be as small as the percentage ratio between a single and a dual CCD, or will the results be worse due to the excessive amount of data passing through that single IF?
Posted on Reply
#97
Sol_Badguy
Following my previous posts #52 and #77 I have done a little exercise in estimating the 9950X3D2's power draw and performance:
Translating the colors:
Magenta is 17% increase to match the % difference between 170/230 W and 200/270 W.
Purple is halfway between stock and max PBO because increasing by 17% would have exceeded max PBO.
These will not be 100% accurate to real life but it's at least ballpark.
In reality stuff like 7Zip, AI NLP and Web Hosting might be less than my estimate.
Also some stuff like Blender, Chess, V-Ray could be slightly higher than my estimate.
Keep in mind this is using late 2024 data. Also the 285K is a stand-in for the 270K.
Regarding app avg:
www.techpowerup.com/review/amd-ryzen-7-9800x3d/23.html
If we have:
285K - 132 W
9950X - 135 W
7950X - 149 W
14900K - 180 W
Then 9850X3D2 would be 147 W based on my estimate above. So 8.9% higher than 9950X.
www.techpowerup.com/review/intel-core-ultra-7-270k-plus/23.html
If I use that percentage for the latest data, relative to the 9950X the 9950X3D2 would be 149 W in that chart.

/////

Now for performance.
First assumption is that going from 9950X3D to 9950X3D2 yields half the increase as when going from 9950X to 9950X3D.
But there is extra TDP involved:
www.techpowerup.com/review/amd-ryzen-9-9950x/27.html
Max PBO is 3% better than stock, however 200/270 W is sort of halfway between stock and max PBO therefore only 1.5% increase due to only increased TDP.
In games 720p only 1.2% better therefore only 0.6% increase.
Now using 2026 data:
www.techpowerup.com/review/intel-core-ultra-7-270k-plus/28.html
First app avg:
9950X3D - 105 %
9950X - 101.5 %
270K (without 200S Boost) - 100 %
Thus 3.45% better.
But only half of this when going from 9950X3D to 9950X3D2. Multiplied by 1.015 to account for increased TDP gives:
9950X3D2 - 106.8 %

Games:
720p:
9800X3D - 119.9 %
9950X3D - 113.1 %
9950X - 101.8 %
270K (without 200S Boost) - 100 %
This is 11.1 % better. Half of this multiplied by 1.006 gives:
9950X3D2 - 119.4 %

Enjoy!
Posted on Reply
#98
R-T-B
AVATARATDoes the 9950x3d2 model have a dual IF? We'll have to wait and see, because if it doesn't, that would be very interesting.
Just for info:
Doubtful, the 9950x standard does not, for comparison, and this is based on that.

Not gonna lie though: I really want this. Whether I can justify it is another matter.
Posted on Reply
#99
AusWolf
AVATARATDoes the 9950x3d2 model have a dual IF?
Definitely not. Not possible on AM5 AFAIK.
Posted on Reply
#100
igormp
AVATARATor will the results be worse due to the excessive amount of data passing through that single IF?
I don't think this will happen. Scheduling should still attempt to group threads belonging to the same process into the same, higher-clocking CCD, and it doesn't really matter if both have the extra cache now. Even on the regular 9950x the best CCD clocks ~300MHz higher than the 2nd CCD by default, I don't think this would change with this new version. The 9950x3D has a 200MHz delta, so at best I could see the gap being a bit smaller compared to the non-3D version.
Posted on Reply
Add your own comment
Nov 22nd, 2026 11:01 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts