AMD Ryzen 9000X3D Series to Keep the Same 64 MB 3D V-Cache Capacity, Offer Overclocking

[AleksandarK]

AMD is preparing to release its next generation of high-performance CPUs, the Ryzen 9000X3D series, and rumors are circulating about potential increases in stacked L3 cache. However, a recent report from Wccftech suggests that the upcoming models will maintain the same 64 MB of additional 3D V-cache as their predecessors. The X3D moniker represents AMD's 3D V-Cache technology, which vertically stacks an extra L3 cache on top of one CPU chiplet. This design has proven particularly effective in enhancing gaming performance, leading AMD to market these processors as the "ultimate gaming" solutions. According to the latest information, the potential Ryzen 9 9950X3D would feature 16 Zen 5 cores with a total of 128 (64+64) MB L3 cache, while a Ryzen 9 9900X3D would offer 12 cores with the same cache capacity. The Ryzen 7 9800X3D is expected to provide 96 (32+64) MB of total L3 cache.

Regarding L2, the CPUs feature one MB of L2 cache per core. Perhaps the most exciting development for overclockers is the reported inclusion of full overclocking support in the new X3D series. This marks a significant evolution from the limited options available in previous generations, potentially allowing enthusiasts to push these gaming-focused chips to new heights of performance. While the release date for the Ryzen 9000X3D series remains unconfirmed, industry speculation suggests a launch window as early as September or October. This timing would coincide with the release of new X870 (E) chipset motherboards. PC enthusiasts would potentially wait to match the next-gen CPU and motherboards, so this should be a significant upgrade cycle for many.



View at TechPowerUp Main Site | Source

 

[oxrufiioxo]

Can't wait to read threads about people frying their chips lol.

 

[GerKNG]

i am pretty sure that VCore is not unlocked.

 

[JWNoctis]

There may not be much achievable in practice, unless they did something to improve the thermal penalty of 3D V-Cache and the thick IHS. I think they are supposed to be iterating the design of the cache chiplet this generation. Maybe they would.

That, and not repeating anything similar to the vSOC whoops of the Ryzen 7000 generation.

 

[Bwaze]

Yeah, overclocking. I bet they're not going to leave performance on the floor, so it will be purely academic. Anyone else remembers early Ryzen "Precision Boost Optimizer" video that showed overclocking you couldn't achieve later even on LN2, and they simply said "the numbers were just an example, not indicative of frequency processors could achieve"?

 

[alwayssts]

Can't wait to read threads about people frying their chips lol.

My *guess* is they will limit it to ~1.3-1.35v. Maybe even 1.2v.

What is it now? 1.1v?

That's 20% (give or take 4/5nm differences); enough to be note-worthy but not enough to fry anything. Probably ~170w TDP and that's conceivably IF you keep it within their temp limits (of perhaps 89c again?).

If I had to guess, enough to get to 5900mhz with a golden sample, but ain't no way they'll let you have 6ghz. /cynical realism

 

[Jism]

Can't wait to read threads about people frying their chips lol.

Hopefully, they have decoupled the additional cache from it's v-core voltage rail. That was the only reason why the previous generations could not be overclocked. Exceeding 1.35V would guarantee the cache to be toast.

 

[oxrufiioxo]

Hopefully, they have decoupled the additional cache from it's v-core voltage rail. That was the only reason why the previous generations could not be overclocked. Exceeding 1.35V would guarantee the cache to be toast.

My *guess* is they will limit it to ~1.3-1.35v.

What is it now? 1.1v?

That's 20%; enough to be note-worthy but not enough to fry anything. Probably ~170w TDP and that's conceivably IF you keep it within their temp limits (of perhaps 89c again?).

If I had to guess, enough to get to 5900mhz with a golden sample, but ain't no way they'll let you have 6ghz. /cynical realism

There are probably some samples that boost higher than others and might be able to clock 1-200mhz higher than others my guess is that's all some are going to get.

My 7950X3D vcache all cores around 5.3hz but my 7800X3D didn't break 4.9 same motherboard so if you can take a 9800X3D up 2-300mhz people will have less fomo

 

[Dimitriman]

Would anyone here pay additional 100$ on top of original price if the 9950 x3d had a 64mb cache on each ccd? ie. 16 x3d gaming cores instead of 8?

 

[7950X Italia]

There are probably some samples that boost higher than others and might be able to clock 1-200mhz higher than others my guess is that's all some are going to get.

My 7950X3D vcache all cores around 5.3hz but my 7800X3D didn't break 4.9 same motherboard so if you can take a 9800X3D up 2-300mhz people will have less fomo

I'm sorry for my english.
I wouldn't be so pessimistic. The 4nm is given +11% performance vs 5nm by TSMC (obviously with the same transistors) and Zen5 has a 16% increase in IPC (which means 16% lower frequencies for the same performance on Zen4). A Zen5 3D at @5GHz would already perform as well as a Zen4 3D at 4.3GHz, already running at 5.1GHz like Zen4 is all butter. In my opinion the expectation of at least +20% on Zen4 is the minimum of the minimum. What remains to be understood is how much more maximum MT frequency vs Zen4 and the silicon curve of the TSMC N4P. I have a 7950X... coupled with an AIO 360, CO -15 and 101MHz bus (I use MT, I don't play), I'm more than fine... about +1.5/2% vs def, but over 230W it's a nightmare... I got to 260W in manual OC... hung with the temp 95° (tamb 20°). The expectation for OC fun is that the 4nm TSMC can allow +100/200MHz at the same PPT with Zen4, that the silicon curve can perhaps allow 6GHz (psychological and therefore it makes sense to go custom vs AIO

 

[oxrufiioxo]

Would anyone here pay additional 100$ on top of original price if the 9950 x3d had a 64mb cache on each ccd? ie. 16 x3d gaming cores instead of 8?

I would. My 7950X3D is a headache with basically different cores on each CCD.

Once dialed in and setup with process lasso it's fine though but I'd pay 100 more not to deal with that shite lol.

 

[billeman]

If I read the article right, the 2 CCD versions would only have 64MB of L3 cache in total per CCD, and the 1 CCD version would have 96MB.

I find that mildly confusing, and wonder why that would be. Or maybe it's just a 'typo' ?

 

[dgianstefani]

I would. My 7950X3D is a headache with basically different cores on each CCD.

Once dialed in and setup with process lasso it's fine though but I'd pay 100 more not to deal with that shite lol.

You y wouldn't gain any advantage in games from vcache on both cores. Arguably worse.

The issue is chiplet to IO die to chiplet latency.

 

[oxrufiioxo]

You y wouldn't gain any advantage in games from vcache on both cores. Arguably worse.

The issue is chiplet to IO die to chiplet latency.

Gaming performance is awesome better than my 7800X3D once I jumped through 10 hoops...

My 5950X is always faster than my 5800X despite dual CCD so I figured assuming the same clocks a 7950X3D with dual ccd would just be more consistent.

 

[R0H1T]

Well think of it this way how would a car perform with 18" tyres at the front & then 20" at the back?

 

[phints]

I'm sure AMD found the sweet spot in Vcache and there is no reason to increase it further right now. After how mindblowing the performance per watt is on the 7800X3D (3x better than Intel?) it's obvious their engineers know what they are doing. Looking forward to the 9800X3D.

 

[kondamin]

is it the same 7nm sram?

 

[evernessince]

You y wouldn't gain any advantage in games from vcache on both cores. Arguably worse.

The issue is chiplet to IO die to chiplet latency.

Your two sentences are contradictory.

If the issue is latency between the CPU chiplets as you posit, having 3D cache over both CPU chiplets should reduce latency to a level at the very least equal to that of single CCD X3D chips. It would also be interesting to see how the additional cache impacts energy efficiency.

The gains of two X3D caches are likely to be much more minimal in general but I for one would be willing to pay extra for it if it means they have a 16 core chip that's top of the chart in gaming and energy efficiency and has excellent MT.

 

[dgianstefani]

Your two sentences are contradictory.

If the issue is latency between the CPU chiplets as you posit, having 3D cache over both CPU chiplets should reduce latency to a level at the very least equal to that of single CCD X3D chips. It would also be interesting to see how the additional cache impacts energy efficiency.

The gains of two X3D caches are likely to be much more minimal in general but I for one would be willing to pay extra for it if it means they have a 16 core chip that's top of the chart in gaming and energy efficiency and has excellent MT.

Nothing stopping you from limiting voltage and clocks to get what you're looking for, that's the only difference in "efficiency" between an X3D part and a non-X3D part. Similar to how K and non K series chips work from Intel.

Your assertion I wrote something "contradictory" implies a fundamental misunderstanding of the architecture, its drawbacks, and why X3D counters those drawbacks and therefore works well for games.

Two sets of 3DVCache increases latency between cores on different chiplets, not decreases it, lol.

 

[evernessince]

I'm sure AMD found the sweet spot in Vcache and there is no reason to increase it further right now. After how mindblowing the performance per watt is on the 7800X3D (3x better than Intel?) it's obvious their engineers know what they are doing. Looking forward to the 9800X3D.

Oh I'm sure a single X3D cache is the current sweet spot but that inherently implies there's more performance being left on the table. After all, calling it the sweet spot implies there's returns to be had above, just at diminishing levels.

Nothing stopping you from limiting voltage and clocks to get what you're looking for.

Your assertion I wrote something "contradictory" implies a fundamental misunderstanding of the architecture, its drawbacks, and why X3D works well for games.

Nonsense, limiting voltage and clocks are not going to have the same impact as a stack of X3D cache on all CPU dies.

 

[ADB1979]

I think they are supposed to be iterating the design of the cache chiplet this generation.

Yes this is, so I would expect a number of small changes that all add up to a better product overall as this is the third generation 3D V-Cache design and has a lower CPU clock speed penalty than the first or second generations and seemingly allows overclocking, I expect there to be a number of other small benefits they tell us about whilst releasing it.

 

[Ruru]

Just my 2 cents, but does anyone even actually manually overclock Ryzens? PBO magic with undervolting/CO works great already.

 

[dgianstefani]

Just my 2 cents, but does anyone even actually manually overclock Ryzens? PBO magic with undervolting/CO works great already.

PBO magic still results in fluctuating clocks, that can sometimes drop below what a static OC can achieve.

 

[evernessince]

Two sets of 3DVCache increases latency between cores on different chiplets, not decreases it, lol.

The vast majority of workloads are not bouncing threads between CCDs.

Where exactly is your source for such info anyways? There are zero consumer processors with two sets of 3D cache.

 

[dgianstefani]

The vast majority of workloads are not bouncing threads between CCDs.

So explain how gaming performance would be improved with a dual X3D chip then (hint, it isn't).

The gains of two X3D caches are likely to be much more minimal in general but I for one would be willing to pay extra for it if it means they have a 16 core chip that's top of the chart in gaming and energy efficiency and has excellent MT.

The gaming perf is the same, you want to be cache resident, and once you split into two caches, you don't get the gaming uplift - Amit Mehra, head of OC and IO/Core testing

Where exactly is your source for such info anyways? There are zero consumer processors with two sets of 3D cache.

Basic logic my dude. Also, FYI the dual CCD X3D chips were tested and dropped, because they did not offer improvements over single CCD or 1+1 of each type. Moving across chiplets has a latency cost since they do not talk directly to each other, they move through the IF on the IO die, which essentially hasn't changed since Zen 3, still clocking at 2000 MHz, moving to non native chiplet cache has another latency cost, plus you are now running at lower clocks dictated by the lower voltage cap of X3D, so latency is further increased.

If you want an all around CPU you go with the 7950X3D or the 14900K. Although the 7950X3D is slower than the 7800X3D in games due to that dual CCD compromise.

Your confusion comes from the fact that you assume X3D cache is the reason for efficiency increasing, from some inherent property, when the reality is that it's the voltage limitation that comes from the stacked cache forcing AMD to not release chips tuned way past their efficiency sweetspot to get within 10-20% of Intel single core performance.

Wishful thinking along with some active imagination is what causes people to long for the hypothetical performance/efficiency of a part that did not release to market (for a good reason).