AMD Ryzen 9 9950X3D & 9900X3D CPUs Benched, Leak Suggests Pleasing Single-core Performance Improvements

[T0@st]

AMD's Ryzen 9 9950X3D and Ryzen 9 9900X3D "Zen 5" processors are due for launch next month, but an exact date has not been announced. Currently, Team Red's 3D V-Cache-equipped Ryzen 9000 CPU series is composed of a single SKU: the popular eight-core Ryzen 7 9800X3D model. A new leak points to a possible imminent lineup expansion; the sixteen-core Ryzen 9 9950X3D and twelve-core Ryzen 9 9900X3D CPUs have finally popped up on Geekbench Browser. Both candidates seemed to be tested on the same PC platform; utilizing a GIGABYTE X670 AORUS ELITE AX motherboard and 32 GB (2 x 16 GB) of DDR5-4800 MT/s RAM. Notebookcheck

The Ryzen 9 9950X3D sample scored overall with 3363 (single-core) and 20,465 points (multi-core) in Geekbench 6.3. As expected, the Ryzen 9 9900X3D candidate's overall tallies came in slightly lower—it achieved 3274 (single-core) and 19,227 points (multi-core) overall. Press outlets were quick to compare these figures to prior generation outputs (refer to Wccftech's chart, below). On average, the incoming "Zen 5" parts surpass "Zen 4" equivalents by an average of 15%—in terms of single-core performance. Multi-core performance improvements are less significant; coming in at an average of 7%. The range-topping Ryzen 9 9950X3D CPU's multi-core performance score is "lower than expected," but closer-to-launch optimizations could rectify this matter. Geekbench results often do not reflect the true potential of tested silicon; gamers tend to dismiss or completely ignore these data points. Last month, an AMD executive revealed that the two upcoming X3D Granite Ridge desktop chips: "will provide similar overall gaming performance to the Ryzen 7 9800X3D." Speculative price points—of $699 (16-core) & $599 (12-core)—leaked online last week.



View at TechPowerUp Main Site | Source

 

[GhostRyder]

It will be intersting to see a few things with these chips:

1: If their multiplier will be unlocked like the 9800X3D (I assume so).
2: How high the CCD's turbo
3: If both CCD's have 3D Cache (I seem to remember AMD saying there was no point but I could be misremembering)
4: How the Scheduler handles things this go round (Especially if only one CCD has the cache)

I felt like the last time these chips were kinda an oddity in the lineup. Hopefully this round they are better!

 

[Chaitanya]

It will be intersting to see a few things with these chips:

1: If their multiplier will be unlocked like the 9800X3D (I assume so).
2: How high the CCD's turbo
3: If both CCD's have 3D Cache (I seem to remember AMD saying there was no point but I could be misremembering)
4: How the Scheduler handles things this go round (Especially if only one CCD has the cache)

I felt like the last time these chips were kinda an oddity in the lineup. Hopefully this round they are better!

Given this gen AMD has recomended using that Xbox Game bar for non-X3D parts and the drama we saw with Win11 dont get your hopes high about these X3D CPUs. Overall these will be a bit of mess compared to single CCD 9800X3D.

 

[Octavean]

3: If both CCD's have 3D Cache (I seem to remember AMD saying there was no point but I could be misremembering)

IIRC, AMD cited 3D cache as being less then cost effective for both CCD's.

 

[Makaveli]

3: If both CCD's have 3D Cache (I seem to remember AMD saying there was no point but I could be misremembering)

That rumor was put to rest awhile ago there is no v-cache on dual CCD's.

 

[Cheeseball]

Ehh, wasn't this expected? Both the 9900X3D and 9950X3D have higher boost clocks than the 9800X3D, so any single-thread/minimal multithread tasks should be faster due to clock speed alone.

This was also seen with the 7800X3D and the 7950X3D. If you disable/park CCD1 in the 7950X3D, you can observe the 3D cache CCD0 cores turbo to 5.4 GHz (and sometimes 5.6 GHz) without any other PBO/CO settings set. I believe this should be achievable with Process Lasso as well with the proper chipset drivers installed under 24H2.

 

[Quicks]

They should just increase the cache from 32mb to 64mb, probably wouldn't even need the 3D Cache part anymore. But hey why not add it to get a total of 128mb on a single die just to see if it makes much if a difference then.

 

[_roman_]

Cache is expensive. I think the biggest reason it was kept so small on mobile chips.

It annoys me already to have the urge or need to create for every game a windows 11 pro 24h2 amd gpu driver gaming profile. I do not want to bother now creating profiles for any software i use with "process lasso". That stuff should be done flawless by a decent paid operating system out of the box. It is a kernel problem - not a userspace problem.

 

[r.h.p]

They should just increase the cache from 32mb to 64mb, probably wouldn't even need the 3D Cache part anymore. But hey why not add it to get a total of 128mb on a single die just to see if it makes much if a difference then.

yeah that would be interesting

 

[Rightness_1]

AMD's release schedule a glacial at best.

 

[efikkan]

Well, this comparison without the versions without 3D V-Cache as a baseline is pretty uninteresting, even more in such a useless benchmark as Geekbench. We know from before that such chips have lower overall performance than their "regular" counterparts, and really only show significant benefits in a few edge cases. Strangely enough, people will pay more for a gimmicky inferior product, even though the benefits from large L3 caches will only decrease as workloads get more computationally intense over time, and the CPUs with more computational power will increase their lead.

Meanwhile there are some actually interesting news which should excite any (prospective) owner of Zen 5; AVX-512 improvements in FFmpeg. While it may be a small step, improvements such as these will continue to add value to products which supports it.

 

[Chrispy_]

Geekbench single-core is only measuring the peak performance of the core without vcache, making this the most pointless leak in the history of leaks.

We don't need a leak to know that a zen5 goes faster in Geekbench when it's clocked higher...