AMD Ryzen 7 9800X3D Linux Performance: Zen 5 With 3D V-Cache

[phoronix]

Phoronix: AMD Ryzen 7 9800X3D Linux Performance: Zen 5 With 3D V-Cache

Ahead of tomorrow's availability of the AMD Ryzen 7 9800X3D processor as the first Zen 5 CPU released with 3D V-Cache, today the review embargo lifts. Here is a look at how this 8-core / 16-thread Zen 5 CPU with 64MB of 3D V-Cache is performing under Ubuntu Linux compared to a variety of other Intel Core and AMD Ryzen desktop processors.

https://www.phoronix.com/review/amd-ryzen-7-9800x3d-linux

[Smurphy]

Well, looking at the last playthrough video from Squadron42 - when the time is right I may upgrade to that CPU our gaming rigs

[uid313]

I would like to see some benchmarks of the integrated GPU (if it has one) and see how it fares at gaming without a dedicated graphics card.

[WonkoTheSaneUK]

I would like to see some benchmarks of the integrated GPU (if it has one) and see how it fares at gaming without a dedicated graphics card.

Techpowerup covered that in their review of the chip. Here's the relevant page (Windows 11 23H2 used, obvs):-

https://www.techpowerup.com/review/amd-ryzen-7-9800x3d/22.html

[skeevy420]

Well, looking at the last playthrough video from Squadron42 - when the time is right I may upgrade to that CPU our gaming rigs

Keep in mind that while the 9800X3D is faster than the 7800X3D, that speed does come at the expense of using almost 50% more electricity. It should also be noted that the 7800X3D can be configured to use less electricity and potentially use its boosted speeds longer than it will with the default UEFI settings. These results don't necessarily show the 7800X3D at its best...Set the CPU to 65w eco mode and set Curve Optimizer to -20. Use Windows and Ryzen Master to fine tune it further from there...

I'd assume that the 9800X3D can be reigned in to use less power and still be equal to or better than the 7800X3D.

[Anux]

Nice, better than I expected. It seems to do especially well in video compression.

[rodja]

Thanks for the review.
I have mixed feeling about this chip. As skeevy420 pointed out this chip uses substantially more power than 7800X3D, so I am not sure how fair is is to say that it "delivered 1.34x the performance of 7800X3D". Also "Compared to the Ryzen 7 9700X, the Ryzen 7 9800X3D offered 9% better performance." This also is not fair comparison, it should be compared against the 9700X at cTDP 105W. Here the the gain is just shy of 4%.
Yes, there are a handful of productivity tasks where the extra cache helps (not just games), however, they are not that many and the 9700X3D is substantially more expensive than 9700X (and the 12 core 9900X too!).

I really appreciate your in depth reviews, if I have one suggestion perhaps it would be to split the geomean in two parts: one for single core tasks and one for multi-core tasks. I think the many single core tasks skew the geomean in favor of CPUs with lower cores. Another issue may be tasks where the execution time is dominated by the memory bandwidth more than the CPU speed.

[NekoTrix]

single core tasks skew the geomean in favor of CPUs with lower cores

The higher core count AMD and Intel SKUs have consistently demonstrated higher single core performance due to their comparatively higher single core frequency (and better binning), so that is not true.

[rodja]

The higher core count AMD and Intel SKUs have consistently demonstrated higher single core performance due to their comparatively higher single core frequency (and better binning), so that is not true.

I disagree with you. I don't think the extra 200mhz (5.5Gghz 9700x vs 5.7 9950x) are that meaningful. It's a 3% difference and it's probably gonna look bad when power-normalized since you are working in the inefficient area of the power-frequency curve.

[sobrus]

Something is odd with Zen3 results. 5950x is often beaten by lower clocked, lower binned 8 core 5800x? In video encoding?

Maybe I should disable one CCD and see if my rig is faster without it

Also, again, IMHO far too many tests artificially favor AVX-512 enabled parts, which isn't really representative for general use case workloads.
I'm not talking about video encoding and likes, this is (in my opinion) a common scenario. But who taxes their rigs with things like simdjson?