Ryzen 7950X3D with One CCD Disabled

[Radxge]

Unsurprisingly, the 7800X3D does not do much at 4K as most games are GPU bound.
For 4K folks like me, 7800X3D looks too expensive compared to 7700 (~$125 and free cooler and Jedi game) for ~5% average additional performance (which is mostly driven by one game, Borderlands 3).

But those enjoying 1080p with a 4090, what a delight!

 

[Upgrayedd]

Unsurprisingly, the 7800X3D does not do much at 4K as most games are GPU bound.
For 4K folks like me, 7800X3D looks too expensive compared to 7700 (~$125 and free cooler and Jedi game) for ~5% average additional performance (which is mostly driven by one game, Borderlands 3).

But those enjoying 1080p with a 4090, what a delight!

Ehh. Native 4K maybe not so much of a difference but still shows in Spiderman RT 4K even. With DLSS/FSR being popular at 4K your CPU definitely matters if you want HFR with RT. If you want HFR with RT then you need more than a 7700 but HFR with no RT then sure a 7700 is mostly fine.

What I'd like to see are some GPU usage graphs mixed in with low res CPU RT benchmarks. Spiderman RT 4K still seems a little CPU bottlenecked. CP2077 RT 720p graphs show the 7700 at 102fps while the 7950X3D shows up with 163fps. Running CP2077 in 4K RT with DLSS Balanced should put the render resolution at around 1080p right? Which is where the 7700 will bottleneck and the 7950X3D would be 30fps faster at 133fps or you could use Ultra Performance DLSS and see 163fps with a 7950X3D but with the regular 7700 nothing will happen except the picture will look worse at the same fps since its a bottleneck.
The 1% RT minimums for CP2077 were 78fps for the 7700 and 112fps for 7950X3D.

I'm curious if the 4090 is at +90% usage in CP2077 720p RT. CPUs have had a rough time recently with RT and a GPU bottleneck is a much smoother problem than a CPU bottleneck.

 

[W1zzard]

but still shows in Spiderman RT 4K even

That's one of the reasons I added RT this round

What I'd like to see are some GPU usage graphs mixed in with low res CPU RT benchmarks

These should pretty much match the FPS, assuming the GPU usage sensor is linear, which it might not be

 

[BoboOOZ]

Odd decision indeed to delay 7800X3D, but now with simulated results really excited to see how the real 7800X ends up performing.

Not odd at all, AMD are upselling impatient gamers with much money to the 7950x3d instead of the 7800x3d. If they released them in the same time, very few people would buy the 16 core, or it seems that it's selling like hot cakes right now.

 

[AusWolf]

I thought this was going to be a total win, like the 5800X3D was compared to the 5800X, but apparently, it's not. Compared to the 7700X, some games gain from the extra cache, while others lose from the clock speed deficiency. Very interesting.

Eagerly waiting for the real 7800X3D review, though.

 

[Bensam123]

Just want to say given how weird benchmarking these particular processors are, the 7900X3D and 7950X3D, this website has probably done the best job out of all the tech reviewers out there including Tubing. I would say somethings that are missing that would've been nice is seeing a simulated 7900X3D (which will probably be terrible), W10 vs W11 even if a couple tests, as well as a partially bloated windows install.

The problems with the 900/950X3D seems to stem almost completely from the thread scheduler, so when there are multiple programs vying for CPU time they're going to start running into thread contention and performance losses. It doesn't need to be a super bloated Windows install, but a lot of gamers run multiple programs on their computer while they're playing games and it doesn't need to be the main benchmarking method. However installing maybe 8 or so programs and having them run all the time is a good place to start, like Discord, Twitch window with a active stream/chat, chrome tabs open, LED software, motherboard software, steam/b.net... there are a bunch of options.

Keep up the good work!

 

[yeeeeman]

they delayed 7800x3d cause people would probably that one instead of the 7950x3d, so they will give the 7950x3d a bit of alone time to sell.

 

[alexcheveau]

I registered here only to congrat the guy who did this review, amazing job

 

[Naito]

This kind of content is one of the many reasons I find TPU so awesome and keep coming back after so many years! Thanks @W1zzard ! Appreciate the time and effort you put into this!

 

[Tsukiyomi91]

real 7800X3D vs 5800X3D. that would be interesting.

 

[W1zzard]

I registered here only to congrat the guy who did this review, amazing job

This kind of content is one of the many reasons I find TPU so awesome and keep coming back after so many years! Thanks @W1zzard ! Appreciate the time and effort you put into this!

Thank you, I spent the whole month of January just retesting CPUs, GPUs and SSDs, your feedback is greatly appreciated

 

[Gasaraki]

Odd decision indeed to delay 7800X3D, but now with simulated results really excited to see how the real 7800X ends up performing.

It's not odd. This processor will cannibalize the sales of the 7950X3D because the performance will not be that different. Have the people buy the more expensive processor first.

 

[Tek-Check]

This processor will cannibalize the sales of the 7950X3D because the performance will not be that different

I don't think AMD is naive about it and I am sure they must have thought about wafer allocation.

7950X3D is mostly a collectionist halo product for Ego, performance, gaming, bragging rights, fps junkies, professionals, etc. They should sell two or three hundred thousand units, at least. There are always such buyers.
7800X3D is a mainstream SKU that will sell a few million units.

 

[jayjr1105]

Those power consumption/performance per watt numbers are just bat shit crazy.

 

[medi01]

Shocked to be shocked by power efficiency of the same CPU for the second time.

 

[rusTORK]

Hm... sometime i think about one thing:

Why it's impossible to do next thing - in CPU with two CCDs (7950X3D in this case) use 8 cores of CCD with 3D V-Cache for fast multicore performance and 1 core of regular CCD (somewhere is the middle of CCD or closest to memory controller\fabric) for just single core performance? Windows can't or don't know how to use it?

We already have a product that is possible to configure (shut down cores), but for some reason it's not working this way or there are no proper tools?

 

[mouacyk]

7800x3d vs 5800x3d ~ 20% uplift @1080p

4.85GHz vs 4.35GHz ~ 11.5%, so about 8.5% in IPC uplift. This is better than what Intel gets in gen-over-gen IPC increase

 

[A Computer Guy]

Another question I had was how does the 7950x CCD's compare to the individual 7950x3D non-3D CCD?
Is the 7950x3D non-3D CCD first or second hand quality compared to the 2 provided in 7950x?

I just wanted to say thanks @W1zzard for all the work it took to make this review.
Perhaps on the Test Setup page you can clearly indicate which CCD has the 3d cache. (I'm re-reading it because I don't know at first glance when looking at the charts.)
Also perhaps if you do CCD testing like this in the future instead of saying in the chart "CCD# disabled" say "CCD# only" which I think makes it more clear, anyway just an idea. Otherwise the reader has to do a bit of on the fly gymnastics in the brain "CCD0 disabled = CCD1" distracting from interpreting the content.

 

[luches]

Guess i will stick with my 5800x3D since im playing on a LGC2 which can do 4k120hz max and 5% uplift ain't worth switching to a new platform. Although im pretty astonished that it has even better efficiency than 5800X3D. 44W while gaming is insane !

 

[QuietBob]

Another superb article from our resident W1zzard. A big thank you!
The 7800X3D is shaping up to be the best gaming CPU of the generation. The power consumption figures and efficiency of the single CCD are downright amazing!

I measured the power consumption of my 5800X3D with -30 CO on all cores at 1080p maxed out. The average across eleven games was 60 W.
With the games tested here at the same resolution the simulated 7800X3D uses 15% less power than the 5800X3D while being 22% faster

 

[pressing on]

As I see it, both the 7950X3D and 7900X3D use a combination of BIOS updates, chipset drivers and a Windows driver that are geared to two separate blocks of cores, one with 3D V-Cache and one without. Conveniently with the 7950X3D/7900X3D the V-Cache and ordinary cores are on physically separate CCDs. The 7800X3D as a single eight core CCD with 3D V-Cache on all eight cores does not fit into this system. It could be made to fit but this would seem to mean V-Cache on less than eight cores, maybe V-Cache on six cores (like the 7900X3D) with two ordinary cores on the single CCD. This 6 + 2 configuration has echoes of Apple and Intel's Performance and Efficient core configurations. It would probably make the 7800X3D less of a general purpose CPU but this is unlikely to matter to its target market where gaming performance would be the main priority.

 

[Minus Infinity]

v-cache makes more sense for the 7800 as you would buy this for gaming. But the main thing is how weak it performs in productivity. Take away encryption and emulation and the performance drop is too much. The power drop is nice though, but I would buy a 13700 non K still as gaming performance was already fine for Zen 4 and Raptor Lake and 13700 is far better for productivity.

 

[Hxx]

I hope microcenter has plenty of these in stock on release day because i know what im buying as my next gaming chip

 

[A Computer Guy]

In the UEFI/BIOS is there an option to turn off the 7950x3D's cache?

 

[Hxx]

Unsurprisingly, the 7800X3D does not do much at 4K as most games are GPU bound.
For 4K folks like me, 7800X3D looks too expensive compared to 7700 (~$125 and free cooler and Jedi game) for ~5% average additional performance (which is mostly driven by one game, Borderlands 3).

But those enjoying 1080p with a 4090, what a delight!

1440p shows pretty good gains which is probably the most significant as that’s where most high refresh displays are at