One CPU To Rule Them All - Ryzen 7 9800X3D Review

[Plouffe]

With a small bump to boost clocks and the teensiest increase of L1 cache, how good could AMD's new Ryzen 7 9800X3D possibly be?

 

Buy an AMD Ryzen 7 9800X3D: https://geni.us/CNfv3S

Check out the rest of the AMD Desktop CPU Processors: https://geni.us/DPwe

Purchases made through some store links may provide some compensation to Linus Media Group.

 

 

[Senzelian]

I'm still running a 5800X and was hoping to upgrade to this new chip, but it seems I will just get a 5800X3D used on eBay. Hopefully, enough people will be upgrading so I can have a good deal on one.

[Agall]

8 minutes ago, Senzelian said:

I'm still running a 5800X and was hoping to upgrade to this new chip, but it seems I will just get a 5800X3D used on eBay. Hopefully, enough people will be upgrading so I can have a good deal on one.

If you can't afford to jump platforms, sure, but the 9800x3D is really the one true king. As soon as they announced that they were flipping the 3D v-cache SRAM and CCD stack, it was game over.

[ulookuglynoob]

1 hour ago, Plouffe said:

With a small bump to boost clocks and the teensiest increase of L1 cache, how good could AMD's new Ryzen 7 9800X3D possibly be?

 

Buy an AMD Ryzen 7 9800X3D: https://geni.us/CNfv3S

Check out the rest of the AMD Desktop CPU Processors: https://geni.us/DPwe

Purchases made through some store links may provide some compensation to Linus Media Group.

 

 

Can't wait to buy it. And sell my kidney . Idk how else ima afford it. Its like half the price of my whole setup. And the prices will go up fast.

 

[Kilrah]

17 minutes ago, Agall said:

9800x3D is really the one true king

* of low res high framerate gaming exclusively.

Doesn't seem very enticing for anything else...

[Haswellx86]

Transistor count:

 

7700x - 6,570 million

9700x - 8,315 million

 

7800x3d - 11,270 million

9800x3d - 8,315 million

 

7950x - 13,140 million

9950x - 16,630 million

 

Zen 5 processors tend to have 26.54% more transistors (which is still contradictory to their power efficiency and performance uplift besides AVX 512, so maybe the architectural improvements, or the newer process node, or both might have improved the power efficiency while countering this increase in transistor count, but there might be inefficient use of more transistors, but AVX 512 performance improved drastically which maybe could be worth it, but besides that, other changes such as deeper branch prediction and all might be also where all these additional transistors might have been used, which unfortunately didn't actually result in that big of a performance improvement in gaming) over Zen 4, but interestingly the 9800x3d has 35.5% less transistors, and also what is worth noting that it is pretty much similar in transistor count to the 9700x, to which the 9800x3d is just higher binned with 3D-vcache, but so is the 7800x3d to the 7700x, but there we have a massive transistor count difference.

 

There seems to be seem huge efficiency of transistor usage, where the new 3D-vcache doesn't need that many transistors to construct.

 

Or the reporting is just flawed, in which case, I am not to blame.

[leadeater]

@Plouffe Same comment as I left over on FP, I believe the transistor counts in the tables are not correct.

 

Zen 4 CCD = 6.57b, Zen 5 CCD = 8.315b, VCache Die = 4.7b, IOD = 3.37b (Ryzen 7000 & 9000)

 

9800X3D = 16.385b

7800X3D = 14.64b

[Senzelian]

55 minutes ago, Agall said:

If you can't afford to jump platforms, sure, but the 9800x3D is really the one true king.

Affording it isn't the issue, but justifying the purchase is. Why spend money for no performance gain? I play at 4K...

[Haswellx86]

5 minutes ago, leadeater said:

@Plouffe Same comment as I left over on FP, I believe the transistor counts in the tables are not correct.

 

Zen 4 CCD = 6.57b, Zen 5 CCD = 8.315b, VCache Die = 4.7b, IOD = 3.37b (Ryzen 7000 & 9000)

 

9800X3D = 16.385b

7800X3D = 14.64b

huh? sources? Read my comment earlier.

[Agall]

9 minutes ago, Senzelian said:

Affording it isn't the issue, but justifying the purchase is. Why spend money for no performance gain? I play at 4K...

The whole 'CPU choice doesn't matter at 4K' is something I spent quite a bit of time debunking. Its more dependent on what games you play and not resolution.

 

If you play nothing but MMO and MMO-like multiplayer games that are negotiating with a server and a bunch of players, 3D v-cache alone is ~50% increase over its non 3D counterpart.

 

If you play nothing but AAA single player games, then sure, it matters quite a lot less since you're likely GPU limited, even with an overclocked 4090. You'd still get more performance out of a better CPU, but probably not enough to justify the cost.

 

48 minutes ago, Kilrah said:

* of low res high framerate gaming exclusively.

Doesn't seem very enticing for anything else...

Still, this fallacy of CPU choice not mattering for 4K.

 

Everyone focusing in on the academic type benchmarks because they're far easier to control variables. With all the people I know personally, they don't spend majority of their game time playing any of the games that GN, HWUB, LTT, etc benchmark, they're playing multiplayer games that negotiate with players and a server. 

 

As much as I wish places like GN, HWUB, LTT would provide realistic CPU benchmarking, its simply not possible within their standards of measurement. That's where you have to find smaller channels, usually focusing in on one specific game, with comparisons.

 

I still use my example of Warframe as a control. 4K max settings, endurance mission with 3 other players and effectively maximum enemy count for a reapeatable mission. The difference between CCD0 and CCD1 of a 7950x3D alone was 160 fps vs 240 fps, practically speaking the difference between a 7800x3D and R7 7700. this ~50% increase was repeatable in other games too like Diablo 4.

[leadeater]

10 minutes ago, Haswellx86 said:

huh? sources? Read my comment earlier.

Yea I read it but worth waiting before addressing other commenters. Those other than the 3D V-Cache dies are all official AMD numbers, the V-Cache die is an estimate by multiple tech analysts and info likely came from AMD itself. 

 

Edit:

@Haswellx86 You can quite easily see the data errors from your post and this video

 

7700X and 7800X3D use the same CCD so transistor count is identical, 6.57 + 4.7 = 11.27 

 

The information is just simply not correct for any of them 

[RejZoR]

Thanks for also testing at realistic image quality settings. I'm so annoyed with all the benchmarks showing only 1080p Ultra Low settings benchmarks and showing huge performance difference between 5800X3D and 9800X3D. I didn't buy 5800X3D to run games at Ultra Ugly mode. I bought it to max out my settings and have great FPS. This review finally shown my 5800X3D is still very much relevant and perfectly fine. Thanks for doing a proper review.

[Spotty]

51 minutes ago, Haswellx86 said:

Transistor count:

 

7700x - 6,570 million

9700x - 8,315 million

 

7800x3d - 11,270 million

9800x3d - 8,315 million

 

7950x - 13,140 million

9950x - 16,630 million

 

5 minutes ago, leadeater said:

@Plouffe Same comment as I left over on FP, I believe the transistor counts in the tables are not correct.

 

Zen 4 CCD = 6.57b, Zen 5 CCD = 8.315b, VCache Die = 4.7b, IOD = 3.37b (Ryzen 7000 & 9000)

 

9800X3D = 16.385b

7800X3D = 14.64b

It looks like the 7800X3D transistor count shown in the LTT video is including the 3Dvcache but the 9800X3D is not including the 3DVCache. Assuming the 3DVCache being 4.7B transistors, 11.2B - 4.7B = 6.5B transistors.

If you look at reviews for the 7800X3D they say 6.5B transistors, the same transistor count as the 7700X/7800X CCDs which would be correct if it's not including the 3DVCache.

https://www.windowscentral.com/hardware/computers-desktops/amd-ryzen-7-7800x3d-review

https://hwbusters.com/cpu/amd-ryzen-7-7800x3d-revisit-the-best-gaming-and-not-only-cpu/

 

I don't think the I/O Die is normally included in transistor count?

[leadeater]

1 minute ago, Spotty said:

I don't think the I/O Die is normally included in transistor count?

Some places do some don't, it's not that important so long as it's consistent in the comparison.

[UnawareBlaire]

Why did the non-x3d processors disappear from the benchmarks for the "real world" benchmarks? I would have liked to see how the performance differs there

[Haswellx86]

46 minutes ago, Spotty said:

It looks like the 7800X3D transistor count shown in the LTT video is including the 3Dvcache but the 9800X3D is not including the 3DVCache. Assuming the 3DVCache being 4.7B transistors, 11.2B - 4.7B = 6.5B transistors.

If you look at reviews for the 7800X3D they say 6.5B transistors, the same transistor count as the 7700X/7800X CCDs which would be correct if it's not including the 3DVCache.

https://www.windowscentral.com/hardware/computers-desktops/amd-ryzen-7-7800x3d-review

https://hwbusters.com/cpu/amd-ryzen-7-7800x3d-revisit-the-best-gaming-and-not-only-cpu/

 

I don't think the I/O Die is normally included in transistor count?

So you mean my data of 7800x3d and 7700x is accurate, right? It's just that the 9800x3D's transistor count is shown without taking in the 3D-vcache. That would make it around 13015 million transistors with 3D-vcache.

 

It was kind of dumb of me to realize that there is barely any transistor count increase between the current X3D and non-X3D. Cache is built using SRAM, but those need transistors to operate as well. The 7800x3D required a whopping 71% bump in transistor count for that 3D-vcache over the 7700x.

 

Who the hell wrote 8315 million on TechPowerUp ? Even other sources stated the same. It makes sense, the 9800X3D is so new, that there was barely any information about it. Somebody go and inform TechPowerUp to make the change.

 

@leadeater Maybe delete my original post if it's wrong.

[xnamkcor]

Can we stop changing the X axis scale multiple times when doing comparisons?

[pestermess]

Hopefully, they’ve addressed the CPU temperature issues I faced with the 5800X3D, especially with the 3D V-Cache heat dilemma. My EK 280 AIO just couldn’t keep up at times, and it definitely scared me a few times. Even with custom water cooling, I could barely keep temps under 80°C. That’s actually why I decided to skip the 7800X3D altogether.

[pestermess]

Has Linux released the 3D files for their CPU holders yet? I'm dying to print something similar for the CPUs I have lying around without clamshell cases.

[CyberneticTitan]

Not sure what is going on with YouTube or if this is intended but wtf is this thumbnail. Who made his eyes 3x larger

 

[pestermess]

19 minutes ago, CyberneticTitan said:

Not sure what is going on with YouTube or if this is intended but wtf is this thumbnail. Who made his eyes 3x larger

 

Clearly you don't understand the LOTR reference eh?

[CyberneticTitan]

8 minutes ago, pestermess said:

Clearly you don't understand the LOTR reference eh?

Guess not, haven't watched LOTR. Is this supposed to be Gollum? Could have done a far better job if that was the goal tbh.

[pestermess]

1 minute ago, CyberneticTitan said:

Guess not, haven't watched LOTR. Is this supposed to be Gollum? Could have done a far better job if that was the goal tbh.

Yeah, it's either Smeagol or Gollum. I haven’t watched LOTR since high school, which feels like ages ago now. You’re right, they could’ve done better, but honestly, LTT’s quality has been a bit hit or miss lately. We’re still recovering from the big drop in quality and all the drama that went down last year.

[ADo_9000]

Feedback to the thumbnail team.

I get a/b testing, but please never make something like this ever again.

It is actual nightmare fuel.

[Jaw709]

nice thanks for including the 5800x3d in the comparison metrics.. I don't think JayZ did, and it's important for tracking generational X3D uplift, also I'm biased. thanks