AMD Ryzen 9 9950X

[AnotherReader]

For AVX512 workloads it`s certainly memory bandwidth starved but it already can't use the entire DDR-6000 bandwidth(which is about 96 GB/s).

This is almost certainly why it barely had improvements over the 7950X in Y-Cruncher, it just doesn`t have the bandwidth to do more.

View attachment 358933

At least for y-cruncher, it's being held back by memory bandwidth. Compare the single threaded speedup over Zen 4 to the minimal multithreaded speedup.

 

[tfdsaf]

Nothing unexpected in the end. Zen 5 has extremely selective improvements and is largely castrated by the memory bandwidth

It is much better for server/workstation type processes, in fact anywhere from 10% to 25% faster, but yeah for desktop it is a disappointment. I think AMD will need to decouple their development for the server/workstation and desktop parts as it is clear that the two are very different and being on average 20% faster in server/workstation type apps is not going to translate well into desktop apps and games.

AMD also needs to work more closely with Microsoft to make sure their processors operate at optimal rate.

I'd say depending on what you need it can be both great and bad, if you are looking into gaming you are better off with a 7800x3d, if you are looking into value then the 7000 series are better, but if you are looking into more professional work, into workstation type work then it can be a great choice.

From my point of view it is a disappointment for your common desktop use and playing games, but from a server/workstation type view its a major advancement over the 7000 series, it is a way more forward looking architecture and it is going to age much better.

 

[1sanpedro1]

Really curious how this would do in rendering workloads vs. the 7950x3d at the same power. In the states now, and I'm itching to upgrade my 5950x, but now having trouble deciding what to go for. If the 9950x isn't 10-15% faster at the same power, I'd just enjoy the x3d and get better gaming to boot

 

[Tomgang]

Meh. Another dissapointment.

Power consumption is 40 watt lower than 7950X and temperature is also satisfactioned lower.

But performance boost is a lagbuster. Not impressed at all.

This makes lose hope that zen 5 3D will be anything special either. Unless amd has something secret about 3D like extremely fast 3D chase that could boost game fps.

Else cpu right is mess. Den 5 failure to impress and intels 13 14 gen cpu problems.

I will be staying with my 5950X and still fantasize about that 5959X3D we probably never get

 

[jmeistr]

Zen5 is so weird, i don't understand why AMD even released it. I'm saying this as a 7800X3D user, in case anyone was thinking "intel fanboy". Also AMD need to cut the crap with their misleading marketing slides.

 

[R0H1T]

Oh the meltdowns

Also AMD need to cut the crap with their misleading marketing slides.

True but I will go back to the same ol' tried & tested formula ~ don't need to upgrade every gen!

 

[oZ65]

It is much better for server/workstation type processes, in fact anywhere from 10% to 25% faster, but yeah for desktop it is a disappointment. I think AMD will need to decouple their development for the server/workstation and desktop parts as it is clear that the two are very different and being on average 20% faster in server/workstation type apps is not going to translate well into desktop apps and games.

AMD also needs to work more closely with Microsoft to make sure their processors operate at optimal rate.

I'd say depending on what you need it can be both great and bad, if you are looking into gaming you are better off with a 7800x3d, if you are looking into value then the 7000 series are better, but if you are looking into more professional work, into workstation type work then it can be a great choice.

From my point of view it is a disappointment for your common desktop use and playing games, but from a server/workstation type view its a major advancement over the 7000 series, it is a way more forward looking architecture and it is going to age much better.

As I wrote under the 9700X review I was looking for a new processor for Stockfish.

The improvements per watt are about 10% (we'll have to see with the next versions), but with the 7900 at 375€ in Europe it was an easy choice

 

[z1n0x]

Powerful core been constrained by interconnect and packaging. For how long is AMD going to milk the PCB copper routing? Its high time for higher performance and energy efficient solutions.

 

[R0H1T]

Like what light/fiber optics?

 

[AnotherReader]

Like what light/fiber optics?

Just doubling the number of GMI links would be enough. They do it for Zen 4c and EPYC {link to PDF). However, this would have to be coupled with a more capable memory controller.

INTERNAL INFINITY FABRIC INTERFACES connect the I/O die with each CPU die using 36 Gb/s Infinity Fabric links (This is known internally as the Global Memory Interface [GMI] and is labeled this way in many figures). In EPYC 9004 and 8004 Series processors with four or fewer CPU dies, two links connect to each CPU die for up to 72 Gb/s of connectivity

 

[Xuper]

Zen5 = Zen 4 + Fully native AVX-512 + some specific optimizations.

The AMD Ryzen 9 9950X and Ryzen 9 9900X Review: Flagship Zen 5 Soars - and Stalls

www.anandtech.com

Based on what we know about the Zen 5 architecture, as well as the Granite Ridge chip overall, the reasonable guess here is that we’re seeing AMD’s uncore – the memory controllers and attached Infinity Fabric – stretched to its limit. Since AMD reused the original Ryzen 7000 IOD for Ryzen 9000, the only additional resources available to feed the CPU cores is the slightly higher bandwidth of DDR5-5600 memory. All other cache and interconnect bottlenecks remain.

 

[Octavean]

I likely was never going to upgrade to a 9950X but there might have been some performance envy. I never thought I would have absolutely no interest though. The good news is that I can be really content with my RyZen 7950X / ASRock B650E PG Riptide WiFi based system for a few more years at least.

 

[Braegnok]

Great Review W1zzard, thanks for all the work, well done sir,..

I can OC my 7950X PBO MAX/ +150/ CO -20, with 8000CL34 memory.. and match, or beat the 9950X in all those review benches.

Not really enthusiastic to play retail lottery for 9950X with sufficiently good SP, IMC, core VIDs.

 

[R0H1T]

Just doubling the number of GMI links would be enough. They do it for Zen 4c and EPYC {link to PDF). However, this would have to be coupled with a more capable memory controller.

That may still not be enough, wonder how they'll work around this with zen6/128 bit (wide) DDR5 limits? Because it will be a major flop if they stick to just DDR5 6400 or even 7200(MT/s) with it.

 

[W1zzard]

@W1zzard Are the prices correct on the table in the first page? The 7900X3D costs 70$ more than the 7950X3D??? (the same values are on the table on the 9900X review)

View attachment 358934

Fixed. Since I don't have it in my test group, I didn't look up its price and never updated it

 

[DemonicRyzen666]

No one gave use a cpu-z screen shot of Zen 5 at all so far.

 

[W1zzard]

No one gave use a cpu-z screen shot of Zen 5 at all so far.

I gave you four, check the OC page on each review

 

[Spilly44]

Wondering why no 3950X included in the tests given that the 5950X and 7950X were in there?

 

[sergionography]

Multithread efficiency WORSE THAN PREVIOUS GEN?! I wonder how they managed that...

I thought about that too, but then the SMT review TPU did answered some of the questions. In theory, the better and more efficiently you utilize your core in single threading, the less room you have for SMT. If branch predictors have gotten so good that SMT is rendered almost useless. Maybe that explains intels approach to remove SMT altogether. Maybe they predicted we will eventually get there?

Another theory is that windows OS scheduler is failing miserably. This is more likely the case as the Linux reviews paint a completely different picture

 

[W1zzard]

Wondering why no 3950X included in the tests given that the 5950X and 7950X were in there?

AMD never provided one for review, buying it makes very little sense, and now that it's cheap it's just too old.

 

[rv8000]

I thought about that too, but then the SMT review TPU did answered some of the questions. In theory, the better and more efficiently you utilize your core in single threading, the less room you have for SMT. If branch predictors have gotten so good that SMT is rendered almost useless. Maybe that explains intels approach to remove SMT altogether. Maybe they predicted we will eventually get there?

Another theory is that windows OS scheduler is failing miserably. This is more likely the case as the Linux reviews paint a completely different picture

Anandtech seems to have mentioned there were core parking issues in a handful of scenarios. Wouldn’t be surprising tbh.

 

[KarymidoN]

hi @W1zzard, First of all thanks for the amazing review, you work is always the best.
I Would love to see memory scaling on ZEN5, how does it behave with higher frequencies (does it even support it?), better timings with lower clocks? or higher clocks with looser timings? i know its a lot of work, but it shure would be a nice piece of content, thank you.

 

[mb194dc]

There are some productivity workloads where it can make sense. For the vast majority of users and gaming at modern resolutions, no point in spending the money.

 

[Spilly44]

Thanks Wizard.Seems weird they gave you a 3900X Superb review and write up .

 

[nexxusty]

Was really hoping these chips could easily do 6600mhz 1:1 at least. I already have this speed achieved on my 7800x3D.