• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Zen 5/8950X etc... rumors, news, benchmarks....

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

EarthDog

Gulper Nozzle Co-Owner
Joined
Dec 15, 2008
Location
Cbus!
I'll start!

Comparing the numbers that RedGamingTech (RGT) gives for the Ryzen 9 8950X, we see between a 12% and 21% increase in single-threaded Geekbench 6, while the multi-threaded benchmark gain is larger, between 18% and 40%. The variance here comes down to the range of available benchmark results in the Geekbench database; if we use the numbers on the official Geekbench CPU charts, it's 16% single-thread and 25% multi-thread.
More in the article....
 
If they can really top it out, I may consider updating my build and giving my other kiddo the existing one.
 

Mysticial is the writer of y-cruncher. Based on a recent patch from AMD to GCC, some architectural features of Zen 5 could be deduced.

Doubling of FPU execution from Zen 4 is massive. This might be the first time in the Zen era that AMD will have a "better" FPU than Intel when considering all products. Zen 4 is already strong in that area but it doesn't quite keep up with the two unit Intel AVX-512 implementations (Skylake-X, higher server/workstation offerings). Doubling that will pass Intel. The other possible improvements to shuffles and stores will help the data be in the right place at the right time. Moving data to feed this will be a challenge! I hope they release X3D versions sooner than later.

Within foreseeable future, there is no way out of Intel's regression to AVX2 in the consumer space so this gap will be massive. Intel's good AVX-512 implementations are out of reach of consumers.
 
Ouch, those potential price tags are hefty! Upgrading might not be worth it if they come true, especially if your current setup is still running well.
 
The core of Zen 5 is brand new, pretty much nothing left untouched. I expect massive performance uplifts. Since it's basically on the same process (slightly better), I expect clocks to be lower and/or power to increase but the increase in IPC should still allow for a large performance improvement and improved efficiency.
 
$1750/$2500 🤷🏻‍♂️

EDIT: no, wait, $1999/$2999 would be more in line with Nvidia's usual prices...

Don't you put that evil on me, Ricky Bobby!

My youngest son is asking for his own computer so he can play games with his brother (on their own computer that I built for them about a year ago). If these all end up being worthwhile I'll look to do a new build for myself with the new chips and 5*** series Nvidia and give them my current.
 

"Core for core Zen5 is >40% faster than Zen4 in SPEC"

I'm not familiar with SPEC other than it is an industry standard benchmark. While it is standardised quite strictly it is hard to relate to user tasks.

Also note the comparison given is core for core, not clock for clock which is the more interesting one for me. IPC vs clock gains.
 
SpecViewperf is a graphical benchmark.

The CPU one looks like it was last updated in 2017: https://spec.org/cpu2017/
The list price is $1000 so not surprising it isn't popular in the enthusiast space.

Some results for 7950X vs 5950X vs 3950X and 12900k (ST linked, MT next page). Each AMD generation gets faster than the last as you would expect. Harder to tell how much (on average?)

Edit: I typed up the ST results and got the following average improvements.
7950X vs 5950X: 23% faster int, 25% faster fp
7950X vs 12900k: 16% faster int, 8% faster fp
5950X vs 3950X: 35% faster int, 28% faster fp

Have to say I wasn't expecting this scaling. Zen 2 and 3 have very close max turbo clocks (on paper) and I didn't feel it had that much architectural improvement, perhaps of the ball park 10% or so, with much of its real world improvement from the 8 core CCX. But this is ST so that doesn't matter. Zen 4 has a much higher max turbo but I felt it had more architectural gain, without owning one myself. Maybe in all these cases it isn't necessarily hitting the max boost, even ST, so power efficiency still kicks in at some point.
 
Last edited:
I went from roughly 11 mins and change to roughly 5 mins and change in SuperPi 32M with a 3600XT to a 5900X. Single core only of course.
 

"Core for core Zen5 is >40% faster than Zen4 in SPEC"

I'm not familiar with SPEC other than it is an industry standard benchmark. While it is standardised quite strictly it is hard to relate to user tasks.

Also note the comparison given is core for core, not clock for clock which is the more interesting one for me. IPC vs clock gains.

SPEC is mainly used by server people, but it's been found to be a pretty good benchmark for measuring performance per clock when you run the 1T rate configuration of the test.

You can download the 2020 benchmark - https://gwpg.spec.org/benchmarks/benchmark/specviewperf-2020-v3-1/
For me, it's the same as it used to be with the Blender benchmark. Barely anyone heard about it (out of specific users), and AMD was using it for comparisons vs Intel. Until we see more benchmarks or actual real-world results, then it says nothing.

@mackerel already pointed out that it is SPEC and not SpecViewperf, but SPEC and Blender are very different. SPEC is a whole suite of benchmarks that together give you a final score in the end. It actually uses Blender as one of the benchmarks in the rate test. It also has compiler benchmarks, path finding, fluid dynamics, video encoding, and many more. It's not a perfect test and you can definitely manipulate it a bit with compiler settings, but to an honest reviewer, it is a great tool. Anandtech had been using it to measure performance and performance per clock for the last few generations and the results closely matched what others found when measuring IPC across their own custom suite of benchmarks (gaming, cinebench, office, browser, etc.).

SpecViewperf is a graphical benchmark.

The CPU one looks like it was last updated in 2017: https://spec.org/cpu2017/
The list price is $1000 so not surprising it isn't popular in the enthusiast space.

Some results for 7950X vs 5950X vs 3950X and 12900k (ST linked, MT next page). Each AMD generation gets faster than the last as you would expect. Harder to tell how much (on average?)

Edit: I typed up the ST results and got the following average improvements.
7950X vs 5950X: 23% faster int, 25% faster fp
7950X vs 12900k: 16% faster int, 8% faster fp
5950X vs 3950X: 35% faster int, 28% faster fp

Have to say I wasn't expecting this scaling. Zen 2 and 3 have very close max turbo clocks (on paper) and I didn't feel it had that much architectural improvement, perhaps of the ball park 10% or so, with much of its real world improvement from the 8 core CCX. But this is ST so that doesn't matter. Zen 4 has a much higher max turbo but I felt it had more architectural gain, without owning one myself. Maybe in all these cases it isn't necessarily hitting the max boost, even ST, so power efficiency still kicks in at some point.

Zen 3 had a 19% average PPC (performance per clock) increase over Zen 2. It was also able to hit and sustain it's 1T max turbo clocks much better than Zen 2, which would only hit it's max clocks for a very short time at best. So Zen 3 had a 19% PPC increase and a significant effective max clock increase as well which resulted in the results you see at Anandtech. That, combined with the switch from 4 to 8 core CCX design, is what caused Zen 3 to be so much faster at gaming than Zen 2.

Edit: Zen 4 PPC over Zen 3 was only ~12% on average but it had significantly higher clock speeds.
 
Last edited:
Fresh new rumor on Zen 5 IPC improvement over Zen 4.


Fortnite 7%
Passmark 11%
R23 1T 20%
7-Zip 28%
V-Ray CPU 33%
Metro Exodus 38%
Dolphin Bench 71%
WPrime 86%
Avg 3?%.

Then followed up with:


April fool. The Zen 5 IPC figures aren't too far from reality. The reality is even higher than this. Please wait till June for official figures.


So, according to the rumor, the first "leak" for IPC increase was somewhat made up but close to reality, only the reality is even higher than shown. I'm not sure on individual tests, but others are reporting that the IPC increase is as high or higher (on average) than the original, "April Fool's" tweet. Fmax reportedly has also increased but that doesn't necessarily translate to effective clocks.
 
Back