I’d just like to preface this with what I had in mind as my most desirable CPU on paper. I always envisioned a 12 Core, multi-Threaded CPU with a base frequency of 4.0GHz and a revolutionary amount of IPC. Something to the affect of Intel when they introduced Nehalem.
I just read the Ryzen 7000 news last night. And then promptly shit my pants. 16 multi-threaded cores. And a base frequency of 4.5GHz? AMD Really have outdone themselves. And they haven’t lost ANY momentum since they introduced Ryzen.
I am even more curious to see what Nova Lake will turn out to be like in response to what AMD has been doing. WOW! What an exciting time to witness emerging CPU tech.
Someone can correct me if I’m wrong, but I believe the ‘base’ frequencies are the maximum clock speed with with all cores boosting, and the 5.7GHz listed for the 7950X would be on a single core.
This is probably with PBO on to allow temperature/power to be the limiting factor. We’ll see what performance is like in prebuilts and the like with cheap motherboards.
If it is actually boosting to 5.7GHz on 16 cores at 170W, that would be insane.
Being able to get high clock speeds on demand is certainly of value. But it always comes at the cost of power draw and heat. TSMC 5nm will compensate for a lot, but in the end even with a smaller manufacturing node, the TDP still increases by ~70% (105 vs 170) which is a lot.
So it comes at a cost. Nevertheless it will use way less power than a Ryzen 5000 series at comparable clock speeds.
Yeah, Boost clock values are for single core usage like with any other CPU out there. And this drops the more cores are under load.
AVX and stuff is really intensive on the thermals. Check Intel CPUs with AVX512 running, they drop the clocks like crazy to stay within PPT limits. So it is dependent on your work load. But with stock power limits, it’s pretty much a guaranteed clock speed.
I wouldn’t read too much into clock speeds because IPC is still important. A race for highest clocks always reminds me of Pentium 4 days and this strategy didn’t end well.
I probably cut the peak clocks down to 5GHz on my upcoming Ryzen system, just to save like 30-40% power and corresponding thermals.
I’m definitely interested in them. We’ve been looking at getting a render server for my stuff at work, but if I can just upgrade 4 or 5 PCs to a 7950X and use distributed rendering, I see no reason to spend even more money on Threadripper Pro/EPYC.
I’m looking forward to getting a 5950X workstation when finance can justify, I am tempted to wait for the new though. Use will be video editing, CAD and 3D modelling.
Bear in mind I also wanted to get a 3950, but money has to go elsewhere
Will be curious on how well, does this socket holds up…
I wouldn’t be surprised, with this '2025 talk, being more regarding PCIe 5 → 6 move
AM4 had its PCIe 3 → 4 move, in a piecemealed fashion [CPU, GPU, Mainboards]
As the owner of a 3950X with 4.7GHz printed on the box which has never, and will never, hit anything close to that on a single core, I don’t really take claims like this at face value even when AMD explicitly advertise it.
You missed when Zen3 also didn’t hit its advertised clocks because it entirely exceeded them. The “4.9GHz” 5950X actually turbos to 5.05GHz, the “4.8GHz” 5900X to 4.95GHz, the “4.7GHz” 5800X to 4.825GHz, and the “4.6GHz” 5600X to 4.65GHz. PB2+XFR2 does a lot of work behind the scenes on Zen3, and AMD chose not to advertise it because they didn’t want to be giving anyone with merely minimum spec cooling any false impressions.
The fact that other people got what they paid for doesn’t make me feel less burnt. I am glad that AMD have stopped advertising entirely mythical “minimum specs” that many chips were never going to hit. It is unacceptable that they ever did this at all.
You could also be a victim of the power bug that affected some boards and early 2019 AGESA versions. I.E. not enough power is supplied to reach 4.7GHz despite the CPU attempting to allocate the core to that clock domain. Look into it. Nothing about the turbo bin ratings on Zen2 are mythical unless you’re operating outside the environment rating of the chip, or have some other blocker in the way of the chip reaching those clocks.
No Base is what you drop down to with bad cooling just before thermal throttle.
As long as you can cool it all core boost can be anything up to fmax depending on thermals and current/voltage limits for that workload.
(unless they put more hardlimits in like they did with 5800x3d…)
Nah, it was covered by the tech press after launch: the 4.7GHz was always just fictitious apart from in some extremely rare samples. Ngl if Nvidia or Intel pulled what AMD pulled putting numbers like this on the box we’d never hear the end of it, but AMD always gets a free pass whenever they pull this crap for some reason.
IIRC, the boost frequency printed on the label of Ryzen 3000 series boxes is the standard boost frequency under stock settings. There was another frequency i.e. up to +200MHz boost once PBO is enabled.
The ‘up to 200MHz’ overclock boost with PBO enabled was rarely possible in the launch verions of AGESA. It’s after a few revision, then a single core under extremely light workload (pretty much useless) can barely touch it. At least, that’s the case for early production silicons.
The standard boost frequency printed on the label is easily reached for most people. And 100% of the time after PBO+50MHz enabled. Some people on Reddit did complain their Ryzen 3000 can’t boost much for unknown reason.
AMD introduced a new boost algorithm in Ryzen 3000. And TSMC was used on their CPU for the first time. I believe everyone in the DIY/overclock market was basically their genius pig throughout the whole product lifecycle. Their AGESA releases were mess after mess, reflecting the challenge from first time TSMC N7 and the new algorithm. Eventually Ryzen 3000 stabilized after more than a year since launch as far as I remembered. With only occasional hiccup such as fTPM stuttering being the latest example.
I think ‘Ryzen 3000 experience’ is once in a lifetime kind. Nevertheless, I would never want to experience it again
Yes, directly after launch. Time has moved on, updates have been released, issues have been fixed. Anandtech even amended their boost clock figures from their 3950X review in a later review since the updates had solved the problem you’re describing.
