The AMD Ryzen 9 7900, Ryzen 7 7700, and Ryzen 5 5 7600 Review: Zen 4 at 65 Watts
anandtech.com
anandtech.com
It's probably reasonable as a relative comparison point to other recent AMD parts, but that's about it.
See e.g.: https://www.gamersnexus.net/guides/3525-amd-ryzen-tdp-explai...
(Maybe I'm wrong, I thought the TDP figure is the maximum sustained power that has to be dissipated by the cooling setup, the CPU can output more heat briefly but over a long period of time it'll average 65 W assuming it's fully utilized?)
Also, modern CPUs and GPUs pretty effectively throttle to the capacity of their cooling solution. I would not necessarily recommend it, but you could probably get by without any issue slapping a solution that reliably dissipates 65W of heat on any Ryzen, and have a stable computer. It would just happen to run up to its thermal limit regularly and clock down.
If you actually want a lower power/heat setup, you can undervolt and underclock the CPU and not worry about bouncing off the thermal limits.
Intel had products with TDP of 65w and 125w, so AMD "matched" that by making up a TDP formula to get the results they wanted. At the time, Intel's TDP was actually mostly accurate. It would turbo above that, but only for a few seconds before dropping to the TDP limit.
Then when AMD started winning, Intel had to get match them in return. If AMD's "125w TDP" parts could turbo forever, then why shouldn't Intel's? And so that's what they did. And just like that over a few generations TDP became entirely stupid & without a shred of meaning.
This is from the Conclusion page. Still pretty impressive.
(This is on Anandtech's first page).
Not making any judgements re: whether this chip or others are power-efficient, just making it clear that the power number in the title and body was chosen by a marketing team and has no other meaning.
https://www.anandtech.com/show/17601/intel-core-i9-13900k-an...
Intel Core i9 13900K has a TDP of 125/253 (base/turbo) but draws 334W (peak). 81W or 32% over max turbo TDP, 209W or 167% over base.
AMD Ryzen 9 5950X has a TDP of 105W but draws 142W (peak). 37W or 35% over.
AMD Ryzen 9 7950X has a TDP of 170W but draws 221W (peak). 51W or 30% over.
AMD Ryzen 9 7900 has a TDP of 65W but draws 90W (peak). 25W or 38% over.
So slightly higher % wise on some chips, but not outside the ballpark by any means. I don't think Intel is doing us any favors!
Hence why I specified "lightly threaded" not "single threaded" workloads. There's a whole huge market of workloads that sit at around 4-8 threads. They're called "video games" and it's what Intel has been bragging about being the best at for a while. The new crop of non-K's look prime to suffer the most in this workload relative to AMD's non-X competition.
But also again remember the 13600 non-K has much less L2 than the 13600k. It's not just a power/turbo restriction.
But we'll see for sure when the 13th gen non-K reviews happen. That Intel didn't sample these to reviewers isn't a great sign, though.
Reviewers should put this in BIG ALL-CAPITALS BOLD ITALICS at the top of their reviews, all the time. The TDP rating of any given CPU in most cases is completely unrelated to the actual performance of said CPU. This is most egregious in notebooks, where it's actually the PL1/PL2 limits that matter. In my notebook's case: it contains a Xeon W-11955M (TDP supposedly 45 W), but while gaming or rendering, the CPU regularly draws nearly twice that. In fact, in notebooks, the thermal setup frequently affects the performance of the CPU and GPU by more than ±20-30% compared to the baseline.
CPUs (and now GPUs, too) also include a variety of clock-scaling mechanisms like P-states, C-states, SpeedStep, Speed Shift (hardware P-states), etc that allow the CPU to sip a mere 0.2–0.5 W while doing nothing, and ramp up to 150+ W when demanded.
For accurate comparisons, reviewers should lock CPUs to a maximum power draw, and test CPUs at that fixed power draw.
This is done by some reviewers, but it also only accurately tests efficiency rather than performance (or potential for performance given suitable conditions).
This is also why you mostly see laptop performance reviews instead of laptop CPU reviews.
It is lying about its TDP a bit, because I see much more draw than the supposed 170W. But draw is still low enough that with a good cooler, it basically never exits that 5.8 GHz boost clock.
PopOS never felt so responsive before!
I compiled the same codebase on a home server using a quad core skylake cpu the other day. It took over 2 minutes to build the same code. If you haven’t upgraded for awhile, you’re in for a treat.
https://news.ycombinator.com/item?id=34310505
https://www.phoronix.com/review/ryzen-7600-7700-7900-linux
All Other tests: AMD Ryzen 7900/7700/7600 non-X Desktop CPU Review Roundup
https://videocardz.com/147338/amd-ryzen-7900-7700-7600-non-x...
It's amazing that I can get that kind of improvement without increasing power usage.
I often wondered what kind of performance we'd get out of new Ryzens if we limited the power, and now we're starting to find out :)
Still, seeing the desktop at 65-90W should be rather telling as those laptop parts will be "55W" but likely able to pull up to ~80W depending on system.
Almost doubling the TDP for a 25%-30% frequency increase isn't doing most any good (appeals mainly to benchmark obsessed folks with poor sense of scale). But it's the game that marketing people have to play---so hence the 105 watt X-variants.
When I upgrade to AM5 I'll go with 65W no doubt.
I resemble this remark. But I will not change :-)
People are still buying Intel so might as well try and convince them that AMD is just as "fast"
Since C2D days anything over 65W provided diminishing results for a regular user. But then you can't participate in the di^W gigaherz measuring party.
There's plenty of 4X / RTS / Simulation games where the CPU will be the bottleneck.
Disclosure: I own a 7600X, but strictly because I wanted a new desktop for the Christmas holidays, so I couldn't wait for either the 65W budget option or the X3D high-end (I do play Paradox games a fair bit). Hopefully I'll be able to sell it and upgrade without too much loss, once the full lineup is available.
But the people who (think that they) really need those extra few % performance also tend to be the ones willing to spend. Personally I think this is fine: the big spenders who buy the top $$$ and TDP parts basically subsidize a lot of the costs for the more price conscious who benefit from all the R&D and still get great performance.
You don't buy high end parts to be fiscally efficient, efficiency has always been the realm of mid to low end parts. This applies to nearly everything.
Another advantage there was that I would not have to build another computer to use two cheaper processors. But that only works as I don't think I am limited by any other factors like memory or disk bandwidth.
I did however limit the processor to a 65W TDP to reduce heat output and power usage, which only slowed down my very parallel programs about 10%, and didn't affect less parallel programs at all. (The duplicated other components of two machines would also draw more power, though perhaps not that much in comparison to processors here.)
[1] For a fuzzer you could easily give separate machines multiple random seeds, so really that is no issue. But for model checking/exhaustive search there needs to be a proper form of work distribution. I haven't a clue what the network bandwidth requirement of e.g. TLC, so that might also require faster NICs than one gets on motherboards to not stall processors on the network.
- "I value 30% faster compilation far more than..."
It takes 10,000 hours of wall-clock CPU time to reach your 200 euro figure. That tradeoff is against a 30% decrease in compilation time, or 3,000 hours.
Say what you will about idle power or whatever; but we're talking about the case of 100% all-core CPU for a programmer who is compiling something, and whether or not they should throttle those CPU cores to save money on power bills. The math is an easy "no".
Drops on average less than 10% multi core performance and no single core performance at all. Much quieter, cheaper, etc.
I also have a 35W 5750GE (8c16t too) and that’s also a very respectable performer compared to the 105W 5800X!
We’re so far off the ideal part of the perf/W curve it’s not funny, and I swear being closer to that is part of where Apple et al have done so well.
Edit: and from compute benchmarks on my 3090, the best perf/watt - measured in whole system total power drawn to complete a variety of cuda tasks, that card is most efficient around 225W or so… which is crazy considering they’re sold at 350W or more!
Having 25-30% faster compile times, faster video renders, higher frame rates, snappier responses is huge. I don’t understand why you’re suggesting it’s not doing anyone any good.
Our CPUs don’t run at 100% all of the time. If my CPU takes 200W instead of 65W for the small fraction of the day that I’m doing something CPU-intensive, the difference is absolutely negligible. I’m going to use more total power when I cook my dinner, probably.
If you are the kind of person willing to give up 1/4 of your performance for slightly lower power consumption under load, then go for those slower chips. You’re not the target market for the premium grade performance parts.
In my experience it's often not the case. Edge with its sleeping tabs function is a big improvement and example when app writers do care a bit about not wasting loads of energy. But it's really a very mixed bag. Many browsers do not.
And there is plenty of nuissance javascript that will peg the CPU and RAM all day long. If you're a Steam user, you know that almost all games, when you idle them in the background, and even in pause mode, peg one or a few CPU threads as well as the GPU.
The problem is greatly reduced for those editing their windows power profiles (or for linux users, cpufreq-set) to have their processors run at reduced frequencies. For those it might justify getting 105W X's over the 65W.
Now the almost 30% peak perf gain that the X models can sustain are potentially a big advantage. Yet real life usage speedup isn't as big as one might expect from looking at benchmark multithread compute scores; compiling has a lot of waiting on IO. And the single thread differences between the mellow frequency non-X and the souped up X versions are not very big, percentage wise. So you might get 30% speedup for some cases, but probably more typical difference might closer to ~15%.
iGPU throttling is an area where the 65W models would be disadvantaged at, sometimes significantly.
So in some cases (workstation type usage), especially if running without a dedicated video card, the 105W choice might admittedly make good sense.
65W vs 105W bottom line: I guess it's a case of YMMV.
https://wccftech.com/amd-and-intel-cpus-remain-vulnerable-to...
https://www.amd.com/en/corporate/product-security
Likely need to dig into PDFs to get answers.
I guess it was stated somewhere, and I missed it.
Meltdown never affected AMD CPUs in the first place.
Little cores (real or logical), with small to no reorder buffer, can be nicely asked to forgo any speculative execution.
Browsers worth their salt will take advantage of the fact that high risk threads (such as javascript, wasm, and other interpreted/jit code) can be asked by default to run on little cores. That is, unless a user left clicks on a tab to manually override the default, any scripts arising from said tab is taskset to small cores only. Browsers might also have a whitelist option regarding scripts originating from whitelisted internet domain and their ability to run on big cores.
Likewise the OS should allow the user to easily taskset any application to their desired set of cores (unix already does, via taskset command).
So instead of SMT2, I'd really like Zen 5 to have MT4, or MT5, where those multithreading schemes have just one big thread (massive reorder buffer) and three to four little threads (no ROB or tiny ROB) with an option to disable speculative execution---the latter keeping all those idle pipelines a little less idle.