2. But then there is the test which you linked to, where 7950X and 7950X3D gets roughly the same performance when operating at the same TDP (e.g. performance index 93 vs 91 when both @ 142W, and 80 vs 82 when both @ 88 W). Extracted from this diagram:
Ryzen 9 7950X3D im Test gegen 5800X3D und 13900K(S): Leistung in Anwendungen, Einrichtung und Fazit
AMD Ryzen 9 7950X3D im Test: Leistung in Anwendungen, Einrichtung und Fazit / Leistung und Effizienz in Anwendungenwww.computerbase.de
Computerbase are the most accurate reviews, they are the only ones who test the CPUs at different power levels, if you want more numbers click on "+ xx eintrage" on the graphs, that being said they always use top notch MBs wich are not the most economical idle power wise.
As for power 142W eco mode would be the recomended setting for a 7950X, there s not much benefit to use the standard stock setting while the 7950X3D can be used at stock since it s stock limited at 162W due to its rated 120W TDP.
Basically yeah. The CCD with V-cache is lower clocked while other CCD without huge cache clocks much higher. It would only help you if some workload is very cache dependent but only needs max 16 threads. But then you are better off with 7800X3D.Is all that extra cache useless, or only beneficial for some specific gaming use cases?
The advantage of the X3D is perfs in games, not only it perform 19% better (at 720P though) but it also does so at a lower power, 72W on average in games while the non X is at 105W average despite said lower perfs, to compare with the 13900K at 141W average, the numbers are on the previous page :Just so that I understand this correctly: When 7950X and 7950X3D use the same power setting (e.g. both at 120W), their performance is almost the same (within ~2 index points according to the Computerbase article for 88W and 142W). In that case, what's the point of paying more for 7950X3D? Is all that extra cache useless, or only beneficial for some specific gaming use cases?
Basically yeah. The CCD with V-cache is lower clocked while other CCD without huge cache clocks much higher. It would only help you if some workload is very cache dependent but only needs max 16 threads. But then you are better off with 7800X3D.
7950X3D seems to consume less total energy for completing a task though (even when 7950X3D @ 120W is compared to 7950X power limited to 105W):
Ryzen 7950X3D review: An expensive but incredibly efficient 16-core CPU
AMD's second run at a 3D V-Cache CPU is easier to recommend than the first.arstechnica.com
True. The huge V-cache makes it power efficient as it keeps frequently used data more readily available for the CPU cores to work on and the memory transfers from system RAM are reduced which leads to even more power efficiency. However, you pay for the power hungry V-cache with higher idle power draw of 97W vs 62W.7950X3D seems to consume less total energy for completing a task though (even when 7950X3D @ 120W is compared to 7950X power limited to 105W)
Their handbrake test suggest that the X3D is 27% faster than the 7950X at about same power, that s properly impossible, it would imply almost 27% higher frequency, their methodology is more than doubtfull, such numbers are impossible.
At Computerbase they find that at same 142W TDP the 7950X is slightly faster in Handbrake, wich is in line with all other tests, you can look at individual tests in their average by clicking "bearbeiten" in the multicore perfs graph.
Where did you get that 27% number from w.r.t. Handbrake performance difference between 7950X3D vs 7950X in the Ars technica article?
Pretty exciting if you will regularly convert 4K files to 1080p. Unfortunately, they don't mention which codec they used.Here, the 7950X3D encodes a 4K source video file into 1080p@30 at 124 frames per second, which is the fastest I have ever seen any chip other than an AMD Threadripper accomplish. It is nearly 35% faster than the Intel Core i9-13900K, widely considered the best consumer processor for creatives out there, and 36% faster than the Ryzen 9 7950X. This latter comparison is the most fascinating since it shows very clearly how much that extra cache memory alone can impact performance.
I made a slight mistake in the actual number, in the energy/task graph the 7950X consume 27% higher power to execute the task than the 7950X3D despite the two chips using about the same power and hence being at about same frequency.Where did you get that 27% number from w.r.t. Handbrake performance difference between 7950X3D vs 7950X in the Ars technica article? This is what I could find in that article, which does not indicate such a huge performance difference:
Strange that the results differ so much. That TechRadar article says the Handbrake performance difference is 36.76% between 7950X3D vs 7950X, but the Ars Technica article says there is not that much performance difference at all (but a big difference in energy consumption though).AMD Ryzen 9 7950X3D review: Team Red retakes the lead with its best CPU ever
The AMD Ryzen 9 7950X3D is the best gaming CPU there is, and it's not closewww.techradar.com
Pretty exciting if you will regularly convert 4K files to 1080p. Unfortunately, they don't mention which codec they used.
Not really. Using different source files, resolutions and codecs can account for the difference.Strange that the results differ so much.
I made a slight mistake in the actual number, in the energy/task graph the 7950X consume 27% higher power to execute the task than the 7950X3D despite the two chips using about the same power and hence being at about same frequency.
This 27% difference mean that the 7950X was working at 0/1.27 = 0.787x the frequency of the X3D, wich is not possible, at computerbase there s very few difference between the stock X3D and the 7950X@142W, actually even at 105WTDP/142WPPT the latter roughly match the X3D@120WTDP/162WPPT.
And another remark is that Computerbase measured the 7950X power in Handbrake, at stock 170WTDP/230WPPT setting it use only 189W, so even when limited to 142W it would get a similar score than at stock and perf/Watt would be close.
The only possibility for this graph at AST to be right is if they used a codec and a file that take big advantage of the V-cache, but so far it doesnt seems to be common if we are to look at Handrake tests here and there, but even then a higher throughput would forcibly materialize with a high power compsumption and power is limited here, wich lead me to think that there s a flaw in their methodology.
Edit : at Computerbase the 7950X has 48% better perf/watt in Handbrake than the the 13900K, in the graph below the advantage is only 36%; wich is yet another dubbious number.
True. The huge V-cache makes it power efficient as it keeps frequently used data more readily available for the CPU cores to work on and the memory transfers from system RAM are reduced which leads to even more power efficiency. However, you pay for the power hungry V-cache with higher idle power draw of 97W vs 62W.
Couldn't simply be that for Handbrake encoding 7950X3D has better perf/watt than 7950X? So even if they both have roughly the same performance and thus finish the task at approximately the same time, 7950X3D will have consumed less energy while doing so.
I mean just because a CPU is power limited to X Watts does not mean it uses X Watts all the time even at full load. So just as an example, if both have a power limit set to 120W, it could be that 7950X3D consumes 70W on average for the duration of the encoding task, while 7950X consumes 90W.
Their handbrake test suggest that the X3D is 27% faster than the 7950X at about same power, that s properly impossible, it would imply almost 27% higher frequency, their methodology is more than doubtfull, such numbers are impossible.
Indeed. Waiting for data is the single biggest bottleneck for CPUs. It's not impossible that running at a higher frequency just lets the CPU wait for data more often, and the V-cache can hold said data for specific workloads allowing the CPUs to compute more of its time where without the V-cache it instead has to wait for memory.Overcoming a 400 MHz deficit couldn't be anything other than compute efficiency enhanced by cache, right?
A real life example I can think of is pair programming and I have experienced the efficiency of that myself. I'll think about ways to do stuff with code instead of actually doing it when I'm alone, just wasting brain cycles because my brain can't provide input as well as process simultaneously as quickly as better programmers. But pair me up with someone and they keep asking questions and offering suggestions so that within a short amount of time, they will nudge me into writing the code that does exactly what we set out to do.So no, more performance doesn't always require higher frequency. It absolutely requires more time spent computing instead waiting, and V-cache can help with that depending on workloads.
Results vary wildly for idle power usage. I’ve seen anywhere from 20W to 60W. The primary factors I’ve seen are motherboard, DDR5 kit and whether or not it’s a dual CCD.From my own testing, when properly setup for power saving ( C-States manually enabled etc ) the different between last several gens of Intel/AMD was 25W.
AMD used to have nasty problem of not having iGPU for desktop CPUs, now it is fixed, but Zen4 is not exactly idle power efficiency "champion". I think ~1.04 AGESA they had nasty problems with core C6 CPU states, but it is fixed now and package C6 is broken even now.
understatement of the year with asus.motherboard that doesn’t crank vSOC it can be pretty reasonable.
Results vary wildly for idle power usage. I’ve seen anywhere from 20W to 60W. The primary factors I’ve seen are motherboard, DDR5 kit and whether or not it’s a dual CCD.
My personal system I had previous to this was running 40-55W at idle but it was due to a bad synergy of components (7950X, Asus X670E Strix & 64Gb DDR5-6000 CL30). If you’ve got a single ccd chip and a motherboard that doesn’t crank vSOC it can be pretty reasonable.