MusselsThat's just ridiculous.
That's not how efficiency works in the slightest - that's IPC.
It's also a terrible idea, because components are designed to work in sync with each other and if you set them outside their architectures optimal ranges, they'll generally perform far worse.
It also has nothing to do with how any of these products are intended to run, so it's a data point for IPC and otherwise utterly useless to everyone. Double so when that IPC varies per workload per design - SSE, AVX, AVX-512, etc etc.


Efficiency is time taken to complete a task. A faster, higher wattage part can complete the task quicker. THAT is efficiency. Energy efficiency is when you math the time taken with energy consumed.

Testing at same wattage is IPC? What?

fevgatosCalling it just a "synthetic" to discard the performance numbers is wrong.

You're missing the point. Calling a benchmark "synthetic" is just qualifying it as non-real-world test. Generally, synthetic benchmarks are wonderfully reliable as they make comparing performance very objective. The problem with them is relying on any single benchmark for an overall viewpoint. A lot of people make that mistake. The key is to look at what benchmarks mimic closely what tasks are important to you.

fevgatosTesting at same wattage is IPC? What?

sorry, same *clock speed* is IPC.

same wattage is something entirely different, much like amperage you can't use that if the core counts aren't the same

5800x, 5900x and 5950x are all same wattage at stock - is that the comaprison you were looking for? Because unlocked vs locked the MT performance is wildly different, and performance per-core values go all over the place. Under 8 threads, the 5800x is more efficient per watt than the higher core count models because some power is wasted even when the unused cores are idle, but when they're active they run more cores at lower everything (amps, volts, watts, clocks) to achieve more MT performance in the same wattage. That's why they're less efficient than the 5800x with <8 thread tasks, but more with 8+ (Where a real core can finish work faster than SMT, and idle)

It's not a category that can be tested because you'd need to test it in dozens of different ways to have enough data to make any sort of conclusion, and they'd never agree on a 'best' outcome.

Musselssorry, same *clock speed* is IPC.

same wattage is something entirely different, much like amperage you can't use that if the core counts aren't the same

5800x, 5900x and 5950x are all same wattage at stock - is that the comaprison you were looking for? Because unlocked vs locked the MT performance is wildly different, and performance per-core values go all over the place. Under 8 threads, the 5800x is more efficient per watt than the higher core count models because some power is wasted even when the unused cores are idle, but when they're active they run more cores at lower everything (amps, volts, watts, clocks) to achieve more MT performance in the same wattage. That's why they're less efficient than the 5800x with <8 thread tasks, but more with 8+ (Where a real core can finish work faster than SMT, and idle)

It's not a category that can be tested because you'd need to test it in dozens of different ways to have enough data to make any sort of conclusion, and they'd never agree on a 'best' outcome.

Οk, so it was a typo. I was worried for a second !

The thing is, if you wanna test efficiency, testing at dfferent wattages doesn't tell you anything. The 5950x at 100w is more efficient than the 5950x at 150w. So which CPU is more efficient? Now imagine doing that with different models. Why not just test at iso wattagae...

fevgatosThe 5950x at 100w is more efficient than the 5950x at 150w. So which CPU is more efficient? Now imagine doing that with different models. Why not just test at iso wattagae...

This is where you seem to want an impossible target

Why 100 and 150? Why not every value in between? It's not 'fair' to CPU X Y and Z if the testing is biased towards whichever one is most efficient at the tested wattage
what the hell is 'iso wattage' ?


As i said, ST testing is always more efficient on lower core CPUs and MT testing more efficient at higher wattage. That doesn't change - so the answer to most of what you're asking already exists. You're not asking a complete question but more of a "what if..." and it's so vague it'd take thousands of hours of testing and not get the answer you want, because you don't seem to know what you're asking for

MusselsThis is where you seem to want an impossible target

Why 100 and 150? Why not every value in between? It's not 'fair' to CPU X Y and Z if the testing is biased towards whichever one is most efficient at the tested wattage
what the hell is 'iso wattage' ?


As i said, ST testing is always more efficient on lower core CPUs and MT testing more efficient at higher wattage. That doesn't change - so the answer to most of what you're asking already exists. You're not asking a complete question but more of a "what if..." and it's so vague it'd take thousands of hours of testing and not get the answer you want, because you don't seem to know what you're asking for

It's not hard to pick a proper wattage to test. Just ask creative people what power limits are they running on their cpu and you'll get your answer. You can even see how puget configures their systems. You don't go above 200w on big cpus and you don't go above 150 on the smaller ones and that's it.

fevgatost's not hard to pick a proper wattage to test.

Pick one for AM4, AM5 and intels current socket then and tell me what they are instead of saying why they're so easy to do.

200? That lets any AMD CPU go at full performance but cripples intel severely - why the biased setting?

MusselsPick one for AM4, AM5 and intels current socket then and tell me what they are instead of saying why they're so easy to do.

200? That lets any AMD CPU go at full performance but cripples intel severely - why the biased setting?

Again, it's really not that hard. High end cpus should be capped at something between 150 and 180, the more midrange options at 120. Performance gains after that are negligible.

An efficiency comparison obviously only matters for people that care about.... efficiency. Check puget, they preconfigure systems to balance between efficiency and performance for content creators, the people that actually run long hour multithreaded workloads. Nobody runs those at unlimited power.

fevgatosAgain, it's really not that hard. High end cpus should be capped at something between 150 and 180, the more midrange options at 120. Performance gains after that are negligible.

No they arent.
You're throwing what you want out as if its fact, and you're not backing ANY of this up.

Intel Core i9-12900K Alder Lake Tested at Power Limits between 50 W and 241 W | TechPowerUp


Does performance gains reduce at higher wattages? sure. But that's not what you've been claiming, so try not to move your goalposts around so often.


Whatever you're thinking of, is only relevant to low core count CPUs on older designs and it's extremely biased towards certain sets of hardware.
Just because the hardware you like does well with those limits for your intended use, doesn't mean it makes any sense at all for anyone else, in any other workload.

I'll repeat it again:

• ST workloads dont give a shit about power limits
• MT workloads with a low power limit benefit CPUs with more cores every time
• Optimal wattages varies per CPU design, let alone per generation/socket.
• Low power limits destroy high core count CPUs in low threaded tasks, as they lose power due to the extra cores/cache simply existing.
  
  What you're asking for is a fantasy. It's contradictory and there can never be one answer to this. You're asking for "at X wattage, what CPU is the best" and that cant be determined without testing every single CPU ever released - go do that yourself. Don't forget every possible RAM combination as that comes from the CPU's power on many platforms as well.

I asked about this as it's something i could genuinely have tested for you, but you can't make any argument for it whatsoever other than "I demand reality be this way"

MusselsNo they arent.
You're throwing what you want out as if its fact, and you're not backing ANY of this up.

Intel Core i9-12900K Alder Lake Tested at Power Limits between 50 W and 241 W | TechPowerUp


Does performance gains reduce at higher wattages? sure. But that's not what you've been claiming, so try not to move your goalposts around so often.


Whatever you're thinking of, is only relevant to low core count CPUs on older designs and it's extremely biased towards certain sets of hardware.
Just because the hardware you like does well with those limits for your intended use, doesn't mean it makes any sense at all for anyone else, in any other workload.

I'll repeat it again:

• ST workloads dont give a shit about power limits
• MT workloads with a low power limit benefit CPUs with more cores every time
• Optimal wattages varies per CPU design, let alone per generation/socket.
• Low power limits destroy high core count CPUs in low threaded tasks, as they lose power due to the extra cores/cache simply existing.
  
  What you're asking for is a fantasy. It's contradictory and there can never be one answer to this. You're asking for "at X wattage, what CPU is the best" and that cant be determined without testing every single CPU ever released - go do that yourself. Don't forget every possible RAM combination as that comes from the CPU's power on many platforms as well.

I asked about this as it's something i could genuinely have tested for you, but you can't make any argument for it whatsoever other than "I demand reality be this way"

The graph you posted is wrong though. We've been over this if you remember. You can check any other review with power limited numbers. 12900k @ 125 = 23500-24000 score. At 75w it should be around 19-20k.


I don't know what you mean "the hardware I like". All i'm saying is,there are people who care about efficiency, and people who don't. People who don't, well they are not part of this conversation. So we are left with the people that do care about efficiency. Those people will most like not run blender for 15 hours at 390 watts. And I know cause I know a couple of people that make heavy use of these CPUs, they all have 7950xs and 13900ks, all caped anywhere below 200w. If someone cares about efficiency why would he give a damn about what happens when you run blender at 400 watts? He wouldn't. It's that simple

And just to make it obvious, let's say I ask the following

"I really care about efficiency, I found the 7900x for 400€ and the 7900 for 500€, is it worth to spend the extra 100€ for the more efficient 7900?"
The answer is no, they are both the same CPU, get the 7900x and power limit it to the same wattage and voila.

And btw, this is how we test almost everything. Fans for example, are tested at equal noise levels. If you don't test them at same noise levels then whichever has the highest RPM will score the best temperatures and also the worst noise, but that information doesn't really tell you anything about your daily use cases. Cars, how do you test their efficiency? Running one at 200km/h and the other one at 100km/h, again, doesn't tell you anything about how efficient each car actually is.

But after all, that's my opinion, I think testing for efficiency at different wattages is completely useless information that doesn't really help anyone. Except I guess people that don't know that they can change the power limits with a click of a button, but these people probably don't care about efficiency, cause if they did, they would know that already. If you run at different wattages then youll end up thinking that the 5950x is more efficient than the 7950x, which is just not true, not even close actually.

EG1. I don't need you to test it, I know the numbers already. The 7950x is the most efficient at 125w with a slight edge over the 13900k depending on the workload, averagely it's around 10% faster / more efficient.