Intel Core i5-13600K and Core i7-13700K QS CPUs Benchmarked

[JustBenching]

According to TPU tests on efficiency, the "efficient cores" are only on par with Zen 3 cores.
Unless Zen 4 cores are less efficient, and / or RaptorLake "efficient cores" are much more efficient, otherwise your theory is not correct.

I wasn't talking about the Ecores, the golden cove cores (p cores) are way more efficient than ecores

 

[Deleted member 185088]

Do i have to waste time replying cause you cant understand what you are seeing on a graph? The performance numbers you quoted are from a variety of tests, including single thread tests. Then the power consumption you are quoting is from cinebench. Dont you see the problem?

Testing at the same wattage doesnt make any sense? Really? Is that how you test fans and coolers as well? Dont you normalise for noise first in order to determine which cooler has the best performance to noise ratio? Of course you do. Well that's what you should with cpus as well.

Let's just agree to disagree.

 

[Jimmy_]

Can't for the KS parts
As, 12900KS - 5.5ghz out of the box. I am hoping 13900ks to be 6.0ghz out of box - Thats pretty exciting - the only thing is how much power

 

[Mussels]

How did I come up with what, that amd can't compete in MT performance? It's pretty obvious. The 12600k is already 30 up to 60% faster (47% in CBR20 and 61% in CBR23). So in order to compete with the 12600k, they need to boost performace by up to 60%. Which is almost impossible, and yet the 13600k will be even faster than that, since it double on the ecores. Even the 13400 will probably have ecores this times around.

There is no way in hell amd will be more efficient. Only the 7950x stands a chance at that, the rest of the lineup won't do much when testing at same wattage

Hold on, 12600K is 60% faster than what CPU?


Tryting to keep up with claims here, but forgetting to name a CPU makes it harder

Going by the TPU review, the 5600x?
If so, i dont see the comparison... the 12600K is $130Au more, as well as requiring more expensive boards, RAM and cooling. Miss any one of those, and you do get reduced performance throwing all the equations off.

Unless you've redefined the meaning of the word, Intel are at a severe DISadvantage with energy efficiency, even with the E cores.

Did I misunderstand what you're trying to say, or have you got some warped info from somewhere?
Intel have some raw performance advantages, but performance pretty much lines up as
5800x - 12600k
5900x - 12700k
5950x -12900k (Intel does have an MT advantage here, the rest are more equal)

MT they trade blows pretty well til the very top - where the extra costs can be worth it to someone who needs the extra rendering speed for work
ST intel pulls ahead in certain workloads, but thats due to Zen 3's boost hitting heat density issues - all Zen 3 chips boost to basically the same MHz value. Any improvements to the IHS or overall CPU design could easily improve on that.

seriously i'm so baffled trying to make sense of this, unless you're talking about something like the warped userbench results

The 5800X3D looks to be doing ok for AMD peeps who are upgrading but Alder Lake looks to be the one peeps are going for when doing a new build.

Heh, not here in aus. The price cuts intel made are very US centric

I'm seeing a lot of people gobbling up older B450 stock with 5900x and 32GB of DDR4 with the new GPU prices

Kind of like how AMD is bending poor folks over a stump with Zen 4 not supporting DDR4?

Didn't you hear that they're rumoured to be bringing Zen 4 to the AM4 platform?
AM5 is DDR5, AM4 is DDR4.

Doesn't mean they cant release Zen4 on both, and that seems increasingly likely.

 

[JustBenching]

Hold on, 12600K is 60% faster than what CPU?


Tryting to keep up with claims here, but forgetting to name a CPU makes it harder

Going by the TPU review, the 5600x?
If so, i dont see the comparison... the 12600K is $130Au more, as well as requiring more expensive boards, RAM and cooling. Miss any one of those, and you do get reduced performance throwing all the equations off.

You think the 7600x will cost 200€ on release? No, right? Well that's my point, the 7600x won't be able to even beat the 12600k, let alone the 13600k. If pricing is similar to zen 3, the 7600x will cost roundabout between a 13400 (6p + 4e i assume) and a 13600k. And it will get beaten by both, unless the 7600x is 60% faster than the 5600x, which is absolutely impossible, and even then, it might just tie the 13400.

Unless you've redefined the meaning of the word, Intel are at a severe DISadvantage with energy efficiency, even with the E cores.

Because the tests are ran at different power levels. Limit a 12600k to 65w and will walk all over the 5600x. I know cause I tried, it score around 14k in CBR23. Comparing efficiency at different wattages is just dumb. That's like comparing fans at different noise levels. Whats the point of that? You need to normalize for something else the CPU with lower limits will seem more efficient, even though it's not.

If you could interpret the graphs you would understand it yourself. The 12600k at 125w beats both the 5600x and the 5800x in performance, handily. So if you lower it's power limit in order to match the performance it would handily beat them in efficiency as well. The 12600k can match the 5800x at 70-75W

Regarding efficiency I was talking about the GC cores specifically. Golden cove / Pcores, whatever you wanna call them, walk all over zen 3 in terms of efficiency. Just test 8gc cores to 8 zen 3 cores at same wattage and...yeah. The gap would be huge. 8 zen 3 cores need 150w+ to match 8GC cores at 65w.

 

[LuxZg]

Efficiency is a tricky one to measure.

Suppose that a processor uses a constant 100 watts to complete a workload in one minute, but another one uses 75 watts to complete a workload in one minute and 20 seconds. The 100W processor is then more efficient than the 75W one.

A faster core that chugs power but completes a task quicker is more desirable over one that is leaner on power but takes a longer span of time to complete the same task in terms of total power consumption.

Sorry, missed that comment. But I think it's just wrong thinking when Intel has +50% consumption but doesn't have +50% performance. Likewise, some tasks are pretty much based on time. If you game for 2 hours, Intel will just waste power. Anyway, if you want you can measure Alder Lake vs Zen 3 on "task base" and it won't look good. Race to finish is valid philosophy when it actually saves power. There was a good article about Alder Lake power consumption and efficiency which compared performance and efficiency cores as well as Zen 2 (not Zen 3 unfortunately) and Alder Lake P-cores have best efficiency at 1.5-2 GHz while Zen 2 actually peaks at 3.5 GHz. See this link:
https://chipsandcheese.com/2022/01/28/alder-lakes-power-efficiency-a-complicated-picture/
And at 3.5GHz frequency Zen 2 beat AL more than twice.

Now I couldn't find a test that would show something like encoding 2h video and measure kWh spent for total system. But I'm certain that any Zen 3 would win vs Alder Lake. You can also scroll few comments above where kilojoule power for Cinebench run is graphed from TPU review, Zen 3 wins the efficiency, and isn't all that far in total performance either.

(edit: TPU saved my earlier post after all, yay! made edits to make sense)

 

[Dr. Dro]

Sorry, missed that comment. But I think it's just wrong thinking when Intel has +50% consumption but doesn't have +50% performance. Likewise, some tasks are pretty much based on time. If you game for 2 hours, Intel will just waste power. Anyway, if you want you can measure Alder Lake vs Zen 3 on "task base" and it won't look good. Race to finish is valid philosophy when it actually saves power. There was a good article about Alder Lake power consumption and efficiency which compared performance and efficiency cores as well as Zen 2 (not Zen 3 unfortunately) and Alder Lake P-cores have best efficiency at 1.5-2 GHz while Zen 2 actually peaks at 3.5 GHz. See this link:
https://chipsandcheese.com/2022/01/28/alder-lakes-power-efficiency-a-complicated-picture/
And at 3.5GHz frequency Zen 2 beat AL more than twice.

Now I couldn't find a test that would show something like encoding 2h video and measure kWh spent for total system. But I'm certain that any Zen 3 would win vs Alder Lake. You can also scroll few comments above where kilojoule power for Cinebench run is graphed from TPU review, Zen 3 wins the efficiency, and isn't all that far in total performance either.

(edit: TPU saved my earlier post after all, yay! made edits to make sense)

What I said is architecture agnostic, it's a power mechanic. Small bursts to higher frequencies and smart TDP management, even if unsustainable in the long run, reduce task completion time significantly enough for the overall power usage to be reduced relative to that task's completion. Obviously, this is strictly limited to bursty workloads such as opening a website, or converting a single audio file that would convert in like 5 seconds on a modern CPU, if you're speaking of a continuous full load scenario, this would never work. This is the concept behind Speed Shift and I believe Ryzen's impressive turbo mechanism.

Modern processors can switch power states in a millisecond, Zen 3 for example can change frequency and voltage states over 1000 times a second, so you can easily understand how this mechanic works.

I'm aware of chips and cheese, great place, it's actually written up by an acquaintance of mine, we shared a Discord server. Great guys and always a great read. You should read their IPC projection article, it's actually amazing.

 

[LuxZg]

What I said is architecture agnostic, it's a power mechanic. Small bursts to higher frequencies and smart TDP management, even if unsustainable in the long run, reduce task completion time significantly enough for the overall power usage to be reduced relative to that task's completion. Obviously, this is strictly limited to bursty workloads such as opening a website, or converting a single audio file that would convert in like 5 seconds on a modern CPU, if you're speaking of a continuous full load scenario, this would never work. This is the concept behind Speed Shift and I believe Ryzen's impressive turbo mechanism.

Modern processors can switch power states in a millisecond, Zen 3 for example can change frequency and voltage states over 1000 times a second, so you can easily understand how this mechanic works.

I'm aware of chips and cheese, great place, it's actually written up by an acquaintance of mine, we shared a Discord server. Great guys and always a great read. You should read their IPC projection article, it's actually amazing.

I agree with everything except that it's architecture agnostic. It isn't even model agnostic. I mean, sure - philosophy is agnostic. Yet the implementation results can be wildly different. Race to finish isn't a guarantee of power saving on its own. Eg if task takes 10 seconds on 200W that's 2kJ. But if we "race" and use 320W to finish in 8s then we spend ~2.5kJ. So THAT brakes the philosophic model. And slower CPU is actually more efficient. That's what happens to Alder Lake. It is very efficient at 1500MHz, but from there it shoots up, and if it's around 1700J and 5800J at 5GHz then the efficiency falls. Since it's same CPU, we can say that task competition will be almost linear from 1.5 to 5GHz so 3.33x faster. But it spends 3.2x the energy. So it goes outside efficiency curve and race to finish actually spends more power. On other hand you can take Zen 2 graphs and see that at same task it uses about 1200 J at 1.5GHz but 1100 J at 3.5GHz. So task is done 2.33x faster at same energy cost (bit lower actually) and in that case yes it did make sense of "race to finish" philosophy. That's underlined by TPU graphs as well, where you can clearly see 12600K and 12700K being slightly more efficient by spending less energy to finish that multithreaded Cinebench run. If you look at 12400F review, it too beats 12900K ever so slightly. And eg Ryzen 5600X is likewise even better at this metric than 12400F (so 12600/700/900 as well). Now sure you can underclock, undervolt, and whatever, but looking at CPUs as they cane from factory, Intel obviously missed the sweet spot. Add to all that a fact that at idle Intel system draws slightly more power, racing to finish isn't saving any power in that generation. Now sure, it saves *time* for end user, if they work with tasks that are very time sensitive, but that's not a majority of people, and they won't mind less efficient CPU if they can do more/earn more in one work day. Most people will do a task, then PC will idle waiting for user. So if task X spends 10kJ on 12900K and 9.5kJ on 12400F and 9kJ on 5600X, and then idles at 55W (whole system) for 12900K, 54W for 12400F and 51W for 5600X then you can't say that 12900K is somehow more efficient JUST because it finished quicker. So again, yes, philosophy of race to finish (and idle/low power state) is correct one, but execution gives different results. Plus, again, if you game 4h on eg 12900K and 5600K, with everything else in PC being same, you will get higher power bill (and more heat in your room) with 12900K. Sure you may have 2-3 FPS more but as if that will actually change your experience. Anyway, to finally apply that to 13000 series, if they are to be actually efficient they'll need more than just relying on race to finish. Because eg yesterday or whatever they've leaked supposed 350W 13900K with power limit removed, but it spent 60% more watts to be 14% faster. That would be 60% more watts for 14% less seconds, so about 36% more energy spent to finish a task. Luckily that's overclocking, but still, they are obviously not aiming at actual efficiency. I mean, just look at laptops, they actually regressed in battery life with Alder Lake, now it seems they'll regress again. Sure we get more speed but also less run time. I am similarly disappointed with Zen 4 power rumors, waiting to see if world went all crazy for next gen

 

[Dr. Dro]

I agree with everything except that it's architecture agnostic. It isn't even model agnostic. I mean, sure - philosophy is agnostic. Yet the implementation results can be wildly different. Race to finish isn't a guarantee of power saving on its own. Eg if task takes 10 seconds on 200W that's 2kJ. But if we "race" and use 320W to finish in 8s then we spend ~2.5kJ. So THAT brakes the philosophic model. And slower CPU is actually more efficient. That's what happens to Alder Lake. It is very efficient at 1500MHz, but from there it shoots up, and if it's around 1700J and 5800J at 5GHz then the efficiency falls. Since it's same CPU, we can say that task competition will be almost linear from 1.5 to 5GHz so 3.33x faster. But it spends 3.2x the energy. So it goes outside efficiency curve and race to finish actually spends more power. On other hand you can take Zen 2 graphs and see that at same task it uses about 1200 J at 1.5GHz but 1100 J at 3.5GHz. So task is done 2.33x faster at same energy cost (bit lower actually) and in that case yes it did make sense of "race to finish" philosophy. That's underlined by TPU graphs as well, where you can clearly see 12600K and 12700K being slightly more efficient by spending less energy to finish that multithreaded Cinebench run. If you look at 12400F review, it too beats 12900K ever so slightly. And eg Ryzen 5600X is likewise even better at this metric than 12400F (so 12600/700/900 as well). Now sure you can underclock, undervolt, and whatever, but looking at CPUs as they cane from factory, Intel obviously missed the sweet spot. Add to all that a fact that at idle Intel system draws slightly more power, racing to finish isn't saving any power in that generation. Now sure, it saves *time* for end user, if they work with tasks that are very time sensitive, but that's not a majority of people, and they won't mind less efficient CPU if they can do more/earn more in one work day. Most people will do a task, then PC will idle waiting for user. So if task X spends 10kJ on 12900K and 9.5kJ on 12400F and 9kJ on 5600X, and then idles at 55W (whole system) for 12900K, 54W for 12400F and 51W for 5600X then you can't say that 12900K is somehow more efficient JUST because it finished quicker. So again, yes, philosophy of race to finish (and idle/low power state) is correct one, but execution gives different results. Plus, again, if you game 4h on eg 12900K and 5600K, with everything else in PC being same, you will get higher power bill (and more heat in your room) with 12900K. Sure you may have 2-3 FPS more but as if that will actually change your experience. Anyway, to finally apply that to 13000 series, if they are to be actually efficient they'll need more than just relying on race to finish. Because eg yesterday or whatever they've leaked supposed 350W 13900K with power limit removed, but it spent 60% more watts to be 14% faster. That would be 60% more watts for 14% less seconds, so about 36% more energy spent to finish a task. Luckily that's overclocking, but still, they are obviously not aiming at actual efficiency. I mean, just look at laptops, they actually regressed in battery life with Alder Lake, now it seems they'll regress again. Sure we get more speed but also less run time. I am similarly disappointed with Zen 4 power rumors, waiting to see if world went all crazy for next gen

I mean, math has never been my forte, but you definitely got what I meant lol

It is definitely true that you need to account for each architecture's characteristics and even each sample's operating conditions to come up with a precise estimate, but the general formula is there in every case. In this curve there will be a maximum efficiency and a balanced point, and past a certain point the energy requirements begin spiraling out of control until you need massive increases to tap into minimal potential.

 

[Mussels]

Felt like about 2,000 words there to say

"Completing a task faster doesnt matter if the task doesnt end"

Rendering work? Sure, higher power draw balances out.

Gaming? Hell no. It's a goddamn mess. If you're using 200W more and playing the same game it doesnt matter if the FPS is slightly higher, for a task like that all that matters is the total consumed over the entire time. And for the intel side, that's just insanely worse unless you throw in power limits that reduce the performance anyway.