News Core i9-14900K, Core i7-14700K CPUs Benchmarked

[bit_user]

Show me one benchmark that shows that happening because for the last forever years every single benchmark is with the clocks running pedal to the metal, either with unlimited power or with the a certain power target but always with all cores running as fast as the power allows.

Maybe it would help if we work an example. So, let's consider an all-core integer workload, similar to the 7zip case measured here:

​

​

For the CPU, consider the i9-12900, once it reaches the 65 W point in its operating envelope.
So, for optimal throughput, you'd want to schedule:

8x E-cores @ 3.1 GHz, consuming 27.86 W and providing 40.50 MB/s of throughput
8x P-cores @ 2.5 GHz, consuming 29.38 W and providing 40.82 MB/s of throughput
---
Total: 57.24 W; 81.32 MB/s.
​

That only leaves 7.76 W for package power, which is probably on the low side. So, the easiest thing would probably be to bump the E-cores down a step or two, since they seem to have more frequency steps, in that range, than the P-cores.

If you just ran the 8 P-cores flat out, then you'd have to go with about 56 W for them, which achieves about 56 MB/s of throughput. That's only about 69% of the performance attained with the 8 + 8 combination.

If we consider a hypothetical CPU with 12 P-cores, then you could hit about 54 W at 69 MB/s of throughput. That's still only about 85% of the 8 + 8 combination considered above.

how is the battery life at very low power where the e-cores are the most efficient?

If you're saying they're only more efficient at 1.1 GHz, then realistically nobody uses those frequencies. So, either you're wrong or Intel is wrong.

Look at this picture, if you can allow your CPU to use more than 90W and unless you have to run every core in your CPU at full power all the time then you can replace every single e-core with the equal amount of p-cores and make those replacement p-cores use the same amount of power as the e-cores do now and you would get higher performance for the same amount of total power,

Where do you get 90 W?

Given that example, you could have a 16 P-core CPU that uses about 70 W @ 2.5 GHz and is the same speed as a 16 E-core CPU. That seems pointless, to me. If that's as fast as you're going to run those P-cores, then save money and go with 16x E-cores, instead.

Anyway, it's unrealistic just making an arbitrarily large CPU. You should really compare against something that's approximately the same cost (i.e. area). So, that would mean about 10 P-cores.

Working with a 95 W package power budget, if we allow 10 W for uncore, then here's the optimal 8 + 8 configuration:

8x E-cores @ 27.9 W -> 40.5 MB/s
8x P-cores @ 56.0 W -> 56.0 MB/s
---
Total: 83.9 W; 96.5 MB/s
​

Using 10 P-cores, you'd get only 76.25 MB/s in 85 W. Even with 12 P-cores, I get only 84 MB/s.

unless you are forced to use your desktop PC at below 90W but still with all its cores enabled for some reason

Oh, you mean like the i9-12900 (65 W)?

Product Specifications

quick reference guide including specifications, features, pricing, compatibility, design documentation, ordering codes, spec codes and more.

ark.intel.com

I actually have one of these in my work PC. It's a Dell compact desktop and that's the fastest CPU we could configure it with.

 

[bit_user]

Now, let's do a more interesting example, like 125 W. This one is neat, because it lets us explore running the E-cores faster than 3.1 GHz. Again, assuming 10 W uncore leaves us a 115 W budget for the cores:

8x E-cores @ 48 W -> 46 MB/s
8x P-cores @ 67 W -> 58 MB/s
---
Total: 115 W -> 104 MB/s
​

Using 10 P-cores, there's a datapoint that gives us 117.5 W -> 82.5 MB/s.
Using 12 P-cores, the closest datapoint is 109.5 W -> 93 MB/s.

So, with any kind of reasonable power budget, you're not beating the performance of a hybrid configuration, with any CPU that's close in size.

 

[TerryLaze]

Maybe it would help if we work an example. So, let's consider an all-core integer workload, similar to the 7zip case measured here:

​

​

For the CPU, consider the i9-12900, once it reaches the 65 W point in its operating envelope.
So, for optimal throughput, you'd want to schedule:

​

8x E-cores @ 3.1 GHz, consuming 27.86 W and providing 40.50 MB/s of throughput​

8x P-cores @ 2.5 GHz, consuming 29.38 W and providing 40.82 MB/s of throughput​

---​

Total: 57.24 W; 81.32 MB/s.​

​

That only leaves 7.76 W for package power, which is probably on the low side. So, the easiest thing would probably be to bump the E-cores down a step or two, since they seem to have more frequency steps, in that range, than the P-cores.

If you just ran the 8 P-cores flat out, then you'd have to go with about 56 W for them, which achieves about 56 MB/s of throughput. That's only about 69% of the performance attained with the 8 + 8 combination.

If we consider a hypothetical CPU with 12 P-cores, then you could hit about 54 W at 69 MB/s of throughput. That's still only about 85% of the 8 + 8 combination considered above.

Yes if you don't replace every e-core with an p-core you are in trouble, that explains the cost and area efficiency, when are you going to explain the power efficiency.

Where do you get 90 W?

Sorry, I was thinking about the 13900k that has 16 e-cores and not 8.

Given that example, you could have a 16 P-core CPU that uses about 70 W @ 2.5 GHz and is the same speed as a 16 E-core CPU. That seems pointless, to me. If that's as fast as you're going to run those P-cores, then save money and go with 16x E-cores, instead.

YES! If you take the cross over point were both types are pretty much the same then there is not going to be much of a difference.
Why do you think I was taking the point at which the e-cores provide the most performance???????
The point at which the p-cores are like 20-30% more effective?

Anyway, it's unrealistic just making an arbitrarily large CPU. You should really compare against something that's approximately the same cost (i.e. area). So, that would mean about 10 P-cores.

So again you are talking about cost and size even though the only thing I was against was the power-efficiency quote...
I was pretty much the first that said that intel is doing this for cost reasons.

Oh, you mean like the i9-12900 (65 W)?

​

Product Specifications

quick reference guide including specifications, features, pricing, compatibility, design documentation, ordering codes, spec codes and more.

ark.intel.com

​

I actually have one of these in my work PC. It's a Dell compact desktop and that's the fastest CPU we could configure it with.

Yeah, as I said I confused it with the 13900,
for the 12900 it is 2*4 cores@15w for the e cores and 2*4 cores@15w for the p-cores so 30+30=60

 

[TerryLaze]

Now, let's do a more interesting example, like 125 W. This one is neat, because it lets us explore running the E-cores faster than 3.1 GHz. Again, assuming 10 W uncore leaves us a 115 W budget for the cores:

​

8x E-cores @ 48 W -> 46 MB/s​

8x P-cores @ 67 W -> 58 MB/s​

---​

Total: 115 W -> 104 MB/s​

​

Using 10 P-cores, there's a datapoint that gives us 117.5 W -> 82.5 MB/s.
Using 12 P-cores, the closest datapoint is 109.5 W -> 93 MB/s.

So, with any kind of reasonable power budget, you're not beating the performance of a hybrid configuration, with any CPU that's close in size.

You assume that anything above 125W is not reasonable?????
Also again a cost and size but not a power efficiency thing.
Nobody wants to be using many full cores, it's just way too expensive, why do you think that even AMD is looking into the hybrid CPUs with smaller cores?

 

[bit_user]

Yes if you don't replace every e-core with an p-core you are in trouble, that explains the cost and area efficiency, when are you going to explain the power efficiency.

...

So again you are talking about cost and size even though the only thing I was against was the power-efficiency quote...
I was pretty much the first that said that intel is doing this for cost reasons.

I included cost in the same list of advantages as energy-efficiency. So, I'm not going to argue against myself.

However, comparing against a hypothetical 16x P-core client CPU, as if such a thing were a viable alternative, is not a particularly honest or useful discussion. So, we really should restrict the discussion to what can be sold at the same price point, which is a 10x or 12x P-core CPU. And, from that perspective, I already showed that you get more performance in the i9-12900K's 125 W TDP with 8x E-cores than you would with 10x or even 12x P-cores.

We could go even further, and look at the unlimited scenario. The data goes up to 137 W for 8x P-cores, which should produce about 70.5 MB/s of throughput. That works out to 0.51 MB/s per W. If you add the 8x E-cores, power goes up to 185 W and throughput goes up to 116.5 MB/s, increasing efficiency to 0.63 MB/s per W. So, it's as I said: pick your power target, and adding E-cores into the mix can improve your efficiency.

You assume that anything above 125W is not reasonable?

You have to keep in mind that the majority of these CPUs are not going into overclocked gaming systems. They're shipping in OEM boxes that run at stock power limits, laptops, and other scenarios where you can't just churn out 200+ Watts, indefinitely.

My claim was that E-cores enable performance to scale more efficiently. Remember this graph?

​

It's showing you can't run the P-cores above 3.7 GHz without resulting in worse efficiency than the E-cores. However, by using a mix of cores, you get more performance with efficiency that's merely a blend of the two, instead of being as bad as P-cores alone would deliver.

Q.E.D.

 

[TerryLaze]

Why do you keep repeating things, that I already agreed with several times, over and over again...ohhhhh right, I forgot whom I talking with.

However, comparing against a hypothetical 16x P-core client CPU, as if such a thing were a viable alternative, is not a particularly honest or useful discussion.

So what about them newfangled 5950x and 7950x CPUs? I know I often said that those are just acts of desperation and that they aren't viable and way to costly for AMD to keep making them but I think it was also you that said that those are like awesome.
I guess those are viable because AMD has to pay TSMC an 50% or whatever it is overhead on them and it's not like prices at TSMC are going up all the time or something.

In your example power goes up to 185 W and throughput goes up to 116.5 MB/s and with 16 p-cores the same 185W (divided by 4=46w) would get you about 33MB/sec per cluster for a total of 132Mb/sec for the exact same power, so again not an issue of power-efficiency just a matter of how desperate a company is to make a 16 full core CPU.
Intel just prefers to be making more money by only putting 8 full cores in every CPU they make.

And because they aren't e-cores that would just be the base performance and you could throw efficiency to the winds and crank all them cores up to overdrive, it's only 3-4MB/sec more per cluster but it's more performance you can get out of a single CPU, if you have to, which would be impossible with e-cores because they just don't go above a certain performance level.

 

[Ogotai]

um terry, isnt this all about intels efficiency, and its cores ? what does amd have to do with this ? why drag amd into this ?

 

[TerryLaze]

um terry, isnt this all about intels efficiency, and its cores ? what does amd have to do with this ? why drag amd into this ?

Who else has a desktop CPU with 16 full cores that people can buy?!
If somebody tells you that a 16 full core CPU is not viable then bringing up a consumer CPU with 16 full cores isn't ok?!

 

[bit_user]

So what about them newfangled 5950x and 7950x CPUs?

We already agreed on the matter of E-cores benefiting cost-efficiency, so the most fair way to control for that and simply look at energy-efficiency is to hold the die size roughly constant, irrespective of AMD or anything else. That means comparing an 8 + 8 configuration vs. 10 + 0.

In your example power goes up to 185 W and throughput goes up to 116.5 MB/s and with 16 p-cores

What I'm saying is that, for a given die area and power target, you can use E-cores to add both performance and efficiency. This is in line with my original claim - that E-cores are both about cost and power.

If I get a chance, I'll see if I can compute efficiency scores for the datapoints in ChipsAndCheese' graphs. That should make it clearer how you can dial in a mix of clock speeds between the E and P cores, such that it improves performance at any power target than you'd get with 10 P-cores.

 

[bit_user]

Who else has a desktop CPU with 16 full cores that people can buy?!

Golden Cove + L2 is 7.43 mm^2. Zen 3 is 4.05 mm^2.

Die walkthrough: Alder Lake-S/P and a touch of Zen 3

I try to make at least one video project per month and in January I was bouncing back and forth between three different projects.

locuza.substack.com

I didn't get a precise figure on Zen 4, but the CCD is 84.3% as big as Zen 3's. Because it has the same amount of L3 cache and we know SRAM doesn't scale down well, that suggests the Zen 4 cores are at least that much smaller than Zen 3.

All of that is to say that because AMD can ship 16x P-core CPUs doesn't necessarily mean it's a viable option for Intel to do so. One glance at the price sheet for Xeon W 2400 confirms this:

​

If somebody tells you that a 16 full core CPU is not viable then bringing up a consumer CPU with 16 full cores isn't ok?!

Intel is asking $1389 for 16 P-core CPUs featuring Golden Cove.

 

[TerryLaze]

We already agreed on the matter of E-cores benefiting cost-efficiency, so the most fair way to control for that and simply look at energy-efficiency is to hold the die size roughly constant, irrespective of AMD or anything else. That means comparing an 8 + 8 configuration vs. 10 + 0.


What I'm saying is that, for a given die area and power target, you can use E-cores to add both performance and efficiency. This is in line with my original claim - that E-cores are both about cost and power.

If I get a chance, I'll see if I can compute efficiency scores for the datapoints in ChipsAndCheese' graphs. That should make it clearer how you can dial in a mix of clock speeds between the E and P cores, such that it improves performance at any power target than you'd get with 10 P-cores.

Then at least say area divided by efficiency ,power-efficiency for performance scaling , so getting as much performance as possible from as little power as possible is something completely different from 'we have to fit that much efficiency (performance/power) into that much size' .

And just because you add other stuff to it doesn't make it any more correct.

Intel is asking $1389 for 16 P-core CPUs featuring Golden Cove.

Which is what AMD should be charging and would be charging if they weren't desperate, or at least it would be close to that, but because intel can match their 16 full core CPU performance with a $600 CPU that only has 8 full cores AMD can't charge much more.

Die sizes change all the time so the cores being bigger is a non argument.
Cost is, which everybody agreed on from the beginning, but cost isn't something that would make this impossible.
But yeah, as you discovered yourself intel can get twice the money for a CPU with twice the p-cores so why the heck would they want to change that.

 

[TerryLaze]

Golden Cove + L2 is 7.43 mm^2. Zen 3 is 4.05 mm^2.

​

Die walkthrough: Alder Lake-S/P and a touch of Zen 3

I try to make at least one video project per month and in January I was bouncing back and forth between three different projects.

locuza.substack.com

​

I didn't get a precise figure on Zen 4, but the CCD is 84.3% as big as Zen 3's. Because it has the same amount of L3 cache and we know SRAM doesn't scale down well, that suggests the Zen 4 cores are at least that much smaller than Zen 3.

All of that is to say that because AMD can ship 16x P-core CPUs doesn't necessarily mean it's a viable option for Intel to do so. One glance at the price sheet for Xeon W 2400 confirms this:

​

​

Intel is asking $1389 for 16 P-core CPUs featuring Golden Cove.

Oh yeah, also so this is not viable even though it exists right?

 

[bit_user]

Oh yeah, also so this is not viable even though it exists right?

Do you see the Xeon W5-2465X penetrating the same market segments where the i9-13900K plays? I don't.

So, no. It doesn't appear viable for mainstream. What sells it as a workstation processor are the workstation platform features, like memory channels, capacity, and PCIe lanes.

 

[bit_user]

Then at least say area divided by efficiency ,power-efficiency for performance scaling , so getting as much performance as possible from as little power as possible is something completely different from 'we have to fit that much efficiency (performance/power) into that much size' .

And just because you add other stuff to it doesn't make it any more correct.

I'll decide what I am claiming, but thanks for the suggestions.

The point I'm defending is that, for roughly the same area, E-cores provide more performance and greater efficiency than P-cores.

Die sizes change all the time so the cores being bigger is a non argument.

Yes, die sizes across different manufacturers and process nodes aren't directly comparable, but you were effectively doing that by pointing to AMD and saying they offer a 16x P-core CPU so that proves it's viable for Intel to do with Golden Cove. We don't know exactly how much Intel or AMD's manufacturing costs are, but the fact that AMD's cores are smaller stands as a plausible explanation of why AMD managed to do it. We can't go any further down this path without data that we lack, but suffice to say it's a false-equivalence.

cost isn't something that would make this impossible.

Sorry, I don't agree. Selling a product with negative margins is unsustainable. I don't know how close they would be to doing that, but cost actually does matter. Investors are also keenly interested in things like gross margins and profitability, and you certainly don't help those metrics by doing things like selling mainstay products at-cost.

 

[TerryLaze]

I'll decide what I am claiming, but thanks for the suggestions.

No, your grammar and what you write decides what you are claiming, if they don't match what's going on in your head then it's not the readers fault.

The point I'm defending is that, for roughly the same area, E-cores provide more performance and greater efficiency than P-cores.

Yes, now you are.
Initially you separated these things with commas meaning that you made a separate claim for each one of these things.
You claimed that the e-cores are the most power-efficient. Full stop.
You claimed that the e-cores are the most area-efficient. Full stop.
You claimed that the e-cores are therefore the most cost-efficient. Full stop.
You missed a 'for a given area' there at the end and people wouldn't be arguing with you for so long.

Now that they've taken the plunge, Intel is never giving up hybrid CPUs because E-cores are the most power-efficient, area-efficient, and therefore cost-effective way to scale multi-threaded performance.

 

[TerryLaze]

Do you see the Xeon W5-2465X penetrating the same market segments where the i9-13900K plays? I don't.

So, no. It doesn't appear viable for mainstream. What sells it as a workstation processor are the workstation platform features, like memory channels, capacity, and PCIe lanes.

Does that make the w5 not exist anymore?
Nobody disagreed with that it's a cost thing and that it would never happen

Yes, die sizes across different manufacturers and process nodes aren't directly comparable, but you were effectively doing that by pointing to AMD and saying they offer a 16x P-core CPU so that proves it's viable for Intel to do with Golden Cove. We don't know exactly how much Intel or AMD's manufacturing costs are, but the fact that AMD's cores are smaller stands as a plausible explanation of why AMD managed to do it. We can't go any further down this path without data that we lack, but suffice to say it's a false-equivalence.

Let me guess, it's only a false-equivalence for my point because you know that bigger must mean more expensive even though intel makes cores at cost while AMD pays a 50% (whatever it is) margin to TSMC.
If wafer cost were an issue then no intel CPU would have an iGPU let alone all the super cheap formerly celeron/pentium ones.

Sorry, I don't agree. Selling a product with negative margins is unsustainable. I don't know how close they would be to doing that, but cost actually does matter. Investors are also keenly interested in things like gross margins and profitability, and you certainly don't help those metrics by doing things like selling mainstay products at-cost.

Exactly you (and I) don't know how much it would cost for intel and for how much they would be able to sell them....

 

[bit_user]

No, your grammar and what you write decides what you are claiming, if they don't match what's going on in your head then it's not the readers fault.

You're free to point out any inconsistencies or take issue with what I say. For sure, there could be a misunderstanding, which is why it's good to cite what you're talking about.

Initially you separated these things with commas meaning that you made a separate claim for each one of these things.

Exact quotes would be best, but I think we can work with your paraphrasing.

You claimed that the e-cores are the most power-efficient. Full stop.

Yes, and I stand by that.

What I'm not asserting is that they're more efficient by whatever definition of efficiency you pick. That's why I had to make the precise claim I made above - because you're trying to take them out of the context of how they're used.

I'm reminded of a video I once saw where a Toyota Prius was driven around a race track along with a sports car. The sports car normally had much worse fuel efficiency than the Prius, but actually had better efficiency around the track.

It's the same way with E-cores. If you insist on a naive approach of using the same clock speed for both P-cores and E-cores, until the E-cores are maxed out, then there's a window where the E-cores are less efficient on some workloads. That's why they don't clock them that way! CPU cores are sophisticated machines and not designed to be "idiot-proof". They're designed with the assumption that a reasonable frequency-scaling algorithm is being used.

So, if you're arguing that there's a way they can be used that makes them less efficient, then sure. However, that's a very artificial and pointless argument, because it's not how they're used in real life.

You claimed that the e-cores are the most area-efficient. Full stop.

I also stand by this. They're roughly 54% as fast as a single-threaded P-core on floating point, and about 64% as fast on integer. So, 4 of them easily outpace a single dual-threaded P-core.

Now, can you find some corner case where that doesn't hold? Maybe. Does Alder Lake have some scaling issues and a sub-optimal implementation of the E-cores (i.e. the ring bus frequency issue)? Sure. That's an indictment on their integration with Alder Lake, not the E-cores themselves.

You claimed that the e-cores are therefore the most cost-efficient. Full stop.

Area -> cost. So, yes.

 

[bit_user]

Let me guess, it's only a false-equivalence for my point because you know that bigger must mean more expensive

I said it's a plausible explanation. That's all.

even though intel makes cores at cost while AMD pays a 50% (whatever it is) margin to TSMC.

Source?

 

[thestryker]

I just want to put a pin in the area/silicon cost part of the discussion as for ADL Intel used the same die for everything with e-cores on desktop. Up to 10p with no e-cores would have led to a smaller die size it's only when you go above 10 that it gets bigger. What that says to me is that it was more cost effective to use the bigger die and harvest accordingly than it was to get more die per wafer. It also indicates that the e-core clusters were more likely to have manufacturing issues than the p-cores.

It wasn't until RPL that they started using more than one die on the desktop side for e-core parts. Also of note to me is that RPL massively skews efficiency in favor of e-cores as Intel was able to double the number of e-cores in the 13900K while only raising the PL2 12W and increased clocks across the board. While I think the area argument holds more weight than efficiency for ADL desktop for RPL the advantage of e-cores is universal.

Just as thought experiment based off of what measurements I was able to find a 12p ADL part would have led to ~12% fewer die/wafer assuming identical error rates than 8p/8e.

 

[bit_user]

What that says to me is that it was more cost effective to use the bigger die and harvest accordingly than it was to get more die per wafer.

This depends on yield. If yield was fairly low, then they could have enough dies with bad P-cores or E-core tiles that they could fill most of the lower bins with them, avoiding the need for another E-core die.

It also indicates that the e-core clusters were more likely to have manufacturing issues than the p-cores.

That requires you to know something about their sales volume. You can't determine that just by counting the number of SKUs with 4 E-cores and 8 P-cores. Plus, it's fairly "cheap" to disable a perfectly good E-core tile, for market segmentation purposes. Less wasted silicon than if you disable a pair of P-cores.

It wasn't until RPL that they started using more than one die on the desktop side for e-core parts.

By that, I presume you mean they rebadged the Alder Lake "big" dies to fill out some of their product stack? AFAIK, there was only one true Raptor Lake die. I don't know if they even made one for laptops.

 

[thestryker]

By that, I presume you mean they rebadged the Alder Lake "big" dies to fill out some of their product stack? AFAIK, there was only one true Raptor Lake die. I don't know if they even made one for laptops.

I may be wrong, but I thought there were two die made as Raptor Cove cores appeared all the way down in 6/4 configuration where the reused ADL chips play. That just seems like a lot to disable, but maybe that speaks to increased ability to get bad silicon working.

There are definitely multiple mobile die as anything with over 32 EUs is a completely different die. I believe the mobile die were capped at 6p/8e though.

This depends on yield. If yield was fairly low, then they could have enough dies with bad P-cores or E-core tiles that they could fill most of the lower bins with them, avoiding the need for another E-core die.


That requires you to know something about their sales volume. You can't determine that just by counting the number of SKUs with 4 E-cores and 8 P-cores. Plus, it's fairly "cheap" to disable a perfectly good E-core tile, for market segmentation purposes. Less wasted silicon than if you disable a pair of P-cores.

They had a lot more 8c SKUs than they did anything else with e-cores, though I suppose they could just axe a ton of working e-core clusters for the market segmentation since it's not like they could make a single cluster chip with the design they used.

 

[Ogotai]

If somebody tells you that a 16 full core CPU is not viable then bringing up a consumer CPU with 16 full cores isn't ok?!

for intel it isnt. why ? my guess, power usage, and cooling it. there has to be a reason why intel doesnt make a 16 core cpu that is labeled i9 ( or i7), instead they just added e cores to make up for it. to bring up a xeon workstation cpu, is well, invalid, its a work station cpu, NOT a consumer one, hence the price of the platform, and be fore you say it, Threadripper is also a workstation platform, and again, cause of its price. but you dont need threadripper to get a full 16 core cpu, as amd offers one for consumers.

Which is what AMD should be charging and would be charging if they weren't desperate, or at least it would be close to that, but because intel can match their 16 full core CPU performance with a $600 CPU that only has 8 full cores AMD can't charge much more.

most would consider intel the desperate one here. in order to compete with amd they had to come out with the p core/e core setup, cause if they didnt, intels power usage would be horrid, not to mention trying to keep that cpu cool.....

Oh yeah, also so this is not viable even though it exists right?

its not, why? for consumers, the price alone makes it not viable for consumers.

No, your grammar and what you write decides what you are claiming, if they don't match what's going on in your head then it's not the readers fault.

sorry terry, but the same can be said about you.

Does that make the w5 not exist anymore?
Nobody disagreed with that it's a cost thing and that it would never happen

for mainstream users, not it doesnt exist, cause of cost, same reason why threadripper doesnt.

ok let me ask you this then. IF intel were to release a 16 FULL core cpu, that compared to the 7950x ( or x3D version) provided the same performance across the board, with in 5% of the 7950x im every test ( win some, lost some ) BUT cost 1.5x ( rest of platform is the same price, only difference is the cpu price )or more the price of the 7950x, but also used quite a but more power, which one would you choose ?

 

[bit_user]

Initially you separated these things with commas meaning that you made a separate claim for each one of these things.

So, here's what I actually said:

Now that they've taken the plunge, Intel is never giving up hybrid CPUs because E-cores are the most power-efficient, area-efficient, and therefore cost-effective way to scale multi-threaded performance.

And now, I'm about to prove it!

So, I just extracted the datapoints from ChipsAndCheese' plots. Then, I wrote a little program to compute the optimal combination of clocks for E-cores and P-cores, at a given power level. Finally, I used those clocks to estimate the performance, which I compared against 10P + 0E and 12P + 0E (just for good measure). I estimated package power at 10 W, in all cases (perhaps it should even be a little higher, for the 12P case).

Here's data from the 7zip test case:

​

And here's the data from the x264 test case:

​

As you can clearly see, the 8P + 8E configuration beats 10P + 0E and even 12P + 0E at all power levels!! More performance per Watt is the very definition of efficiency!

As I said, adding E-cores truly is the most power-efficient, area-efficient, and therefore cost-effective way to scale multi-threaded performance.

Q.E.D.

 

[graham006]

We appreciate your enthusiasm, but you might want to re-check the video you mention to get your facts straight.
Because its the opposite... motherboard manufacturers giving way too much power beyond specs to the AMD chip.

No other motherboard can push Intel CPUs beyond spec? I've an Asus motherboard that can easily push my 13600k beyond spec. Hasn't burned up yet. I use just a 95mm Noctua air cooler.

 

[thestryker]

So, here's what I actually said:

​

​

​

And now, I'm about to prove it!

So, I just extracted the datapoints from ChipsAndCheese' plots. Then, I wrote a little program to compute the optimal combination of clocks for E-cores and P-cores, at a given power level. Finally, I used those clocks to estimate the performance, which I compared against 10P + 0E and 12P + 0E (just for good measure). I estimated package power at 10 W, in all cases (perhaps it should even be a little higher, for the 12P case).

Here's data from the 7zip test case:

​

​

​

And here's the data from the x264 test case:

​

​

​

As you can clearly see, the 8P + 8E configuration beats 10P + 0E and even 12P + 0E at all power levels!! More performance per Watt is the very definition of efficiency!

As I said, adding E-cores truly is the most power-efficient, area-efficient, and therefore cost-effective way to scale multi-threaded performance.

Q.E.D.

The problem here is that you cannot extrapolate the 12900K out of the 12700K due to the disproportionate change in base turbo (2.4 vs 2.7 and 3.2 vs 3.6) and peak turbo (3.9 vs 3.8 and 5.1 vs 4.9 @ 241W vs 190W). If you ran the numbers comparing 10p to 8p/4e I imagine it would look very similar to how the estimated 12p shows here in the x264 test.

I really wish chips and cheese had run the efficiency tests on both CPUs to see the difference. Just like how I wish they would do the same on the 13700K and 13900K since those both have the same peak turbo I'd love to see how the scaling and efficiency plays out.