News Intel Panther Lake processors could pack up to 16 cores, maximum of four performance cores according to leak

[bit_user]

1/ Then you have to synchronize even more.

Not in the way you're talking about. You have to ensure that certain operations are atomic, but this is different than the kind of "barrier" behavior that you were previously describing.

2/ If that was really the case, then instead of 8 fullsize + 6 limited size, you would rather choose 6 fullsize + 10 limited size, then rather than that you would choose 4 fullsize + 14 limited size.

The tradeoff isn't quite like that. I don't have time to dig up the relative core sizes, but the simple answer to your question of how many C-cores to pair with how many fullsize cores is based on the TDP of the CPU and where the all-core clocks of the fullsize cores crosses into the realm where the C-cores offer similar efficiency. We can absolutely work that out, but I don't have time right now.

The fact that this is not the architecture we observe, mean there is a difference, and that you would prefer to use a full size core rather than several limited size ones.
Hence, there are drawbacks using limited size cores.

The world is full of pragmatic compromises. We make these compromises to achieve a better outcome than if we stubbornly held to uncompromising positions.

The engineering world is ruled by such tradeoffs. What's optimal for one system might not be optimal for another. Numerous parameters have to be tuned, in order to deliver the optimal solution for a given set of constraints or requirements. The mix of core types is ultimately just another one of these tunables.

What makes it challenging is that the concurrency APIs in some modern operating systems haven't caught up (credit to Apple, here) and then applications have to actually use them. That still doesn't make E-cores or C-cores bad, in most cases, but rather just not as good as they could be.

 

[MacZ24]

What is more probable between these two scenarios :

1/
Efficiency cores, which are useful for laptops because it helps with battery usage, and are useful for servers because it helps with their heavy multithreaded workloads, also happen to be (miraculously) also very useful for desktops which have no efficiency requirement and are far less multithreaded than servers.

OR

2/
Yes, let's cram as much power through these 8 cores (the amount required today for gaming) as possible, so that we have good gaming performances and with the power and die area left, let's put some efficiency cores so that we also have a higher core count than the competition ? The marketing will find some way to explain to the suckers it's like the second coming of Jesus or something.

We still live in free countries and you can still believe in Santa if you want.

 

[bit_user]

Funny thing according to TPU and using your own comparison logic the 9950x is less efficient than the 5950x. 4 years later and amds latest and greatest is less efficient than a 2020 CPU. Talk about stagnation right?

LOL, you're missing the forest... because we went down the road an there's now a hill in the way. The forest is still there, it's just not visible. ...or something like that.

Analogies aside, what happened is that AM5 supports much greater power delivery than AM4 did. The main reason the 5950X was so efficient is that its cores were given only a limited amount of power that kept them running at much more efficient clock speeds. With the enhanced power & cooling requirements of AM5, a 16-core CPU can give each core much more power, which lets them run at less efficient clock speeds.

So, it's not that Zen 5 is less efficient than Zen 3. It's just that AM4 was holding back the 5950X to a greater extent than AM5 does with the 9950X. The efficiency gains which occurred since then haven't fully cancelled out that discrepancy.

We could see the same thing if you disable E-cores and compare the 8 P-cores of a Raptor Lake @ stock power to an older 8-core CPU, like the i9-9900K. Both are 8-core CPUs, but one with much higher power delivery that's capable of driving it well past the point where it's less-efficient. This doesn't tell us Intel made no progress, but it does tell us something about how they & their target customers regard efficiency.

 

[TheHerald]

LOL, you're missing the forest... because we went down the road an there's now a hill in the way. The forest is still there, it's just not visible. ...or something like that.

Analogies aside, what happened is that AM5 supports much greater power delivery than AM4 did. The main reason the 5950X was so efficient is that its cores were given only a limited amount of power that kept them running at much more efficient clock speeds. With the enhanced power & cooling requirements of AM5, a 16-core CPU can give each core much more power, which lets them run at less efficient clock speeds.

So, it's not that Zen 5 is less efficient than Zen 4. It's just that AM4 was holding back the 5950X to a greater extent than AM5 does with the 9950X. The efficiency gains which occurred since then haven't fully cancelled out that discrepancy.

We could see the same thing if you disable E-cores and compare the 8 P-cores of a Raptor Lake @ stock power to an older 8-core CPU, like the i9-9900K. Both are 8-core CPUs, but one with much higher power delivery that's capable of driving it well past the point where it's less-efficient. This doesn't tell us Intel made no progress, but it does tell us something about how they & their target customers regard efficiency.

My point is I'm using same logic to conclude that the 9950x is more efficient than the 5950x as I'm using to conclude that the 13700k is more efficient than the 7700x. I don't care about power limits, stock settings and none of that in both cases. I just look which one performs more while using the same power.

It's you who's moving the goal posts to favor amd now here you are claiming that the 9950x is more efficient than the 5950x yet the 13700 isn't more efficient than the 7700x. Makes this make sense....

Everything you just described applies to the 13700k. Intel allowed more power than amd did to the 7700x which makes it run at less efficient clocks etc. Yet you insist the 13700 isn't more efficient but the 9950x is......

 

[bit_user]

My point is I'm using same logic to conclude that the 9950x is more efficient than the 5950x as I'm using to conclude that the 13700k is more efficient than the 7700x.

Oh, but I just pointed out the technical reason. I didn't turn around and try to say that the 9950X is actually the more efficient product!

That's where we differ. I like to understand the "why", but I understand that doesn't change how a user will actually experience a product.

I don't care about power limits, stock settings and none of that in both cases.

Right. You just want to make an artificial experiment that shows Intel looking better.

It's you who's moving the goal posts to favor amd now here you are claiming that the 9950x is more efficient than the 5950x

Nope. I just explained why the 9950X is less efficient.

Yet you insist ... the 9950x is...... (more efficient)

I literally did not.

Again, with the mind games... trying to trick people into agreeing with you.
🙄

 

[TheHerald]

Oh, but I just pointed out the technical reason. I didn't turn around and try to say that the 9950X is actually the more efficient product!

That's where we differ. I like to understand the "why", but I understand that doesn't change how a user will actually experience a product.


Right. You just want to make an artificial experiment that shows Intel looking better.


Nope. I just explained why the 9950X is less efficient.


I literally did not.

So the 9950x, amds latest and greatest is less efficient than a 2020 cpu. Okay, as I've said before stagnation.

 

[bit_user]

So the 9950x, amds latest and greatest is less efficient than a 2020 cpu. Okay, as I've said before stagnation.

Say what you like about their product marketing team and how they prioritize performance over efficiency, as they're ultimately the ones that tell the engineers what power envelope to work with. All legitimate complaints.

The same can be said of Intel.

Just don't try to confuse engineering with product marketing.

 

[TheHerald]

Say what you like about their product marketing team and how they prioritize performance over efficiency, as they're ultimately the ones that tell the engineers what power envelope to work with. All legitimate complaints.

The same can be said of Intel.

Just don't try to confuse engineering with product marketing.

Quite the contrary, you are confusing engineering with product marketing. When it comes to engineering, obviously the 9950x is more efficient than the 5950x by a ton. The same ton the 13700 more efficient than the 7700.

 

[bit_user]

Quite the contrary, you are confusing engineering with product marketing.

To be clear, what I mean is the people at AMD and Intel who define the product specifications - things like the power envelope. The engineers have to build products which conform to the requirements. So, when they're told "build the fastest CPU possible that fits in a 230 W envelope and uses X mm^2 of TSMC N# die area", that's what they do.

Not to say it's a straight waterfall process. For instance, it could be the case that they made the Zen 5 CCD, and then PM later decided to sell the 9700X as a 65W CPU after seeing the performance data.

My main point is to separate architecture from how the specific products are defined. So, I can look at Zen 5's perf/W curve and compare it to Zen 4's perf/W curve, but whatever that looks like is mostly a technical curiosity and doesn't change how the final products behave.

When it comes to engineering, obviously the 9950x is more efficient than the 5950x by a ton.

Just as long as we're clear: that's what you're saying.

I didn't say that. I've said pretty much everything I have to say about that matter.

 

[YSCCC]

First of all any such comparison is meaningless because intel chips have 2 different types of cores. But since you care about single thread vs single thread, surely you realize that eg. the 13900k is much faster and consumes much less power than the 7950x 1 core vs 1 core. Right?


You don't have to limit anything but then you arent really comparing the cpus but the settings. If you don't limit them to same wattage then intel wins by default with the T series being at 35w. That's just silly.


Let me ask you a question. Please, don't avoid answering it. Let's for the sake of argument agree that Intel chips are super inefficient and hopeless and they don't stand a chance and the whole 9 yards.

If intel for whatever reason decided to lock the 13700k to 125w instead of the unlimited / 253w it runs now, would that mean that they are more efficient than their amd counterpart (the 7700x)? So the only thing that makes then not efficient is the power intel decided to lock them at?

Wrong again, say referencing tech power up, single threaded power efficiency chart, measuring via the power cable and not cpu consumption you can see 7600x leads the pack, 7700x ties with 13700k and 13900k wins 7950x, then at MT, with same thread count zen 4 wins, what this means is Intel only wins in efficiency when in c state or idle, where the redundant cores aren’t ramping up, which is known fact that AMD architecture drinks more juice than Intel at idle. It’s the system’s extra cores unused affecting the top end chips in ST, when the extra chip let isn’t there as in 7600x, under full load they did win by a lot.

and by the way, Intel obviously thinks they lost in efficiency or I can’t see why they won’t market any of their rpl parts to be efficiency king with your style of artificial power limitation usage. And they know that if they do so some courts and lawyers are going to find them.

 

[YSCCC]

To be clear, what I mean is the people at AMD and Intel who define the product specifications - things like the power envelope. The engineers have to build products which conform to the requirements. So, when they're told "build the fastest CPU possible that fits in a 230 W envelope and uses X mm^2 of TSMC N# die area", that's what they do.

Not to say it's a straight waterfall process. For instance, it could be the case that they made the Zen 5 CCD, and then PM later decided to sell the 9700X as a 65W CPU after seeing the performance data.

My main point is to separate architecture from how the specific products are defined. So, I can look at Zen 5's perf/W curve and compare it to Zen 4's perf/W curve, but whatever that looks like is mostly a technical curiosity and doesn't change how the final products behave.

I think that every VF curve need to see how the whole package is designed to perform to be meaningful, it’s as useless for desktop highend chips at a stupidly below stock power limit point, nor comparing efficiency for extreme OCed mobile parts. The parts and architectures are built to perform at certain level and optimising the power consumption at that range for the “sweet spot”, using the chips at unintended power level is like trying to bully for the sake of arguement and against what the engineer thinks

 

[TheHerald]

Wrong again, say referencing tech power up, single threaded power efficiency chart, measuring via the power cable and not cpu consumption you can see 7600x leads the pack, 7700x ties with 13700k and 13900k wins 7950x, then at MT, with same thread count zen 4 wins, what this means is Intel only wins in efficiency when in c state or idle, where the redundant cores aren’t ramping up, which is known fact that AMD architecture drinks more juice than Intel at idle. It’s the system’s extra cores unused affecting the top end chips in ST, when the extra chip let isn’t there as in 7600x, under full load they did win by a lot.

and by the way, Intel obviously thinks they lost in efficiency or I can’t see why they won’t market any of their rpl parts to be efficiency king with your style of artificial power limitation usage. And they know that if they do so some courts and lawyers are going to find them.

When you realize that the CPU with the higher power limit will usually be less efficient cause it's pushed beyond the sweetspot then we might have a meaningful discussion. Until then, the intel T series are the kings of efficiency, are they not?

 

[YSCCC]

When you realize that the CPU with the higher power limit will usually be less efficient cause it's pushed beyond the sweetspot then we might have a meaningful discussion. Until then, the intel T series are the kings of efficiency, are they not?

Seems not, 14900T have Pl2 of 106W, Ryzen 9 7900 have power draw reported to be ~90W, which is lower, and for passmark test

https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i9-14900T&id=5903

AMD Ryzen 9 7900 Benchmark

www.cpubenchmark.net

I don't see it is remotely king of efficiency

 

[TheHerald]

Seems not, 14900T have Pl2 of 106W, Ryzen 9 7900 have power draw reported to be ~90W, which is lower, and for passmark test

https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i9-14900T&id=5903

AMD Ryzen 9 7900 Benchmark

www.cpubenchmark.net

I don't see it is remotely king of efficiency

Do you understand what PL2 is?

 

[TheHerald]

That's not accurate, though quite a popular misconception. Tau is indeed a time constant, with units rated in seconds. However, it determines the aspect ratio of an exponential-weighted moving average filter, which is used to smooth out the instantaneous power draw.

PL1 is only enforced when this "average" power exceeds PL1. How long that takes depends entirely on how far above PL1 you go. If you go only a little above PL1, it could boost for quite a bit longer than 54 seconds. I can't say exactly how long, because Intel hasn't published the precise formula they use. However, I have observed this in practice, so I know my understanding of their documentation is accurate.

It's called moving window average or something like that, I know. For heavy MT workload PL2 will last 54 seconds because the CPU will max out the PL2. Say something like CBR23.

One of the few reviews that actually tested these CPUs

https://hardwareand.co/dossiers/cpu...-et-16-microarchitectures-differentes?start=6

The 13700 (non k) topped the entire chart in h264 efficiency (from the wall). It's obvious from the above numbers all the non k and T alderlake and raptorlake would be sitting at top or near the top efficiency. Not a single AMD chip would be on the top 10 of efficiency. Maybe the 7900 (non x) could break the top 10, but still, 1 amd cpu on the top 10 most efficient ones, and here we are arguing that amd is more efficient. Doesn't it seem crazy to you?

 

[YSCCC]

Do you understand what PL2 is?

It tops the power usage in the timeframe, which, is what it would run into in the designated time during the test, same goes for the ryzen, but it only tops at ~90W

Those tests don't last an hour, so the PL2 limit does contribute a lot to how it performs

 

[TheHerald]

It tops the power usage in the timeframe, which, is what it would run into in the designated time during the test, same goes for the ryzen, but it only tops at ~90W

Those tests don't last an hour, so the PL2 limit does contribute a lot to how it performs

Exactly, that's why it's not a great idea to test for 1 minute. Does anyone that care about MT runs workloads for less than a minute?

Also, if they were both running at 90 and 106 respectively then obviously the 13900 would be faster. Are you suggesting that a 7900 at 90w is faster than 13900 at 106? That's crazy mind you. TPU showed that at 95w a 14900k ~= 7900x at 200w.

But even if we pretend your numbers are correct, you have 1 CPU from AMD and about 15 from intel (non k and T) that are topping the efficiency chart. AMD only has 1 high core count low power cpu (the 7900 you mentioned) everything else will be suffering compared to the non k and T.

 

[TheHerald]

This situation reminds me of the a12x25 vs T30 fan. People were going crazy about how the noctua a12x25 is better engineered blablaba and the T30 is just thicker and that's why it outperforms it yadayada. And that's absolutely true, but I don't care, the end result is what matters. In the same way, sure the 7700x is probably better engineered (just like the a12x25) but the 13700 just has more cores (thicker). The end result is it can achieve better performance while using the same power, just like the T30 can achieve better temperatures at same noise levels.

Now granted, thank god people aren't comparing fans while having them all blast at 100%, cause then obviously the fan with the highest max rpm would look like it's the worst, even though it's literally the best. That's the case with the T30, if you put every fan to 100%, the T30 would end up the worst in efficiency (performance / noise). If you actually normalize for noise, it's at the top of the chart, nothing gets close to it. It's exactly the same with the 7700x vs 13700, but proper testing methods haven't reach the CPU crowd yet, so one can only hope.

At least, and quote me on it, I'm not changing my testing methods based on whos winning and whos losing like some people do and will do in the future. Im always comparing at ISO wattage and will continue to do so regardless of who is in the lead.

 

[YSCCC]

Exactly, that's why it's not a great idea to test for 1 minute. Does anyone that care about MT runs workloads for less than a minute?

Also, if they were both running at 90 and 106 respectively then obviously the 13900 would be faster. Are you suggesting that a 7900 at 90w is faster than 13900 at 106? That's crazy mind you. TPU showed that at 95w a 14900k ~= 7900x at 200w.

But even if we pretend your numbers are correct, you have 1 CPU from AMD and about 15 from intel (non k and T) that are topping the efficiency chart. AMD only has 1 high core count low power cpu (the 7900 you mentioned) everything else will be suffering compared to the non k and T.

Ahha, goal post moved again, now you are comparing the max power draw of the 7900 vs the nerfed power draw of the 219W PL2 13900 and not the 14900T at 106W, that's SOOOOO consistant, thanks Chief.

 

[TheHerald]

Ahha, goal post moved again, now you are comparing the max power draw of the 7900 vs the nerfed power draw of the 219W PL2 13900 and not the 14900T at 106W, that's SOOOOO consistant, thanks Chief.

No idea what you are talking about, I haven't moved any goalposts. You are making stuff up. It's okay, it's expected.

 

[bit_user]

It's called moving window average or something like that, I know.

It's a window function, in the mathematical sense. That makes it sound like a box filter, but it's certainly an IIR.

https://tttapa.github.io/Pages/Math...oving Average/Exponential-Moving-Average.html
​

The way to understand this plot is that it's the weight applied to a given sample at successive time intervals. As time goes by, older samples contribute less and less to the running average. IIRs turn out to be inexpensive to compute, requiring only two multiplies and one addition per iteration. Also, because the rate of heat transfer is proportional to the temperature delta, the exponential drop-off of an IIR is well-suited to modelling it.

https://hardwareand.co/dossiers/cpu...-et-16-microarchitectures-differentes?start=6

The 13700 (non k) topped the entire chart in h264 efficiency (from the wall). ... Not a single AMD chip would be on the top 10 of efficiency. Maybe the 7900 (non x) could break the top 10,

The 7950X3D is the second most efficient they tested. 8700G is fourth best.

I'm not sure how much to read into it, but the results for the 12 V rail tell a different story. In that case, the 7800X3D rules supreme!

Also, it is just one test. I'm sure the reason they didn't run more tests was due to the time & effort, not because one test is all you need.

P.S. I do think it's cute that you've got us talking about Zen 4, when the 9700X is almost as efficient as the 7800X3D.

​

 

[TheHerald]

It's a window function, in the mathematical sense. That makes it sound like a box filter, but it's certainly an IIR.

​

​

https://tttapa.github.io/Pages/Mathematics/Systems-and-Control-Theory/Digital-filters/Exponential Moving Average/Exponential-Moving-Average.html​

​

The way to understand this plot is that it's the weight applied to a given sample at successive time intervals. As time goes by, older samples contribute less and less to the running average. IIRs turn out to be inexpensive to compute, requiring only two multiplies and one addition per iteration. Also, because the rate of heat transfer is proportional to the temperature delta, the exponential drop-off of an IIR is well-suited to modelling it.


The 7950X3D is the second most efficient they tested. 8700G is fourth best.

I'm not sure how much to read into it, but the results for the 12 V rail tell a different story. In that case, the 7800X3D rules supreme!

Also, it is just one test. I'm sure the reason they didn't run more tests was due to the time & effort, not because one test is all you need.

P.S. I do think it's cute that you've got us talking about Zen 4, when the 9700X is almost as efficient as the 7800X3D.

​

​

The 7950x 3d is the 2nd most efficient cause they have only tested one non k or t. If they tested all of them, nothing on amd would have made it on the top 10. That much is obvious but let's pretend it's not the case.

I'm not talking about zen 5 cause it's kinda irrelevant. When you don't accept that rpl is more efficient than Zen 4 obviously you won't accept its more efficient than Zen 5 either.

Wall is what matters but I don't admit going with 12v either. Even with 12v not a single amd would make it on the top 10 if they included more non k cpus. Don't you agree with that?

 

[YSCCC]

The 7950x 3d is the 2nd most efficient cause they have only tested one non k or t. If they tested all of them, nothing on amd would have made it on the top 10. That much is obvious but let's pretend it's not the case.

I'm not talking about zen 5 cause it's kinda irrelevant. When you don't accept that rpl is more efficient than Zen 4 obviously you won't accept its more efficient than Zen 5 either.

Wall is what matters but I don't admit going with 12v either. Even with 12v not a single amd would make it on the top 10 if they included more non k cpus. Don't you agree with that?

If... If they've tested a K6... O wait, if they've tested a 386.. Or if they tested a threadripper... it's all if, no proof

 

[TheHerald]

If... If they've tested a K6... O wait, if they've tested a 386.. Or if they tested a threadripper... it's all if, no proof

The 13700 is already the most efficient on their chart bud. Isn't that proof?

 

[YSCCC]

The 13700 is already the most efficient on their chart bud. Isn't that proof?

From the chart of your fav Cinebench MT the 7900 non x tops it, more than double of 13700 bud