Discussion Intel current and future Lakes & Rapids thread

[DrMrLordX]

The 5950X will use 250+ watts too if you want to extract the last 10% of all-core performance.

Not without PBO or static OC it won't. You can put it on water and it'll stop @ 142W anyway.

 

[RTX2080]

It seems that there is no N+0 SKUs for K series for initial release, it means that if you want an unlock processor, you have to buy these small cores.(facepalm)
I hope this list is wrong though.

 

[TESKATLIPOKA]

Honestly, I don't care that much about desktop models, mobile in my opinion is more interesting.
I always wondered why Intel stopped at 2+8 config for 9 and 15W TDP. I always thought 4+8 config would be better at least for 15W, but that may not be necessarily true.

For example Core i7-1185G7 4C8T has

Base Frequency at 12 W	1200 MHz
Base Frequency at 15 W	1800 MHz
Base Frequency at 28 W	3000 MHz

Let's say at full load 10W is used only by cores from 12W TDP, that means 2.5W per core including HT and 1.2Ghz base frequency. Actually It should be less, If we consider a 3W higher TDP increases the base frequency by 50%, but let's ignore It, It's not that important.

Now let's apply It to Alder Lake, and It's 2+8 cores, which would mean 1W per core(10W for all cores) at 12W TDP.
I don't think Golden Cove core using 1W can clock at 1GHz, I think 800Mhz should be more realistic.
Now the question is how high can Gracemont clock using only 1W.
I think 1.2Ghz is not unrealistic, and that's 50% more than Golden Cove!
What would It mean for performance?
Let's say Golden Cove has 50% more IPC and HT brings another 20% performance. So Golden Cove would have 80% higher throughput per core than Gracemont at the same frequency.
Golden Cove 100*1,5*1,2 = 180 points
On the other hand Gracemont can clock 50% higher.
Gracemont 100*1.5 = 150 points
The difference in now only 20%.
Maybe the IPC difference is only 40% and Gracemont can clock to 1300Mhz, that would mean 1C2T Golden Cove performs as 1C1T Gracemont.

 

[Accord99]

Not without PBO or static OC it won't. You can put it on water and it'll stop @ 142W anyway.

That's what I meant with getting the last bit of performance. I'm sure there'll be OEM models that hard-limit the 12900 to TDP too.

The question is does it take 250W for Alder Lake to beat the 142W 5950X (in multi-core performance) or can it also beat the 250W 5950X.

 

[DrMrLordX]

That's what I meant with getting the last bit of performance. I'm sure there'll be OEM models that hard-limit the 12900 to TDP too.

There probably will be some checks put in place by OEMs, but if you get a processor-in-a-box and put it in a DiY motherboard, you'll have to plug those limits in by yourself in the UEFI.

The question is does it take 250W for Alder Lake to beat the 142W 5950X (in multi-core performance) or can it also beat the 250W 5950X.

At this point it's all up in the air. Raichu has been a pretty reliable source of information, and the implication is that it requires watercooling to beat a stock 5950X, but we don't know for sure.

 

[TESKATLIPOKA]

Honestly, I don't care that much about desktop models, mobile in my opinion is more interesting.
I always wondered why Intel stopped at 2+8 config for 9 and 15W TDP. I always thought 4+8 config would be better at least for 15W, but that may not be necessarily true.

For example Core i7-1185G7 4C8T has

Base Frequency at 12 W	1200 MHz
Base Frequency at 15 W	1800 MHz
Base Frequency at 28 W	3000 MHz

Let's say at full load 10W is used only by cores from 12W TDP, that means 2.5W per core including HT and 1.2Ghz base frequency. Actually It should be less, If we consider a 3W higher TDP increases the base frequency by 50%, but let's ignore It, It's not that important.

Now let's apply It to Alder Lake, and It's 2+8 cores, which would mean 1W per core(10W for all cores) at 12W TDP.
I don't think Golden Cove core using 1W can clock at 1GHz, I think 800Mhz should be more realistic.
Now the question is how high can Gracemont clock using only 1W.
I think 1.2Ghz is not unrealistic, and that's 50% more than Golden Cove!
What would It mean for performance?
Let's say Golden Cove has 50% more IPC and HT brings another 20% performance. So Golden Cove would have 80% higher throughput per core than Gracemont at the same frequency.
Golden Cove 100*1,5*1,2 = 180 points
On the other hand Gracemont can clock 50% higher.
Gracemont 100*1.5 = 150 points
The difference in now only 20%.
Maybe the IPC difference is only 40% and Gracemont can clock to 1300Mhz, that would mean 1C2T Golden Cove performs as 1C1T Gracemont.

Let's continue. What about performance at full load compared to above-mentioned Tiger Lake with 12W TDP, where 10W is divided among only 4 cores.
Tiger Lake core having +25% more IPC, +20% thanks to HT and same clocks as Gracemont -> 100*1.25*1.2*1.5 = 225 points
Yes, Tiger Lake ended up 25% more powerful than even Golden Cove, but that's because a bigger power budget was allocated to each core.

Performance(throughput) of the whole chip:
Alder Lake 2C4T+8C8T -> 2*180+8*150 = 1560 points
Tiger Lake 4C8T -> 4*225 = 900 points

Alder Lake has 73% higher performance within the same limited TDP. Looking good in my opinion.

 

[Asterox]

That's what I meant with getting the last bit of performance. I'm sure there'll be OEM models that hard-limit the 12900 to TDP too.

The question is does it take 250W for Alder Lake to beat the 142W 5950X (in multi-core performance) or can it also beat the 250W 5950X.

Well, this is just Aida64 CPU stres test.But nonetheless, from this we can extract good information.That was probably stock i9 Alder Lake K, but i would be very bad if it was stock i7 Alder Lake K model.

For comparison i did short test, Aida64 FPU only and all three available CPU combo options.

CPU Package Power, R5 4650G

- AIDA64 FPU, up to 85-87W
- Aida64 3u1, 75W
- Cinebench R20, up to 75W







Stock Alder Lake i9 compared to stock R9 5950X, well in the same conditions(GCN) it will be around 230-250W for Alder Lake.100W+ for what, 10% higher MT score is huh or lol cpu power efficiency.

 

[Zucker2k]

Well, this is just Aida64 CPU stres test.But nonetheless, from this we can extract good information.That was probably stock i9 Alder Lake K, but i would be very bad if it was stock i7 Alder Lake K model.

For comparison i did short test, Aida64 FPU only and all three available CPU combo options.

CPU Package Power, R5 4650G

- AIDA64 FPU, up to 85-87W
- Aida64 3u1, 75W
- Cinebench R20, up to 75W

View attachment 49503

View attachment 49504

View attachment 49505

Stock Alder Lake i9 compared to stock R9 5950X, well in the same conditions(GCN) it will be around 230-250W for Alder Lake.100W+ for what, 10% higher MT score is huh or lol cpu power efficiency.

View attachment 49506

If you're this busy already comparing your favourite manufacturer chips to an unreleased chip how are we ever going to survive your full onslaught once ADL actually gets released? Are all forum rules out the window in this particular thread? #StopTheSpamming

 

[eek2121]

If you're this busy already comparing your favourite manufacturer chips to an unreleased chip how are we ever going to survive your full onslaught once ADL actually gets released? Are all forum rules out the window in this particular thread? #StopTheSpamming

Don't even get me started on the fact that he's comparing what is essentially a mobile SKU (cut down Zen 2) with a SKU targeting at desktop/high end. I'm willing to bet that Gracemont has greater single threaded IPC (and perf/watt) than any Ryzen 4xxx g/h/u chip.

 

[IntelUser2000]

Let's say at full load 10W is used only by cores from 12W TDP, that means 2.5W per core including HT and 1.2Ghz base frequency. Actually It should be less, If we consider a 3W higher TDP increases the base frequency by 50%, but let's ignore It, It's not that important.

It shouldn't matter to your analysis but I know the 6W versions of Goldmont Plus/Refresh/Tremont can sustain 2GHz frequencies under Cinebench MT workload. Reviews show the Goldmont Plus devices maintaining 1.6GHz CPU and 4-500MHz iGPU when both are simultaneously stressed under Prime95 and Furmark. At 10W and CPU-only it can sometimes reach 3GHz.

I wouldn't be surprised if Gracemont does the same.

I think Cove cores are capable of higher frequencies than 800MHz with 1W per core. The efficiency is lost quite quickly as you ramp up the clocks.

 

[jpiniero]

It seems that there is no N+0 SKUs for K series for initial release, it means that if you want an unlock processor, you have to buy these small cores.(facepalm)
I hope this list is wrong though.

There might be some models with no small but those will be locked.

 

[Exist50]

There might be some models with no small but those will be locked.

It's been pretty thoroughly leaked that there's a 6+0 desktop die. Surprised they didn't make the whole i5 lineup out of it, tbh. Certainly all the locked i5s will be.

 

[jpiniero]

It's been pretty thoroughly leaked that there's a 6+0 desktop die. Surprised they didn't make the whole i5 lineup out of it, tbh. Certainly all the locked i5s will be.

There isn't any downside to including small cores so you may as well include it if yields allow. If that ends up being only the 12600K for i5 consider it an extra perk.

I wouldn't be surprised if there was an 8+0 model. That might be Xeon E only.

 

[CakeMonster]

If the small cores work as well as we hope, wouldn't it be a downside to not have any enabled in a CPU for the consumer segment?

 

[jpiniero]

If the small cores work as well as we hope, wouldn't it be a downside to not have any enabled in a CPU for the consumer segment?

That's what I am saying. Being 10 nm there's going to be plenty of busted small cores.

 

[Hulk]

You are basically saying “It barely beats the 5950X so it must be bad.” You are wrong. What we are about to see is an intriguing shift in computing where even the “lowly” 12600k (which will likely retail for under $300) has more than 8 cores.

AMD will have to step it up in order to compete. That is a good thing.

No. I'm saying I don't trust leaked scores.

 

[TESKATLIPOKA]

It shouldn't matter to your analysis but I know the 6W versions of Goldmont Plus/Refresh/Tremont can sustain 2GHz frequencies under Cinebench MT workload. Reviews show the Goldmont Plus devices maintaining 1.6GHz CPU and 4-500MHz iGPU when both are simultaneously stressed under Prime95 and Furmark. At 10W and CPU-only it can sometimes reach 3GHz.

I wouldn't be surprised if Gracemont does the same.

I think Cove cores are capable of higher frequencies than 800MHz with 1W per core. The efficiency is lost quite quickly as you ramp up the clocks.

Thanks for the info. Personally, I would have used clocks sustainable during Cinebench for example, but I didn't know them, so I used possible base clocks instead.
Cove can most likely sustain more than 800Mhz with 1W power budget for example in Cinebench.

 

[Abwx]

Cove can most likely sustain more than 800Mhz with 1W power budget for example in Cinebench.

It s more like 0.5W@800MHz, something like 15W at 4GHz and 25W@5Ghz.

 

[TESKATLIPOKA]

Do you mean in Cinebench?

 

[Abwx]

Do you mean in Cinebench?

Yes, 1C/2T.

 

[dullard]

It seems that there is no N+0 SKUs for K series for initial release, it means that if you want an unlock processor, you have to buy these small cores.(facepalm)
I hope this list is wrong though.

Why the facepalm, do you not like improved multithreaded performance?

 

[dullard]

Also I've been thinking about Intel's claim that Golden Cove will be the P core for the next decade. I'm thinking they mean that as software transitions more and more to MT and therefore ST performance becomes less important, Golden Cove (and future derivatives) might "see them through" the Big core era. Perhaps the future is lots of efficiency cores and Intel is predicting that will occur in the next 10 years?

1) Lots of software cannot transition to more MT. That is because many tasks simply are not possible to be multithreaded. Any task that is user facing must go down to one thread at some point. Having dozens of threads sitting around doing nothing waiting for the mouse to move does not improve performance, only one thread is needed for that. Also, only one thread can display on the screen (the UI thread). Having drawing calls from multiple different threads is a guaranteed way to crash the software. As for math problems, many calculations rely on the result of the previous calculation and thus must be ST. Sure, some tasks can be MT, but many cannot.

2) Think about the number of cores in the next 10 years. Rumors have it that Intel is already planning on 40 cores in ~3 to ~4 years (Arrow Lake). Rocket lake in 125 W over 8 cores gets at the very most 15.6 W per core (actually a bit less since the uncore takes some power). But with 40 cores, that translates to an absolute max of 3.1 W per core. You just cannot have the high speed that you have with 15 W at your disposal when you now only have 3 W. The only possible way forward is to keep adding power or to switch to efficiency cores.

3) Combine #1 and #2 and you get Intel's strategy. Some tasks must be single threaded, so you might as well have a couple high power cores for the tasks that must be single threaded. But, math is forcing you to less and less power per core, so you might as well use cores that are efficient for those tasks.

 

[coercitiv]

Why the facepalm, do you not like improved multithreaded performance?

Reminds of a now famous Blizzard employee quote: "Do you guys not have phones?".

There's a number of forum users here that consider the MT performance of the hybrid solution will come with a price in performance consistency. As long as 6+0 or 8+0 SKUs are available, they would rather avoid paying for the E-cores considering their workloads fit P-cores better. Some would obviously like their chips unlocked to push P-cores further.

 

[Hitman928]

Reminds of a now famous Blizzard employee quote: "Do you guys not have phones?".

There's a number of forum users here that consider the MT performance of the hybrid solution will come with a price in performance consistency. As long as 6+0 or 8+0 SKUs are available, they would rather avoid paying for the E-cores considering their workloads fit P-cores better. Some would obviously like their chips unlocked to push P-cores further.

It's also just a matter of scaling. When I upgraded to Zen 3, I could have purchased a 5950x, but chose the 5900x because what I use it for, at least for the foreseeable future, the extra 4 cores / 8 threads wouldn't have really benefited me much, if at all. If your work load only scales to 6 - 12 threads and you care mostly about top performance, then the E-cores don't really have a place for you over a full P-core die. For others, 2-4 P-cores and as many E-cores as they can fit may be the preferred solution, but that won't be the typical desktop crowd.

 

[JoeRambo]

There's a number of forum users here that consider the MT performance of the hybrid solution will come with a price in performance consistency. As long as 6+0 or 8+0 SKUs are available, they would rather avoid paying for the E-cores considering their workloads fit P-cores better. Some would obviously like their chips unlocked to push P-cores further.

I am one of those users. Actually i don't mind the small cores, as long as they bring with them 2.5MB of L3 per slice. 5MB of extra L3 is good deal for me versus 25MB of L3 for hypothetical 8+0 CPU.
I am now running 10900K with disabled HT and static OC to 5.1Ghz and plenty happy with it as main desktop and gaming machine. As long as Golden Cove has 25% IPC advance over Skylake and i can get something like DDR5 6400CL36 or so i will be happy to disable small cores and HT, set clock to 5-5.1ghz and enjoy real smooth and responsive system.

My dream is simple: 2000 in GB5 ST without VAES style shenanigans would be advance of 33% for me. And i have very little faith in schedulers even in easier setting of HT, heterogenous setup is disaster waiting to bite.

1) Lots of software cannot transition to more MT. That is because many tasks simply are not possible to be multithreaded. Any task that is user facing must go down to one thread at some point. Having dozens of threads sitting around doing nothing waiting for the mouse to move does not improve performance, only one thread is needed for that. Also, only one thread can display on the screen (the UI thread). Having drawing calls from multiple different threads is a guaranteed way to crash the software. As for math problems, many calculations rely on the result of the previous calculation and thus must be ST. Sure, some tasks can be MT, but many cannot.

I think ~10 years ago i had hilarious MT problem with one of our servers, that after upgrade to 2S Sandy Bridge started having nasty periodic slowdowns running exactly same workload versus Core2 FSB based server. After a looooooong investigation involving a lot of digging, it turned out that culprit was periodic scheduled invocation of ImageMagic command line utility to do some misc stuff. Said command was using multithreading on all CPUs to "speed up" things, except once number of CPU's has risen and locking became contended, 99.99% of time started being spent on lock contention and cache line pingpongs betwen CPUs and over inter socket QPI links. And that created NASTY slowdown on whole system, destroying QoL of our service big time.
It took quite some time to investigate and notice patterns and pin the command to single CPU to fix it.

So not everything can be multithreaded and sometimes you don't even control the quality of multithreading in software involved.

I think with heterogenous cores with such wide performance gap, scheduling will steal part of IPC increase from big cores and create problems with various legacy apps. Not complaining much as long as small cores can be disabled tho.