Discussion Intel current and future Lakes & Rapids thread

[mikk]

Only a few tests rely on AVX512, this is clearly visible on the comparisons, besides Sandra has been quite accurate for comparisons within a same brand, assuming you take account of the instructions used on thoses tests.

Almost all rely on AVX512, this is clearly visible from the test. RKL-S is on par or faster than ADL-S and RKL-S usually isn't faster than 10C CML-S in MT. There is a huge AVX512 speedup versus AVX2, that's why.

We saw in the “RocketLake” review (Intel 11th Gen Core RocketLake AVX512 Performance Improvement vs AVX2/FMA3) that AVX512 makes RKL almost 40% faster vs. AVX2/FMA3

 

[inf64]

Almost all rely on AVX512, this is clearly visible from the test. RKL-S is on par or faster than ADL-S and RKL-S usually isn't faster than 10C CML-S in MT. There is a huge AVX512 speedup versus AVX2, that's why.

The problem is that it cannot beat 12 Zen3 cores while running at much higher clocks (8x GoldenCove). Zen3 has no AVX512 support, it has the same 2x256bit AVX2 units per core.

 

[mikk]

The problem is that it cannot beat 12 Zen3 cores while running at much higher clocks (8x GoldenCove). Zen3 has no AVX512 support, it has the same 2x256bit AVX2 units per core.

Yes I know, if there is nothing wrong with the ADL-S system it might be somehow AVX related, maybe AVX2 on Gracemont is slow and not a big help. Another theory from Sisoft is the memory subsystem.

The memory sub-system is crucial here, and these (streaming) tests show that using the Atom LITTLE/E cores is perhaps not worth it and just consume memory bandwidth that the big/P cores would put to better use. VAES/256 still helps a little as it almost ties with Zen3 using normal AES.

With compute, SHA HWA helps but again cannot beat multi-buffer AVX512 in RKL – here again is questionable how much the Atom cores help. With its 12C, Zen3 needs even faster memory to feed all the cores in streaming tests.

Streaming (bandwidth bound) tests don’t benefit from the LITTLE/E cores thus here only the big/P cores should be used. Other algorithms should just use cores (P & E) but not SMT threads on big/P cores. And since different-core (aka between P and E cores) transfer latencies are higher than same-core (aka between P-cores) threads sharing memory should stay on the same type of cores.

 

[inf64]

I think there is something off with the ADL system. Even if there is nothing wrong, Sisoft is not really that useful. We need some real world encoding, rendering, compiling, and gaming benchmarks.

 

[Asterox]

I think there is something off with the ADL system. Even if there is nothing wrong, Sisoft is not really that useful. We need some real world encoding, rendering, compiling, and gaming benchmarks.

Blast from the the past. Old weird times when Intel showed Conroe Preview, even six months before the official launch.

Spring IDF 2006 Conroe Preview: Intel Regains the Performance Crown

www.anandtech.com

Who stole Alder Lake Big/Little Preview, magic and extremely revolutionary Microsoft Windows 11? Or more precisely, whether Intel is even hard thinking about the Desktop CPU segment.An interesting history, as we know Intel was very confident in Conroe CPU success or real world performance.

Alder Lake hm, i dont see(not even close) same self-confidens from Intel in Alder Lake CPU success.CPU power efficiency, Alder Lake it does not stand well in that field.

 

[Hulk]

Alder Lake it does not stand well in that field.

PL2 isn't the only metric to determine power efficiency. We know that Intel big designs can clock very high with commensurate power and cooling. What we don't know, especially when it comes to the 10ESF Alder Lake process is what frequencies it will run before the voltage vs frequency plot goes wildly nonlinear? Will 4.5GHz all-core show good (low-ish) power usage? If so then with the IPC improvement of Golden Cove such frequencies will be competitive with Zen 3.

 

[DrMrLordX]

Supporting latest hardware tech properly is now a bad thing.

It was when Bulldozer hit the scene. It's not like people were falling all over themselves to support FMA4 or XOP, or engage in any other coding tricks to make the most out of BD or PD. And for good reason. We all expected AMD's "exotic" new module system to run existing software properly. When it didn't, people were rightfully critical.

Don't expect anything different for Alder Lake.

 

[igor_kavinski]

I keep wondering that Alder Lake was expressly designed by Intel to keep Apple happy (who purposely didn't inform Intel of their plans to kick Intel's butt with M1). Who knows? Maybe Apple will also have an Alder Lake Macbook line, for people who can't get enough of their bootcamp dualboot setups? Alder Lake's architecture makes sense in a laptop but in a desktop? After reliably providing the world with dependable CPUs that double as room heaters, now suddenly Intel wants to save Mother Earth? Something's not right here.

 

[DrMrLordX]

After reliably providing the world with dependable CPUs that double as room heaters, now suddenly Intel wants to save Mother Earth? Something's not right here.

Alder Lake-S will likely consume more power than Coffee Lake-S. Maybe the same as Comet Lake-S, maybe a little less . . . we'll see. But the existing power specs for it point to TDPs up to 125W and PL2 values of 225W or 241W (not sure whom to believe yet). In any case, that isn't exactly "saving Mother Earth".

 

[RanFodar]

It was when Bulldozer hit the scene. It's not like people were falling all over themselves to support FMA4 or XOP, or engage in any other coding tricks to make the most out of BD or PD. And for good reason. We all expected AMD's "exotic" new module system to run existing software properly. When it didn't, people were rightfully critical.

Don't expect anything different for Alder Lake.

With a company that has thousands of employees more than AMD, you'd think that Intel would, at the very least, mitigate many bugs or failures that might have gone with all the new stuff. With the way the world progresses nowadays, Intel's circumstances is very different than AMD during their time of Bulldozer.

 

[jpiniero]

I keep wondering that Alder Lake was expressly designed by Intel to keep Apple happy (who purposely didn't inform Intel of their plans to kick Intel's butt with M1)

Oh I'm sure Intel knew about Apple's plans to make their own Mac processors well before it was made public. Alder Lake might be Intel's reaction to it.

 

[firewolfsm]

Alder lake is just the only way Intel can increase core count due to power hungry big cores.

 

[igor_kavinski]

Oh I'm sure Intel knew about Apple's plans to make their own Mac processors well before it was made public. Alder Lake might be Intel's reaction to it.

Don't know about that. How does that help them compete with M1? M1X will further cement Apple's supremacy as the most powerful and power-efficient processor maker. Intel needs a decade to get anywhere close to that and by then, Apple will be further still.

 

[igor_kavinski]

Alder Lake-S will likely consume more power than Coffee Lake-S. Maybe the same as Comet Lake-S, maybe a little less . . . we'll see. But the existing power specs for it point to TDPs up to 125W and PL2 values of 225W or 241W (not sure whom to believe yet). In any case, that isn't exactly "saving Mother Earth".

Hmm...going above 200W is just insane. I think they did some analysis of common software usage patterns and figured that the low power cores would be sufficient most of the time for the majority of users. That way, no one can blame their CPUs for being power hungry. Oh, look. The average user will now get a lower electricity bill, they will tell us.

 

[uzzi38]

Don't know about that. How does that help them compete with M1? M1X will further cement Apple's supremacy as the most powerful and power-efficient processor maker. Intel needs a decade to get anywhere close to that and by then, Apple will be further still.

A decade?

Ahahahahahahahaha, no.

Apple have a great lead, but it is no way shape or form a decade ahead of everyone else. That's what we call delusion.

 

[Abwx]

Almost all rely on AVX512, this is clearly visible from the test. RKL-S is on par or faster than ADL-S and RKL-S usually isn't faster than 10C CML-S in MT. There is a huge AVX512 speedup versus AVX2, that's why.

In the financial computation ADL is 55% ahead of RKL in two tests out of three, that s about the expected throughput improvement in non AVX512 computation.

Also it match the 5900X in one test, is ahead by 10% in another one and losing the third by 10%, so that s a tie albeit with only 3 tests.

That being said the best "benchmark" is the way it will be priced, and if rumours are true then the leaked prices are set in respect of the competition current pricing, with eventualy a premium for the novelty, but in no way Intel will undercut AMD in price/perf ratio, this has never been their way of doing business if we except some competitive offerings in the low end from time to time.

Seesawing Intel Core i9-12900K sample with disabled AVX-512 teeters against the i9-11900K in preview benchmarks and fails to overwhelm the AMD Ryzen 9 5900X

A benchmark preview of the Intel Core i9-12900K has been accidentally posted online, and it shows the 12th Gen Alder Lake hybrid chip offering a seesaw-like performance against the Rocket Lake i9-11900K. While the engineering sample does show promise even without the AVX-512 instruction set, it...

www.notebookcheck.net

 

[NTMBK]

A decade?

Ahahahahahahahaha, no.

Apple have a great lead, but it is no way shape or form a decade ahead of everyone else. That's what we call delusion.

I remember when people were saying that AMD would never catch up to Intel. It was the height of the Bulldozer malaise, just before Zen was announced. AMD had just brought out yet another disappointing APU, it looked like they were never going to replace the Piledriver high performance/server CPUs... people were writing AMD's obituary.

Things change!

 

[mikk]

In the financial computation ADL is 55% ahead of RKL in two tests out of three, that s about the expected throughput improvement in non AVX512 computation.

What about the other tests, they doesn't matter for you?

 

[Abwx]

What about the other tests, they doesn't matter for you?

Since AVX512 blur the lines in the SIMD/Vector tests it s better to look only at numbers that do not seems to use those instructions, at least not in an efficent way.
Btw what is missing in those Sandra numbers, and of more relevant importance, are the arithmetic tests.

In this case the financial computation is more or less 55-60% throughput improvement over RKL, wich sound right given the uarch.
If we take Puget s numbers of 11.4% better perf in rendering for 8 PC and 5GHz all core (4.85GHz for RKL without AB) then the big cores will provide 15-16% improvement to wich 45% are added by the small cores (assuming IPC comparable to SKL).

 

[DrMrLordX]

With a company that has thousands of employees more than AMD, you'd think that Intel would, at the very least, mitigate many bugs or failures that might have gone with all the new stuff. With the way the world progresses nowadays, Intel's circumstances is very different than AMD during their time of Bulldozer.

It isn't about "bugs" or "failures". It's about introducing a new computing paradigm while expecting software developers to accommodate your changes immediately. Intel is a larger organization with more financial backing, so it stands to reason that they can "money hat" developers to following along with whatever eccentric designs they dream up to replace their existing heterogeneous core solutions.

Hmm...going above 200W is just insane. I think they did some analysis of common software usage patterns and figured that the low power cores would be sufficient most of the time for the majority of users. That way, no one can blame their CPUs for being power hungry. Oh, look. The average user will now get a lower electricity bill, they will tell us.

For low-power users, yes, the Gracemont cores could probably suffice. For anyone running software where absolute performance is important, however . . .

 

[Zucker2k]

Here we go again....

New Benchmark Leak Reveals Impressive Performance Of Intel's 12900K, 12700K And 12600K Alder Lake Processors

Faster than AMD's Ryzen 9 5950X and Core i5 could be a mid-range killer

www.forbes.com

 

[igor_kavinski]

A decade?

Ahahahahahahahaha, no.

Apple have a great lead, but it is no way shape or form a decade ahead of everyone else. That's what we call delusion.

It took Intel almost 6 years to recover from the misstep that was the Pentium 4. And that was against AMD, a smaller company that has given them bug bites, so to speak, from time to time. As if AMD eating their lunch with Ryzen wasn't bad enough, now they also have to face off against Apple, a behemoth they couldn't have imagined in their worst nightmare. I hope for the sake of x86 that Intel is able to put up a real fight and take back the performance crown. Otherwise, we'll all be running x86 emulators or virtualized Windows ARM on Apple silicon in the not-too-distant future.

 

[igor_kavinski]

Here we go again....

New Benchmark Leak Reveals Impressive Performance Of Intel's 12900K, 12700K And 12600K Alder Lake Processors

Faster than AMD's Ryzen 9 5950X and Core i5 could be a mid-range killer

www.forbes.com

Core i5 beating Ryzen 9 in games would be quite the day.

 

[exquisitechar]

It took Intel almost 6 years to recover from the misstep that was the Pentium 4. And that was against AMD, a smaller company that has given them bug bites, so to speak, from time to time. As if AMD eating their lunch with Ryzen wasn't bad enough, now they also have to face off against Apple, a behemoth they couldn't have imagined in their worst nightmare. I hope for the sake of x86 that Intel is able to put up a real fight and take back the performance crown. Otherwise, we'll all be running x86 emulators or virtualized Windows ARM on Apple silicon in the not-too-distant future.

A15 is a dud, CPU-wise, and Intel and AMD aren't that far behind to begin with. Intel will have a solid peak ST performance lead with Golden Cove already. Perf/watt will still be an issue, I suppose.

 

[Asterox]

Here we go again....

New Benchmark Leak Reveals Impressive Performance Of Intel's 12900K, 12700K And 12600K Alder Lake Processors

Faster than AMD's Ryzen 9 5950X and Core i5 could be a mid-range killer

www.forbes.com

Cinebench is ideal marketing bench, if you close one and a half eyes+CPU power consumption figures are forbidden

- not care about system memory speed

- L3 Cache is optional or decoration, big or small L3 cache blah



R5 4650G vs R5 3600, Cinebench R20 very similar scores ST and MT.But in real life R5 3600 can be significantly faster CPU.