[Canard PC Hardware] Intel prepares Ryzen's response behind the scenes

[coercitiv]

As it's the same die as KBL, yeah. But I was talking about the platform, not the CPU.

And Atari2600 was talking about the CPU, not the platform

 

[dacostafilipe]

And Atari2600 was talking about the CPU, not the platform

And I was talking about what reasons Intel (!) has to do it

 

[SAAA]

So to get a 100 MHz boost in clocks (along with a 20W "boost" in power consumption), an entirely new motherboard is required.

I'm definitely sure that +100MHz won't need 20W more, it's likely a question of allowing more bad samples to pass binning, while on average a 2% overclock shouldn't change power consumption beyond a couple watts on most CPUs. I was indeed expecting higher stock frequencies but right now I'd rather prefer Intel to use solder for these dies than any other change, 5GHz overclocks could be 100% assured that way without need of delidding. Cooling should be way easier too: look at the size of HEDT heatspreader against mainstream one!

 

[Eddward]

Isn't KBL-X dual channel?

As it's the same die as KBL, yeah. But I was talking about the platform, not the CPU.

I don't get it. KBL-X is the same die as KBL and yet they are for completely different sockets and platforms... How is this possible ?

 

[cytg111]

I really hope I'll get little sleep, I'm a big fan of sane competition and I'm tired with Intel behavior.

But do you really think AMD or Apple will come with a chip with better single thread performance than Intel within 2 or 3 years? I don't.

Not a chip perhaps, but mayhaps on a process?

 

[dacostafilipe]

I don't get it. KBL-X is the same die as KBL and yet they are for completely different sockets and platforms... How is this possible ?

TBH, it also makes no sense to me ... maybe Intel is looking for an "entry level" solution to better counter Ryzen in specs?

No idea, really ...

 

[cytg111]

The reason is the migration to the new quad-channel platform ... and that's about it!

And make way for coffeelake++ ? Plus we know skylake scales very well with memory speeds .. going quad will surely net them some extra top bin perf.

 

[lolfail9001]

The only viable theory for KBL-X i have seen so far is that Intel attempts to milk enthusiasts even more and slowly shift out z chipsets altogether.

 

[CentroX]

Intel want you to buy new motherboards as often you can. Their amount of different sockets is ridicilous.

 

[hrga225]

I really hope I'll get little sleep, I'm a big fan of sane competition and I'm tired with Intel behavior.

But do you really think AMD or Apple will come with a chip with better single thread performance than Intel within 2 or 3 years? I don't.

Well,at ISO power Apple is thouroughly beating Intel already.Since most of the Apple profits comes from different market segment,we will probably never see comparison at high TDP.
I give AMD 50/50 chance.I do believe they have an excellent base in Zen architecture.

 

[Lodix]

The answer is obvious: Intel IPC already is quite high and their frequency already is quite high. AMD was starting from a sh*tty CPU so gaining a lot of IPC was trivial. Apply isn't targeting >4 GHz (yet) so gaining IPC is easier.

Wake me when either AMD or Apple has the edge on single thread performance over Intel.

Gaining IPC in a mobile platform with a small TDP of 5W shared with the CPU+GPU+plus other chips easy ? Tell that to intel haha

Just checking some Geekbench results of the 7700k I see that Apple has already beaten intel in IPC.

 

[Eddward]

TBH, it also makes no sense to me ... maybe Intel is looking for an "entry level" solution to better counter Ryzen in specs?

No idea, really ...

"entry level" - don't see a reason but OK,
but I'm more interested in technical aspects of this... KBL is for LGA1151, KBL-X is big a LGA2066 ... LGA2066 is also for SKL-X and it is a quad channel platform, KBL-X is dual channel... just curious what a hybrid Intel did.

 

[dacostafilipe]

"entry level" - don't see a reason but OK

Duno, maybe they will try to position the LGA2066 against Ryzen and not LGA1151?

If you look at the Intel internal QA about Ryzen, it could make sense:

I'm only speculating here.

 

[bjt2]

Very interesting. Thanks.

I'm assuming this is accomplished by software/microcode, correct? Or is there a hardware stepping difference too?

About the Italian; technically I don't speak Italian, but I have a good working knowledge of Latin, so I think I got the larger picture...

According to the paper the update of the statistics for the AVFS is triggered by microcode (i suppose at bootstrap, as part of the boot time calibration) or by the SMU (maybe if it sense an high variation of temperature?). But it is only a signal. The statistics are collected by the 10 units spread on the chip.

EDIT: oh, you meant the turbo algorithm... Well i think it's SMU responsibility change p-state. I don't know if the SMU is microcoded. But probabily yes.

 

[Arachnotronic]

Gaining IPC in a mobile platform with a small TDP of 5W shared with the CPU+GPU+plus other chips easy ? Tell that to intel haha

Just checking some Geekbench results of the 7700k I see that Apple has already beaten intel in IPC.

Performance is clock speed * IPC. Performance per watt is (clock speed*IPC)/power consumption.

 

[Lodix]

Performance is clock speed * IPC. Performance per watt is (clock speed*IPC)/power consumption.

Did I say something wrong? Gaining IPC without increasing power consumption or area is not easy ( things that are very important on a mobile SOC ).

 

[Nothingness]

Gaining IPC in a mobile platform with a small TDP of 5W shared with the CPU+GPU+plus other chips easy ? Tell that to intel haha

Where did I say it's easy? I just said AMD and Apple are not able to beat Intel at ST performance at the moment and that I don't expect them to achieve that within 2 or 3 years.

What some people fail to see is that what Intel has been able to achieve also really is impressive. And that doesn't mean that what Apple and AMD are able to achieve isn't impressive.

Just checking some Geekbench results of the 7700k I see that Apple has already beaten intel in IPC.

Yes, but show me an Apple chip that beats a 7700K at Geekbench...

 

[SPBHM]

I fail to understand how kaby lake x is effective if it's in a more expensive platform and still with 4 cores....

 

[PPB]

I think that what is happening here is that Intel didn't expect AMD to compete, so they scheduled the single best 4c/8t you can buy in the cashgrab platform (errr, socket 2066). So people in the mainstream is happy that their 7700k is the single best processor for lga1151, and if you want to go even beyond you could go with the 7740k on lga2066 without conflicting pricepoints and juxtapositions.

Now that AMD is releasing their Ryzen skus, Intel might be wanting those processors would have been LGA1151 all along. Now releasing a faster 7700K would invalidate the kaby lake X counter part, and LGA2066 will have too much of a markup in platform price to compete straight on with ryzen (the lowest price for a 2011-3 after years of being launch is the same price your usual AM3+ high end prices shipped with when Piledriver was still competitive years ago.

That difference in platform price can't be ignored, specially if you can OC on B350. This is 100 bucks motherboard territory we are talking about aprox. You really have to dig in there to find a 150-160 bucks 2011-3 board and this accounting the platform is getting kind of dated by intel standards. Even without OC, A320 boards with sufficient power delivery could be have had for less and let XFR guide you through good single threaded performance so minimal compromises there.

Intel will for sure keep the 4c/8t single threaded performance lead. But everything else besides that 2c/4t pentium is probably boned, price to performance wise.

Then again, we have to wait and see how the "b-but intel is better!!" effect takes place. Mindshare plays a big role as we can see from Nvidia vs AMD.

 

[sm625]

Where did I say it's easy? I just said AMD and Apple are not able to beat Intel at ST performance at the moment and that I don't expect them to achieve that within 2 or 3 years.

Apple already has Intel beat in terms of web benchmarks. I believe A10 around 10% faster, clock for clock. Given the constant improvement in web benchmarks, it is obvious that apple sees this as the most important (non-gpu) area of focus, to which I agree. At the current rates of improvement, within 3 years iphones will outperform desktop i7s on most single threaded benchmarks. (Assuming max i7 turbo remains at 4.5GHz) Given this, it has to be obvious that Intel is sitting on something.

 

[lolfail9001]

I believe A10 around 10% faster, clock for clock.

ST Performance is clocks * IPC. So, while A10 is faster per clock, it has quite a road to go clock-wise.

At the current rates of improvement, within 3 years iphones will outperform desktop i7s on most single threaded benchmarks.

At the 2011 rates of improvement in desktop space, we would have 50% faster CPUs now. Or something.

 

[Lodix]

Where did I say it's easy? I just said AMD and Apple are not able to beat Intel at ST performance at the moment and that I don't expect them to achieve that within 2 or 3 years.

What some people fail to see is that what Intel has been able to achieve also really is impressive. And that doesn't mean that what Apple and AMD are able to achieve isn't impressive.


Yes, but show me an Apple chip that beats a 7700K at Geekbench...

Don't you realize that comparing the performance of a MOBILE SOC (a lot of stuff together) with a 5W TDP against a 100W TDP CPU is hilarious ? Apple has never designed a CPU to compete in that segment and they don't have the need to right now. But as you guys are repeating performance=IPCxfrequency and I pointed out that Apple has already better IPC/architecture so It is only question of scalling frequency and redesigning the chip.

 

[LTC8K6]

I fail to understand how kaby lake x is effective if it's in a more expensive platform and still with 4 cores....

It's probably pretty simple. KL-X has nothing at all to do with combating RyZen, and it will be released when it was supposed to be released. Most of us are probably connecting some things which actually aren't connected.

 

[Flash831]

I think that what is happening here is that Intel didn't expect AMD to compete, so they scheduled the single best 4c/8t you can buy in the cashgrab platform (errr, socket 2066). So people in the mainstream is happy that their 7700k is the single best processor for lga1151, and if you want to go even beyond you could go with the 7740k on lga2066 without conflicting pricepoints and juxtapositions.

Now that AMD is releasing their Ryzen skus, Intel might be wanting those processors would have been LGA1151 all along. Now releasing a faster 7700K would invalidate the kaby lake X counter part, and LGA2066 will have too much of a markup in platform price to compete straight on with ryzen (the lowest price for a 2011-3 after years of being launch is the same price your usual AM3+ high end prices shipped with when Piledriver was still competitive years ago.

That difference in platform price can't be ignored, specially if you can OC on B350. This is 100 bucks motherboard territory we are talking about aprox. You really have to dig in there to find a 150-160 bucks 2011-3 board and this accounting the platform is getting kind of dated by intel standards. Even without OC, A320 boards with sufficient power delivery could be have had for less and let XFR guide you through good single threaded performance so minimal compromises there.

Intel will for sure keep the 4c/8t single threaded performance lead. But everything else besides that 2c/4t pentium is probably boned, price to performance wise.

Then again, we have to wait and see how the "b-but intel is better!!" effect takes place. Mindshare plays a big role as we can see from Nvidia vs AMD.

I think you should also consider the yield issue. The new SKU's surely are a response to Ryzen. However, Intel cannot churn out infinite numbers of these as surely only a small minority of all chips will qualify. I believe the reason to release the new SKU's to LGA2066 is that there will be less demand for the chip on that platform. Intel might not be able to keep up with demand if it releases the chips to LGA1151.

 

[MajinCry]

Duno, maybe they will try to position the LGA2066 against Ryzen and not LGA1151?

If you look at the Intel internal QA about Ryzen, it could make sense:

I'm only speculating here.

Man, that's just damage control marketing right there.

"Core's don't matter, but the Intel™ Busty Buster™ 1020 has more."