Intel's response to RyZen.

[tamz_msc]

Whatever Intel's response is it won't be another Conroe. An Athlon 64 or Core 2, or in this case Ryzen happens once in a decade.

 

[Charlie22911]

2. process leadership
Zen+ and Zen++ are said to be optimization, optimization (tock-tock) both at 14nm so 2018 and 2019. And then hopefully 2019-2020 GloFo comes out with 7nm cuz they skipping 10nm.
Intel OTOH is releasing 10nm next year. Supposed to be late this year but let's be realistic with volume production and say next year. This means most of 2018 and probably 2019 Intel regains its process leadership at 10nm compared to Zen stuck at 14nm. Makes it easy for Intel to go up to 8 cores and hopefully keep the clocks high while keeping thermals under control.

A process lead doesn't just give you performance on its own, at this point I think it's more for maximizing number of dies (dice?) per wafer.
Take 16nm pascal vs 14nm polaris for example, nvidia isn't at a disadvantage; quite the opposite actually.

Diminishing returns can be quite the boogie man...

 

[IntelUser2000]

Maybe they'll port Icelake cores onto 14nm and call that Coffelake, maybe only Cannonlake cores, I don't know for sure but that's what I would do if I was at Intel.
It should also explain the 15% increase cited here:
Now Kabylake didn't achieve exactly 15% single thread improvement on all chips but it's close: stock 7700K runs 4.4GHz all core turbo while 6700K does all core at 4.0GHz →

Guys.

Kabylake did achieve practically 15%. Just not in the area you guys are thinking of:
http://www.notebookcheck.net/Kaby-Lake-Core-i7-7500U-Review-Skylake-on-Steroids.172692.0.html

Sysmark isn't single threaded. But it isn't very well multi-threaded either. From my memory I know its at least 2. In mobile, Kabylake gained more than Haswell, a tock.

If we take the same logic, then "8th Gen core" is mobile as well.

 

[IntelUser2000]

At RWT they are talking about Zen's SpecCPU2006 scores.

The Base version of the benchmark(not-so-threaded) without the benchmark-breaking libquantum shows that Skylake has about 10% advantage. Meaning its on par with Broadwell, if not better.

Intel has no true answer for at least few years. Man this is a repeat of 1999's Athlon again. The financial impact to Intel is practically zero for couple of years just like with the Athlon. The real ramifications come in a few years when modification to the core arrives like Athlon 64 did. Then they have chance to take real marketshare(AMD briefly took 50% marketshare in retail).

It is a bit more complex than that of course. In terms of marketshare, they need a good APU with Raven Ridge. You NEED an iGPU to sell into most computers.

Oh, and I highly doubt SKL-X and KBL-X will do more than 5%, if that. What secrets do you think they are hiding? It's a fallacy to believe magical big gains are out there.

For how long has Intel known about the potential of ryzen and has it given them enough time to tweak 10nm towards mega herzes .. next shrink may reintroduze the Hz race... that would be nice..

Clock speed wars are dead. CPUs are practically at their maximum clock. To tell you how little about the future even their engineers know about CPUs, Intel was planning 5GHz for Prescott with quarterly 500MHz ramps. They didn't even reach 4GHz with back-then extraordinary heat output and power use.

 

[tamz_msc]

^This is something that should be more widely known but is quite conveniently sidestepped in the discussion on desktop chips. Where Kaby Lake really shines is in power and thermal-limited scenarios. It is also the reason why Data Center will be the priority on the new process nodes for Intel from now on. On desktop a few percent gains from one generation to the next is going to be the norm.

 

[Absolute0]

A process lead doesn't just give you performance on its own, at this point I think it's more for maximizing number of dies (dice?) per wafer.
Take 16nm pascal vs 14nm polaris for example, nvidia isn't at a disadvantage; quite the opposite actually.

Diminishing returns can be quite the boogie man...

With a process advantage Intel has the option to mint more die per wafer, yes, and simply gain economic advantage.

Alternatively, they can exploit the smaller die area to facilitate the transition to mainstream 8C chips. Which, as AMD has shown, does translate to performance.

The point is at 10 nm they have these options, and AMD is stuck with GloFo's 14nm until ~2019, earliest. One might assume Zen+ 2018 14nm, Zen++ 2019 14nm, 2020 revamped uarch @ 7nm.

I think that there are definitely elements of the semi industry (per my reading on semiengineering.com) which have observed the pains transitioning to 10nm and believe that the 7nm transition will likely be delayed... The process requires so many lithographic steps that it becomes quite slow and expensive... And it does not seem like EUV process tech is arriving in time to properly aid the 7 nm transition in the next 2 years. The point is, by skipping 10nm, GloFo is tackling a node that may be difficult to deliver in strong yields in ~2019 and may not see healthy volume until 2020+ (it is a risk). More likely they will push it out more or less "on time" because of the compromises inherent to what most foundries are calling "7nm," which is scarcely smaller than intel's 10nm.

TLDR Intel's most painful point, from a process standpoint, is right now, and things start looking brighter after 10nm cannonlake arrives ~EOY

 

[misuspita]

With a process advantage Intel has the option to mint more die per wafer, yes, and simply gain economic advantage.

Alternatively, they can exploit the smaller die area to facilitate the transition to mainstream 8C chips. Which, as AMD has shown, does translate to performance.

The point is at 10 nm they have these options, and AMD is stuck with GloFo's 14nm until ~2019, earliest.

I think that there are definitely elements of the semi industry (per my reading on semiengineering.com) which have observed the pains transitioning to 10nm and believe that the 7nm transition will likely be delayed... The process requires so many lithographic steps that it becomes quite slow and expensive... And it does not seem like EUV process tech is arriving in time to properly aid the 7 nm transition in the next 2 years. The point is, by skipping 10nm, GloFo is tackling a node that may be difficult to deliver in strong yields in ~2019 and may not see healthy volume until 2020+ (it is a risk). More likely they will push it out more or less "on time" because of the compromises inherent to what most foundries are calling "7nm," which is scarcely smaller than intel's 10nm.

TLDR Intel's most painful point, from a process standpoint, is right now, and things start looking brighter after 10nm cannonlake arrives ~EOY

Nope. Intel's problem is one that is self inflicted. They segmented the market so much, and separated features artificially (avx, virtualization, etc), they stopped the IPC improvement to single digit percentage and were flat-out unprepared for a quality response from AMD. Their problem is that they have been accustomed to the big profits they made the last decade. And AMD delivering a equal or better CPU means they have no solution but lowering prices and subsequently profits. And this will hurt most in server where profits are huge.

 

[itsmydamnation]

Intel has no true answer for at least few years. Man this is a repeat of 1999's Athlon again. The financial impact to Intel is practically zero for couple of years just like with the Athlon. The real ramifications come in a few years when modification to the core arrives like Athlon 64 did. Then they have chance to take real marketshare(AMD briefly took 50% marketshare in retail).

It is a bit more complex than that of course. In terms of marketshare, they need a good APU with Raven Ridge. You NEED an iGPU to sell into most computers.

I dont think it is even remotely the same as back in 99. From 99 onwards is was a slow ramp up and then down right up until neleham and QPI killed AMD's last advantage.

In the space of 6 months AMD will go from not being competitive in any market to heavy competing in all Intel markets.

In the laptop space Zen + Vega will be a better SOC then kabylake
In Server 32 core naples will better then broadwell-EP not only as a processor but as a platform. It will also be very competitive to Skylake-EP (outside people who want AVX-512) in performance and as a platform.
In Highend desktop we can already see the enthusiasm, all the people who are hardcore enthusiast with money now all grew up on the great value of K7 and early K8.
Even with "crappy" excavator AMD has more OEM product then during the golden years thax to the lack of intel bribes.

That really only leaves intel with one area of advantage and that is the high clocked 4 core.


Now add in that AMD will have more manufacturing capacity then it has ever had and that there highend SOC multi chip design methodology has meant they are covering 4 intel chips will 1 ( kaby lake, Xeon-D, broadwell/skylake-E and broadwell/skylake-EP) they can get super aggressive if they need/want to with binning in a way intel can only dream.


This time its going to actually hurt intel because of market conditions and i think far quicker then most people are predicting. How intel will react will be interesting and i think will be determined by what their next products actually look like, price cuts? "cost cutting"? ignoring until its undeniable ?

TLDR Intel's most painful point, from a process standpoint, is right now, and things start looking brighter after 10nm cannonlake arrives ~EOY

No it isn't, nothing is out yet. Process shinks are less and less important, cannonlake is nothing but haswell++++. Manufacturing costs are silly low for both intel and AMD, these chips are small, Zepplin is smaller then P10 and RR will be as well. The costs aren't in manufacturing they are in paying 110k employees vs 10k.

Just remember Piledirver came 12months exactly after bulldozer, there is a real chance of seeing a new Zen core early Q1 next year, even if its on 14nm so long as performance improves (it will the process will be better and there will be low hanging uarch fruit to pick) then what?

 

[vissarix]

if we were to believe rigged Amd benchmarks and rabid fanboys posting new threads on how ryzen is amazing every few seconds then Intel is in trouble...

in reality the i7 6900k does 1575 in cinebench vs ryzen 1800x does 1600 according to Amd so take it with a huge truck load of salt...

best case scenario they are on par on multithread and the i7 being faster on single thread but has worse smt implementation so Ryzen keeps up on the multithread...this means Intel has to lower prices to remain competitive...not that big of a deal and definitely not desperate like some want it to be...

and then Skylake X will come out crushes Ryzen on every metric again..

Trolling is not allowed
Name calling is also not allowed
Markfw
Anandtech Moderator

 

[itsmydamnation]

yep lets see how well a $1500 USD skylake-X's sells...................

 

[LTC8K6]

Guys.

Kabylake did achieve practically 15%. Just not in the area you guys are thinking of:
http://www.notebookcheck.net/Kaby-Lake-Core-i7-7500U-Review-Skylake-on-Steroids.172692.0.html

Sysmark isn't single threaded. But it isn't very well multi-threaded either. From my memory I know its at least 2. In mobile, Kabylake gained more than Haswell, a tock.

If we take the same logic, then "8th Gen core" is mobile as well.

The mobile chips are a bit misleading. They are able to stay at higher boost levels for longer periods than previous chips. The gains in mobile are clock speed gains, not ipc gains.

 

[inf64]

if we were to believe rigged Amd benchmarks and rabid fanboys posting new threads on how ryzen is amazing every few seconds then Intel is in trouble...

in reality the i7 6900k does 1575 in cinebench vs ryzen 1800x does 1600 according to Amd so take it with a huge truck load of salt...

best case scenario they are on par on multithread and the i7 being faster on single thread but has worse smt implementation so Ryzen keeps up on the multithread...this means Intel has to lower prices to remain competitive...not that big of a deal and definitely not desperate like some want it to be...

and then Skylake X will come out crushes Ryzen on every metric again..

Relax man, you seem very excited. Are you doubting the scores are real?

 

[vissarix]

Relax man, you seem very excited. Are you doubting the scores are real?

Yeah man why doubt AMD?

They showed us benchmarks when Fury X was scoring higher then a Titan X...LMAO!!

 

[inf64]

Yeah man why doubt AMD?

They showed us benchmarks when Fury X was scoring higher then a Titan X...LMAO!!

You should just wait a week, March 2nd is coming very soon.

 

[tamz_msc]

The mobile chips are a bit misleading. They are able to stay at higher boost levels for longer periods than previous chips. The gains in mobile are clock speed gains, not ipc gains.

And who's to say that 14nm++ Cannon Lake will not bring even more clock speed gains? That 15% more that they are promising might very well be mobile parts clocking at the 4GHz range, unless they improved the underlying architecture.

 

[coercitiv]

You should just wait a week, March 2nd is coming very soon.

Nah, let the man speak. Finding out Zen has better SMT implementation than Intel's matured HT is a gem of a compliment.

 

[LTC8K6]

And who's to say that 14nm++ Cannon Lake will not bring even more clock speed gains? That 15% more that they are promising might very well be mobile parts clocking at the 4GHz range, unless they improved the underlying architecture.

We are already at 91W tdp with KL desktop chips. We know that 4 core KL-X, with a little more clock, 100mhz more I think, is at 112W. Hard to believe 14nm is going to get much better in terms of clocks, unless you want an even higher tdp rating.

 

[tamz_msc]

if we were to believe rigged Amd benchmarks and rabid fanboys posting new threads on how ryzen is amazing every few seconds then Intel is in trouble...

in reality the i7 6900k does 1575 in cinebench vs ryzen 1800x does 1600 according to Amd so take it with a huge truck load of salt...

best case scenario they are on par on multithread and the i7 being faster on single thread but has worse smt implementation so Ryzen keeps up on the multithread...this means Intel has to lower prices to remain competitive...not that big of a deal and definitely not desperate like some want it to be...

and then Skylake X will come out crushes Ryzen on every metric again..

Intel has no reason to reduce the price of HEDT Broadwell-E because they are already raking in the profits by selling multi-thousand dollar Xeons and the Broadwell-E is derived from the Xeon dies. Even if they lose all the money to AMD that they get from selling chips for the X99 platform, that is a mere drop in the ocean compared to the huge margins that they get from the enterprise market.

What they are worried about is the regular 'enthusiast' desktop lineup and that AMD would carve a significant chunk of that market for themselves. The R7 1700 can easily be a better all-round performer than the i7 7700K.

 

[french toast]

if we were to believe rigged Amd benchmarks and rabid fanboys posting new threads on how ryzen is amazing every few seconds then Intel is in trouble...

in reality the i7 6900k does 1575 in cinebench vs ryzen 1800x does 1600 according to Amd so take it with a huge truck load of salt...

best case scenario they are on par on multithread and the i7 being faster on single thread but has worse smt implementation so Ryzen keeps up on the multithread...this means Intel has to lower prices to remain competitive...not that big of a deal and definitely not desperate like some want it to be...

and then Skylake X will come out crushes Ryzen on every metric again..

please explain how AMD "rigged" the test? -no more open gl benchmarks please

 

[inf64]

please explain how AMD "rigged" the test? -no more open gl benchmarks please

But but Fury X!

 

[vissarix]

please explain how AMD "rigged" the test? -no more open gl benchmarks please

Dude i explained it many times already...

Amd showed Cinebench R15 score

Ryzen R7 1800x scoring 1601

I7 6900k scoring 1474? Actually is 1578 from independent reviews..

thats the problem..

 

[CentroX]

Dude i explained it many times already...

Amd showed Cinebench R15 score

Ryzen R7 1800x scoring 1601

I7 6900k scoring 1474? Actually is 1578 from independent reviews..

thats the problem..

These scores are within error of margin

 

[LTC8K6]

These scores are within error of margin

Could be ram speeds?

 

[itsmydamnation]

Could be ram speeds?

i think it will be cooling and all core turbo, because 1100USD only entitles you to a POS cooler.....

 

[french toast]

Dude i explained it many times already...

Amd showed Cinebench R15 score

Ryzen R7 1800x scoring 1601

I7 6900k scoring 1474? Actually is 1578 from independent reviews..

thats the problem..

Have you taken cooling and differences in ram speed and timings into account between reviews?
Lets take Anandtech review, they used a closed loop water cooler for their 6900k review and it STILL posted lower MT scores than Ryzen on stock air cooler! what kind of boost would ryzen see with the same cooling solution with XFR? AMD seems to have used stock 140w coolers for BOTH platforms, is it AMDs fault that intel cheaped out on their own cooler?
Also AMD used turbo boost 3 @4ghz when Anandtech didn't due to fiddling around in the bios (cant remember) so they already did something to help 6900k scores when some reviews didn't, some consumers are not going to touch the setting in bios, AMD gave them the benefit of the doubt here.
Come on be fair, lets not start spreading fud ok?