Intel Skylake / Kaby Lake

[Arachnotronic]

We will wait and see if actuals match guidance.

They did for Q1 and at this point, they have enough data on Q2 to be able to estimate reasonably well. It's not like they're going into Q2 without any sales data, conversations with OEMs/ODMs, etc.

 

[Jan Olšan]

There is no such rumor, link to your source if you have. Or should we assume you are trolling? Based on your posting history you are neither trustable nor serious.

There certainly is a rumour like that. I have heard it too. But I don't know what is it based on, either.

 

[ShintaiDK]

We will wait and see if actuals match guidance.

Same BS as always isn't it. Never an AMD product that wont sell like hotcakes according to you. So now the forever #waitforamd game continues.

How many years haven't you advocated this crap?

And good luck with your Naples fairytale dreams. Unlike desktop its not so easy for AMD to lie about power consumption or actual performance. And Skylake-SP/EP have shipped since last year for top customers.

And talking about desktop, how many Ryzen chips have one of the top 9 AMD premium retailers with preorders as well even sold? Lets see, 6810 chips in around 2 months.

 

[raghu78]

They did for Q1 and at this point, they have enough data on Q2 to be able to estimate reasonably well. It's not like they're going into Q2 without any sales data, conversations with OEMs/ODMs, etc.

I was talking wrt to full year guidance.

 

[Jan Olšan]

Unlike desktop its not so easy for AMD to lie about power consumption or actual performance.

Can you elaborate? AMD lied about performance or power consumption?

BTW those etailer sale numbers need to be taken in proportion to Intel sales. They sold about 75 000 pieces of i7-6700Ks and i5-6600Ks combined in year and a half since their introduction, it seems. 4500 pieces of Ryzen 7 chips in under two months might still be meaningful marketshare. It is certainly too early to tell, because it is too soon after launch/preorders period and we don't know where will the sales trend to in longer term.

BTW, only 4500 total i7-6850Ks and i7-6800Ks since their launch.

 

[raghu78]

Same BS as always isn't it. Never an AMD product that wont sell like hotcakes according to you. So now the forever #waitforamd game continues.

How many years haven't you advocated this crap?

And good luck with your Naples fairytale dreams. Unlike desktop its not so easy for AMD to lie about power consumption or actual performance. And Skylake-SP/EP have shipped since last year for top customers.

And talking about desktop, how many Ryzen chips have one of the top 9 AMD premium retailers with preorders as well even sold? Lets see, 6810 chips in around 2 months.

dude don't get too so emotional. How well AMD or Intel products sell will eventually reflect in their financials. so cool. lets not argue over that. btw yes server market share gains are hard and take more time. But if you have a competitive product you will slowly and steadily gain share. I am confident AMD can gain share gradually and strengthen their position. btw about performance or power consumption nobody should take a manufacturer's claims as such and should do their own testing. OEMs and server customers are smart enough to do that and compare Naples vs Skylake-EP and make their own purchasing decisions. Don't argue on a forum about such things.

 

[raghu78]

Can you elaborate? AMD lied about performance or power consumption?

BTW those etailer sale numbers need to be taken in proportion to Intel sales. They sold about 75 000 pieces of i7-6700Ks and i5-6600Ks in year and a half since their introduction, it seems. 4500 pieces of Ryzen 7 chips in under two months might still be meaningful marketshare. It is certainly too early to tell, because it is too soon after launch/preorders period and we don't know where will the sales trend to in longer term.

BTW, only 4500 i7-6850Ks and i7-6800Ks since their launch.

Warning : Do not expect a rational conversation with well though out and logical arguments.

 

[ShintaiDK]

Can you elaborate? AMD lied about performance or power consumption?

That the TDP numbers are fairy tale numbers. 95W TDP that uses 120-130W etc.

 

[Crumpet]

Can you elaborate? AMD lied about performance or power consumption?

BTW those etailer sale numbers need to be taken in proportion to Intel sales. They sold about 75 000 pieces of i7-6700Ks and i5-6600Ks combined in year and a half since their introduction, it seems. 4500 pieces of Ryzen 7 chips in under two months might still be meaningful marketshare. It is certainly too early to tell, because it is too soon after launch/preorders period and we don't know where will the sales trend to in longer term.

BTW, only 4500 total i7-6850Ks and i7-6800Ks since their launch.

Only 4500 Ryzen 7 chips?

OverclockersUK, a single store in Britain, had 500 Ryzen 7 preorders in their first hour of availability.

 

[ShintaiDK]

dude don't get too so emotional. How well AMD or Intel products sell will eventually reflect in their financials. so cool. lets not argue over that. btw yes server market share gains are hard and take more time. But if you have a competitive product you will slowly and steadily gain share. I am confident AMD can gain share gradually and strengthen their position. btw about performance or power consumption nobody should take a manufacturer's claims as such and should do their own testing. OEMs and server customers are smart enough to do that and compare Naples vs Skylake-EP and make their own purchasing decisions. Don't argue on a forum about such things.

How many years haven't you advocated AMD products that the next one is the saviour? How many sites and forums dont you post the same nonsense to? Yet Nvidia and Intel shares keeps going UP.

And yes, I can guarantee you OEMs and cloud vendors knows exactly how Naples performs. And its regarded as a complete joke. Extremely poor scaling, low clocks and very poor performance/watt. And we all know how that turns out on sales and financials.

Warning : Do not expect a rational conversation with well though out and logical arguments.

Oh the irony

 

[Jan Olšan]

That the TDP numbers are fairy tale numbers. 95W TDP that uses 120-130W etc.

Your picture is nonsense (sorry). You are conveniently forgetting that the at-wall (!) measurement includes loss in motherboard VRMs and in the ATX PSU. TDP doesn't account for either and as such there is easilly enough room to fit 95W TDP CPU in that delta. I hope you aren't spreading this misinformed claim/"proof" on purpose (I'm kind of amazed somebody long active on this forum doesn'T know this, same goes for that like you received BTW).

 

[moonbogg]

I just trying to say that HEDT and Xeon server prices are NOT the same, on HEDT a 6900K cost $1000, on server a 8C/16T starts at $400, yes 6900K is faster.

Thats the simple fact, 32C Ryzen will be up vs +20C Xeons, Xeons having IPC and AVX avantage.

On HEDT the 16C Ryzen may be a problem for 8C and 10C/12C, yes, still Skylake-X is LCC, it has latency, IPC, clocks, and AVX-512 on their side. Is not gonna be so easy for AMD as you may think.

It will be easy as pie for AMD because a 12 core Skylake will probably cost as much as a used car.

 

[raghu78]

How many years haven't you advocated AMD products that the next one is the saviour? How many sites and forums dont you post the same nonsense to? Yet Nvidia and Intel shares keeps going UP.

And yes, I can guarantee you OEMs and cloud vendors knows exactly how Naples performs. And its regarded as a complete joke. Extremely poor scaling, low clocks and very poor performance/watt. And we all know how that turns out on sales and financials.

Oh wow. So you know everything already. Are you working at a server OEM or are you just talking out of your rear.

 

[xdfg]

It will be easy as pie for AMD because a 12 core Skylake will probably cost as much as a used car.

Forget used cars, Intel is going to charge enough to buy a brand new car, plus taxes. TweakTown: Intel's new Xeon rocks 28C/56T, costs over $12,000. Naples with 16 cores and even 32 for cheap will destroy Intel market share. R7-1700 is currently selling for $320, and with AMD's innovative fabric and MCM, getting to 16 cores will cost merely double, or $640. Even the top 32 core chip is only going to cost a little over a grand.

 

[Jan Olšan]

AMD will also try to charge higher sums for Naples - or better said, as high as possible. If they can sell 32c64t for 5000-7000 USD, they will (or even more). besides using four chips in an expensive package, obviously enterprise and server markets are their chance to ramp up revenue and margins, so they can't sell cheap. They might try to undercut Intel some, but the prices will still be high.

 

[shortylickens]

shorty just upgraded to the i7 7700k.
shorty is pleased.

 

[Markfw]

That the TDP numbers are fairy tale numbers. 95W TDP that uses 120-130W etc.

So comparing the 2 8c/16t CPUs on this screen, the 1800x is slightly lower in load, and way lower in idle power usage than the 6900k, but in performance swaps wins with it. AND its half the price.

So Intel is still better ? And...This is justified how ? Remember, I have 4 E5-2683's and 2 Ryzens, and as I said before, they swap performance, and the Ryzen does it with 6 less cores, so you sure can't call me biased.

 

[.vodka]

https://twitter.com/iBUYPOWER/status/857741376770850818

and...

https://www.reddit.com/r/Amd/comments/67pb15/lisa_su_oems_to_be_using_ryzen_sometime_in_q2/

It's starting.

Rest of the year (Coffeelake, Skylake-X, HEDT Ryzen, Raven Ridge) and 2018 (Pinnacle Ridge, whatever Intel has in the pipeline) should be a fun thing to watch... unlike the past five years.

And yes, I can guarantee you OEMs and cloud vendors knows exactly how Naples performs. And its regarded as a complete joke. Extremely poor scaling, low clocks and very poor performance/watt. And we all know how that turns out on sales and financials.

Okay, I'll bite for a little while.

I assume you haven't seen The Stilt's testing.

As indicated by the Vmin-Fmax curve, Zeppelin's voltage scaling is perfectly linear until 3.3GHz (25mV per 100MHz). The first deviation ("Critical 1") from this linear behavior can be seen at 3.3GHz. The second and the final deviation ("Critical 2") can be seen at 3.5GHz. Beyond this point the voltage scaling is neither linear or recovers even temporarily, and the CPU is requiring higher voltage in increasingly larger steps to scale further.

The ideal frequency range for the process or the design (as a whole) appears to be 2.1 - 3.3GHz (25mV per 100MHz). Above this region (>= 3.3GHz) the voltage scaling gradually deteriorates to 40 - 100mV+ per 100MHz.

This means that at ~3.8GHz pushing further usually becomes extremely costly (power / thermal wise).

and

Zeppelin features a highly advanced power management, as stated many times before. Just like Carrizo / Bristol Ridge, which feature a very similar PM, Zeppelin can infact support cTDP as well. cTDP is not officially supported (or available) on any consumer Zeppelin based SKU (AFAIK). The lack of official support is merely a distraction

850 points in Cinebench 15 at 30W is quite telling. Or not telling, but absolutely massive. Zeppelin can reach absolutely monstrous and unseen levels of efficiency, as long as it operates within its ideal frequency range.

In case you haven't realized it yet, what AMD sells as Ryzen in AM4 is an overclocked, pushed outside its efficient range server die as to compete with high clocked consumer Intel chips... and it's actually quite decent at that. Naples and its high core count models will operate inside Zeppelin's most efficient clock speed and voltage range (see the R7 1700 in perf/w, but better). It seems AMD targeted efficiency first here.

See what happens when you do this and scale core counts to fill the freed TDP quota? I wouldn't be so fast as to dismiss AMD this time, not at the server game. They're actually trying unlike the past five years with faildozer. You could be in for a surprise for the 16 and 32 core server Ryzens or whatever name they come up with for that market vs the Xeon competition, save for particular workloads that take advantage of Intel's SIMD performance.

Again, the rest of the year and 2018 will be fun to watch. Be thankful AMD is back in the game bringing some balance to the whole thing. Their first gen Zen silicon isn't bad at all. Second gen should be a nice improvement overall in both IPC/picking low hanging fruit on a new design and process improvements.

 

[jpiniero]

Oh wow. So you know everything already. Are you working at a server OEM or are you just talking out of your rear.

I'm sure the cloud companies have gotten their hands on pre-release Naples. Shintai of course has no real idea what they think though.

 

[imported_ats]

Forget used cars, Intel is going to charge enough to buy a brand new car, plus taxes. TweakTown: Intel's new Xeon rocks 28C/56T, costs over $12,000. Naples with 16 cores and even 32 for cheap will destroy Intel market share. R7-1700 is currently selling for $320, and with AMD's innovative fabric and MCM, getting to 16 cores will cost merely double, or $640. Even the top 32 core chip is only going to cost a little over a grand.

Umm, if the SS 8 core chip is going for $320, they aren't going to sell the DD 16c package for $640. Esp when it is going to have a higher loss rate (MCM processes aren't perfect), require significantly more validation, significantly more infrastructure, etc. And if you think they are going to sell the QD 32c package for anything close to $1280 or less...

The only way that comes about is if Ryzen completely and absolutely tanks in the non-server market. If you really want to set some real expectations, I would suggest you take the 1800x pricing as your absolute lowest starting point per die, then double/quadruple it, and add 20-100% to cover packaging loss and validation/verification. That puts the 16c at 1k+ and the 32c at 2k+ as a minimum and likely closer to 1.5k+ and 3k+ realistically.

 

[crashtech]

This IS the Skylake/Kaby Lake thread, right? I literally had to double check.

 

[tamz_msc]

This whole AVX512 support favoring Intel story is a bit of overhyping, IMO. Think of applications that are CPU FPU-bound - linear algebra(LINPACK), signal processing(FFTs), data analysis(lots of math functions) - these are the things that would obviously benefit from wider vector SIMD registers. But there are a lot of other workloads that are mostly I/O bound, especially with large data sets. Eulerian CFD, FEM analysis, basically most of the things that involve solving differential equations are I/O bound.

There is a point after which one has to really think, are all these vector computations done on the CPU actually worth the trade-offs in heat and power consumption, due to fitting bigger and bigger registers on the CPU for a specialized task like vectorized data processing? At that point, the logical thing to do would be to move to things like Xeon Phi or GPUs.

 

[RichUK]

What's with all this Ryzen garbage in the Intel thread?

 

[Ajay]

Even when AMD & partners get the ecosystem problems worked out (board/bios/mem issues), they are still capacity limited at Fab 8. There is a 20% wafer rate expansion going on (with newer ASML equipment), they still will not be able to have the impact on global sales that Intel has. I don't know if AMD/GFL have plans to expand further or not - if they do, I would imagine that these plans are contingent on the success of Zen based products.

If AMD can stay competitive with Intel for the next there iterations of their new uAarch and maintain the price/performance lead that they have, then we should see more significant changes. I think server adoption occur more conservatively, though based on Slit's data, high core-count Naples servers may be very attractive (though I haven't heard as much noise as I'd like about platform partners).

Bottom line - AMD has allot of work ahead of it to establish and grow the platform ecosystem (including software support) in mainstream, enthusiast and server markets. As such, I do not think Intel's margins are in danger this year and probably not significantly next year either. Lastly, though caught a bit off guard, Intel is still a moving target in terms of performance and power efficiency (and eventually on price).

The good new for us here, is that competition should increase the performance/$$ ratio and accelerate innovation - re-igniting a stagnant market.

 

[Nothingness]

Haswell launched four years ago, and not only does it seem to have a slight edge in legacy x86 IPC, it has much more advanced SIMD machinery (2x256bit AVX2) as well as TSX (this is actually a feature that enterprises like).

TSX didn't work on Haswell. And neither did it on Broadwell. Not easy to get it right it seems.