Intel Skylake / Kaby Lake

[Shivansps]

Sure forget the fact that it took AMD making a move to force Intel above 12 cores.. but even now that they are forced to compete in a market segment AMD created.. they are doing everything they can to save a penny. Amazing.

Sure... they had 18C dies on a non-server socket waiting in the back just in case AMD had something. May i remind you that X99 had up to 22C on client and now that HEDT and Xeons are running on diferent sockets is not trival to bring a higher SKU? Is not like AMD TR and Naples that are kinda sharing the socket. Just stop what you are doing, is getting silly at this point.

AMD may have stopped Intel for charging about $3000 instead of $2000, nothing more, YET, Intel still increased the price of the top SKU by adding the $2000 tier.
if i look at $1000 tier HEDT has been getting steady 2C increases. Check $1000 HEDT cpus of Nehalem->SB-E->HSW-E->SKL-X.

Of the TIM thing, yes, they did cheaper it out, but im going to wait for actual reviews and OC results, to see how much it does affect.

 

[LTC8K6]

Its absolutely, stunningly mind blowing that Intel cheaped out on the TIM and didn't solder the IHS on these "high end" products. https://www.youtube.com/watch?v=I1Bv8Mxnnlc

Sure forget the fact that it took AMD making a move to force Intel above 12 cores.. but even now that they are forced to compete in a market segment AMD created.. they are doing everything they can to save a penny. Amazing.

It's called business. Both Intel and AMD shave everything down to the penny. Both Intel and AMD cheap out as much as they absolutely possibly can. It's called business.

 

[swilli89]

It's called business. Both Intel and AMD shave everything down to the penny. Both Intel and AMD cheap out as much as they absolutely possibly can. It's called business.

I am so tired of this. Is that your only defense? You just defend profit maximizing no matter what companies do? No matter how much they gouge price or cheap out on materials?

You (a consumer and enthusiast) simply just say "Its called business"? What's the point of any of this discussion if that's your response. We are talking about a CPU that costs a few grand. Its marketed to overclockers. But Intel wants to save a few cents per $2,000 chip. Now those same people have to void their warranty to get the cooling ability that they should be demanding out of the box.

Do you have any actual thoughts as to the issue that I highlighted?

 

[tamz_msc]

Why are you suddenly discussing Zen and Haswell? All I said was that scaling core throughput (by clock frequency) 2x without increasing bandwidth will never yield 2x the performance. Why are you so defensive about this?

I'm yet to get a response from you regarding your purported benchmarks, on the other hand, I'm just mentioning that Zen and Haswell results from FFTW are available if anyone wants real data, just as an example.

I'm still waiting for your data, because without it I cannot verify any of your claims regarding cache frequency.

 

[swilli89]

Sure... they had 18C dies on a non-server socket waiting in the back just in case AMD had something. May i remind you that X99 had up to 22C on client and now that HEDT and Xeons are running on diferent sockets is not trival to bring a higher SKU? Is not like AMD TR and Naples that are kinda sharing the socket. Just stop what you are doing, is getting silly at this point.

AMD may have stopped Intel for charging about $3000 instead of $2000, nothing more, YET, Intel still increased the price of the top SKU by adding the $2000 tier.
if i look at $1000 tier HEDT has been getting steady 2C increases. Check $1000 HEDT cpus of Nehalem->SB-E->HSW-E->SKL-X.

Of the TIM thing, yes, they did cheaper it out, but im going to wait for actual reviews and OC results, to see how much it does affect.

No... For a decade the highest core count HEDT chip was maxed out from whatever the LCC die was. Now.. after all these generations where they could have ALWAYS used their HCC chip.. it just magically coincides with the launch of ThreadRipper and it magically just beats it out on 2 cores? Why are people so against AMD that they can't even admit that they forced Intel to compete with higher core counts?

 

[DrMrLordX]

Sure... they had 18C dies on a non-server socket waiting in the back just in case AMD had something.

I thought Skylake-W was exactly that?

 

[blue11]

I'm yet to get a response from you regarding your purported benchmarks, on the other hand, I'm just mentioning that Zen and Haswell results from FFTW are available if anyone wants real data, just as an example.

I'm still waiting for your data, because without it I cannot verify any of your claims regarding cache frequency.

What are you trying to verify? That HEDT/server CPUs have stock L3 frequencies of 2.4 GHz compared to client 3.6 GHz? You can download CPU-Z or XTU and see that. That increasing core frequency by huge amounts (2x) without increasing uncore frequency leads to bottlenecking? You can underclock your uncore by half and run your own benchmarks. I don't have a tank of liquid nitrogen sitting next to me, so I can't "prove" to you that extreme OC reaches uncore limits, but it should be obvious.

 

[Phynaz]

No... For a decade the highest core count HEDT chip was maxed out from whatever the LCC die was. Now.. after all these generations where they could have ALWAYS used their HCC chip.. it just magically coincides with the launch of ThreadRipper and it magically just beats it out on 2 cores? Why are people so against AMD that they can't even admit that they forced Intel to compete with higher core counts?

Why can't you admit that CPU design cycles are years long, and Skylake-X is not a response to Threadripper at all?

 

[blue11]

Why can't you admit that CPU design cycles are years long, and Skylake-X is not a response to Threadripper at all?

The chips themselves are not a response to anything, but the pricing and SKU segmentation strategy are clearly a response to AMD. Intel did not have to release any 12-18C "desktop" SKUs, even though they could have done so as far back as Haswell. They did so in response to "ThreadRipper", so that AMD could not claim to have the biggest HEDT chip.

 

[RichUK]

Why can't you admit that CPU design cycles are years long, and Skylake-X is not a response to Threadripper at all?

Well, it could be. They are just rebranded Xeon's after all (with reference to the 18c model).

 

[DrMrLordX]

The chips themselves are not a response to anything, but the pricing and SKU segmentation strategy are clearly a response to AMD. Intel didn't have to release any 12-18C "desktop" SKUs, even though they could have done so as far back as Haswell. They did so in response to "ThreadRipper", so that AMD couldn't claim to have the biggest HEDT chip.

Actually, the pace for 10c was set with Broadwell-E, so your statement isn't entirely sensible. The logical conclusion is that Intel had set aside Skylake-W dice for Skylake-X with the plan being to sell up to 12c in the short term and maybe scale up to 18c later to compensate for product delays (there will be more, probably). They went up to 18c in the short term to compete with Threadripper.

 

[tamz_msc]

What are you trying to verify? That HEDT/server CPUs have stock L3 frequencies of 2.4 GHz compared to client 3.6 GHz? You can download CPU-Z or XTU and see that. That increasing core frequency by huge amounts (2x) without increasing uncore frequency leads to bottlenecking? You can underclock your uncore by half and run your own benchmarks. I don't have a tank of liquid nitrogen sitting next to me, so I can't "prove" to you that extreme OC reaches uncore limits, but it should be obvious.

Show me the benchmarks that you're referring to that measures core clock vs uncore scaling. Don't need LN2 speeds to show its effect.

 

[jpiniero]

No... For a decade the highest core count HEDT chip was maxed out from whatever the LCC die was. Now.. after all these generations where they could have ALWAYS used their HCC chip.. it just magically coincides with the launch of ThreadRipper and it magically just beats it out on 2 cores? Why are people so against AMD that they can't even admit that they forced Intel to compete with higher core counts?

TBH Basically Intel only really intends to sell the highest HEDT core to the e-penis people; and yes Threadripper is forcing them to make it 18 cores instead of 10/12. Performance doesn't really matter to these people, it's all about the e-penis. Intel doesn't actually expect to sell any of the 14 and 16 core models.

 

[Topweasel]

Well, it could be. They are just rebranded Xeon's after all (with reference to the 18c model).

That's the big point. No one is suggesting that Intel went out and designed a new die specifically to battle AMD. Intel has had for a long time been making larger core count server CPU's sharing the same socket as the 2011 models for a long time. Intel segments their CPU lineup on purpose. Slow core count increases, socket changes, and so on aren't done for technological reason but because they could. There is a reason that they didn't stop at the base 2011 die and went with a second 18c+ configuration. Again both in development a long time before AMD came out with Ryzen, or admitted the existence of TR. But the segmentation of 2011 chips that Intel had been using was never ever used as HEDT (in terms of branding) before. The 14, 16, and 18c offerings are in response to AMD along with the pricing.

 

[coercitiv]

Why can't you admit that CPU design cycles are years long, and Skylake-X is not a response to Threadripper at all?

Core 2 Extreme 4 cores
Sandy Bridge-E 6 cores
Ivy Bridge-E 6 cores
Haswell-E 8 cores
Broadwell-E 10 cores
Skylake-X 18 cores

Intel was predictable like clockwork until yesterday. Then they jumped by 8 cores instead of 2.

 

[tamz_msc]

Since this thread has turned into a slugfest, and I won't deny that I have also been a participant, lets go back and try to discuss why the need for the revamped cache hierarchy. Well, I found one reason, and it has got to do with AVX2, and more likely AVX512.

Spoiler: Skylake AVX2

Notice how performance falls off a cliff the moment you exceed the L1 limit.
Contrast with Ryzen, looking at low core counts:

Spoiler: Ryzen AVX2

Summing up, the Ryzen CPU cannot keep up with current Intel designs in terms of in-core performance with AVX(2) code. On the other hand, its overlapping cache hierarchy and good memory bandwidth (for a desktop chip) can regain a lot of lost ground despite the non-scalable L3 cache.

 

[Shivansps]

No... For a decade the highest core count HEDT chip was maxed out from whatever the LCC die was. Now.. after all these generations where they could have ALWAYS used their HCC chip.. it just magically coincides with the launch of ThreadRipper and it magically just beats it out on 2 cores? Why are people so against AMD that they can't even admit that they forced Intel to compete with higher core counts?

That could have been true on X99, they just needed to rebrand some Xeon and thats it.

Now they have 2 (two) physically different sockets, for HEDT and server Xeons, they just cant get a "AMD +2" number of cores out of thin air, if they have a non LCC die on LGA2066 is because they planned for it a long time ago.

Hell for a very long time we trought 12C was gonna be the max for LGA2066, and since LCC max is 10C that is impossible to do, the 12C already needs a non-LCC die. They are doing this not because of AMD, they are doing it because on LGA2066 the max number of cores for a workstation is now 18, on X99 was 22C because you could place any Xeon E5 on a client X99 mb and sell it as a workstation. In fact 18C is still not enoght to match max number of core of BDW-E on X99.

 

[Shivansps]

Core 2 Extreme 4 cores
Sandy Bridge-E 6 cores
Ivy Bridge-E 6 cores
Haswell-E 8 cores
Broadwell-E 10 cores
Skylake-X 18 cores

Intel was predictable like clockwork until yesterday. Then they jumped by 8 cores instead of 2.

They also doubled the price from $1000 to $2000 for the top SKU... at $1000 the core count increase is still 2.

BTW, Nehalem was also 4.

 

[Topweasel]

They also doubled the price from $1000 to $2000 for the top SKU... at $1000 the core count increase is still 2.

BTW, Nehalem was also 4.

The doubling of price happened last gen when it went up to $1700. The $1k chip didn't change core wise at all. The 12c chip is $500 cheaper than the 10c last gen. The $1700 chip this gen has an extra 6 cores. As coercitiv stated. You could set a clock to Intel's actions in the past. They really shook up their HEDT lineup here.

 

[JoeRambo]

Now they have 2 (two) physically different sockets, for HEDT and server Xeons, they just cant get a "AMD +2" number of cores out of thin air, if they have a non LCC die on LGA2066 is because they planned for it a long time ago.

Honestly the fact that there are no details about those "non-LCC" cores ( like cache sizes, clocks) pretty much screams that they are rushed to this socket and this HEDT "content" market.

And that argument about max number of cores in a workstation? Intel was not happy with 2nd hand Xeons and took calculated steps to destroy that market for Skylake gen. Then Ryzen happened.
In fact from absurd product that Kabylake-X is, is we can pretty much infer that Intel was about to destroy unlocked CPUs in desktop market, pushing them to HEDT, but 4Ghz Ryzen happened. Now they have to react by pulling-in Coffee Lake 6C.

 

[RichUK]

Intel were originally planning 10c max (and 44 pci-e lanes) for Skylake-X.

 

[Shivansps]

Honestly the fact that there are no details about those "non-LCC" cores ( like cache sizes, clocks) pretty much screams that they are rushed to this socket and this HEDT "content" market.

And that argument about max number of cores in a workstation? Intel was not happy with 2nd hand Xeons and took calculated steps to destroy that market for Skylake gen. Then Ryzen happened.
In fact from absurd product that Kabylake-X is, is we can pretty much infer that Intel was about to destroy unlocked CPUs in desktop market, pushing them to HEDT, but 4Ghz Ryzen happened. Now they have to react by pulling-in Coffee Lake 6C.

Coffe Lake was planned long time ago, in fact i think we know about it before we knew about Ryzen. You may complain about the details of non-LCC, but if they come out this year AMD had nothing to do with them.

 

[AdamK47]

Is there a photo or video of the 18 core with the IHS removed?

 

[Shivansps]

The doubling of price happened last gen when it went up to $1700. The $1k chip didn't change core wise at all. The 12c chip is $500 cheaper than the 10c last gen. The $1700 chip this gen has an extra 6 cores. As coercitiv stated. You could set a clock to Intel's actions in the past. They really shook up their HEDT lineup here.

yes, it happen the last gen and they did it again now with the $2000 one, yes AMD may have caused not to be more than $2000.

But if a look at $1000, what used to be the top SKU until they increased the price on BDW-E, it gets 2 cores increase. 4 on Nehalem-E, 6 on SB-E, 8 on HSW-E and 10 on SKL-X.

 

[dullard]

You could set a clock to Intel's actions in the past. They really shook up their HEDT lineup here.

If you look at a specific price point, Intel did virtually nothing different here with Skylake. Take the ~$1k chip for example (it is the one of the easiest to compare since Intel has been producing it for a decade).

Core 2 (2007): 4 Cores, 3.00 GHz / no turbo, 12 MB L2
Nahalem (2008): 4 cores, 3.20 GHz / 3.46 GHz, 8 MB L3
Nahalem (2009): 4 cores, 3.33 GHz / 3.60 GHz, 8 MB L3
Westmere (2010): 6 cores, 3.33 GHz / 3.60 GHz, 12 MB L3
Westmere (2011): 6 cores, 3.46 GHz / 3.73 GHz, 12 MB L3
Sandy Bridge (2011): 6 cores, 3.3 GHz / 3.9 GHz, 15 MB L3
Sandy Bridge (2012): 6 cores, 3.5 GHz / 4.0 GHz, 15 MB L3
Ivy Bridge (2013): 6 cores, 3.6 GHz / 4.0 GHz, 15 MB L3
Haswell (2014): 8 cores, 3.0 GHz / 3.5 GHz, 20 MB L3
Broadwell (2016): 8 cores, 3.2 GHz / 3.7 GHz, 20 MB L3
Skylake (2017): 10 cores, 3.3 GHz / 4.5 GHz, 13.75 MB L3

Every 3 to 4 years, Intel has added 2 cores to this ~$1k chip. The base clock is generally in the lower 3 GHz range. The cache has moved around a bit, generally getting bigger, but not always. They did the exact same with Skylake-X, added a couple cores every 3 to 4 years, just like they have been doing like clockwork.

I just don't see a massive change here. Not compared to how much they were usually changing it. What they did do was add more price points, some with less features than before and some with more features. But they started doing that in 2016 with the introduction of a new ~$1700 price point.