Intel HEDT Platform to be Forked into Z399 and X599

[Upgrayedd]

Z399 is a step up from X299, X599 is a step up from both. With X599+LGA3647 around, Z399/X299+LGA2066 won't qualify as "extreme" (because there's a more powerful platform above it) and hence Z399 doesn't have an "X" in the name.

So Z399 is a mild refresh seemingly while X599 is a whole new socket designed for XCC only.

 

[XXL_AI]

so.. intel is getting ready to ... its customers more. your era is over intel.

 

[jabbadap]

And they would be called Cascade Lake X not some new "Sky Lake X". Sure there's no much of changes to SkyLake X(some spectre, meltdown HW+SW patches), but it's from different manufacturing process so different Sky Lake sub-architecture... But yeah I presume that earlier Intel sheet about x399 were renamed to z399 and x599 is just server platform on HEDT clothes.

 

[Deleted member 158293]

This is pretty funny actually...

can you see which department has been, and currently still is running Intel right now...

Engineering or Marketing?

 

[LiveOrDie]

Z399 is a step up from X299, X599 is a step up from both. With X599+LGA3647 around, Z399/X299+LGA2066 won't qualify as "extreme" (because there's a more powerful platform above it) and hence Z399 doesn't have an "X" in the name.

They should just do what they did for there CPUs not this crap with replacing X with Z, it really should be X399 and XE599 and call it a day. Why would some one sell there extreme X299 and buy a main stream Z399 it just sounds gay lol.

 

[AnarchoPrimitiv]

Intel is using every finger to plug the holes in the dam, and they're going to run out of fingers soon...

 

[R0H1T]

Intel is using every finger to plug the holes in the dam, and they're going to run out of fingers soon...

Sadly they have enough money to apply a bandaid, or puncture multiple holes in AMD's (recovery) dam.

 

[Slizzo]

So am I not going to get an 8 series option on my X299 then?

 

[efikkan]

These are exciting times, finally there will be some rivalry in the HEDT space.

I would prefer if Intel made one HEDT platform, not two. The X/SP dies are the same, there is no technical reason why they couldn't stick with one socket like AMD did with Threadripper and Epyc.

Still, Cascade Lake-X will bring hardware fixes for Spectre/Meltdown and will be moved to the 14nm++ node.

 

[GoldenX]

These are exciting times, finally there will be some rivalry in the HEDT space.

I would prefer if Intel made one HEDT platform, not two. The X/SP dies are the same, there is no technical reason why they couldn't stick with one socket like AMD did with Threadripper and Epyc.

Still, Cascade Lake-X will bring hardware fixes for Spectre/Meltdown and will be moved to the 14nm++ node.

Die size, Intel's monolithic design is bigger.

 

[jabbadap]

Die size, Intel's monolithic design is bigger.

+1. I doubt XCC chips would fit on lga2066 socket. Maybe if they would have been using lga3647 for HEDT from the beginning, then it would have been quite trivial to just add XCC processors in HEDT. But they were surprised by AMD TRs and had already segmented for different sockets for server and HEDT.

 

[CheapMeat]

"I enjoy confusing market segmentation" said no one ever.

 

[efikkan]

+1. I doubt XCC chips would fit on lga2066 socket. Maybe if they would have been using lga3647 for HEDT from the beginning, then it would have been quite trivial to just add XCC processors in HEDT. But they were surprised by AMD TRs and had already segmented for different sockets for server and HEDT.

Skylake-X/-SP exists in three die sizes; 10, 18 and 28 cores. If it's true Intel can fit >18 cores on LGA 2066, then they can fit the full 28. If not, 18 will be the maximum.

Currently Intel have these high-end platforms:
Consumer HEDT: (X299): 10 and 18 core dies
Xeon W (C422): 10 and 18 core dies
Xeon Scalable (C621): 10, 18 and 28 core dies

The dies themselves are the same, they are compatible with all of these. Rumors claim Intel will end up with the consumer HEDT spread across two sockets, which is completely unnecessary.

 

[yeeeeman]

X599 platform will be monstrously expensive given Intel's track record. But the real wallet breaker will be the CPUs. Intel currently has its 28 core Xeon CPU priced at 8700$ (this is the cheapest one). How much you think they are willing to subtract from that? Before answering, think about this: a 28 core xeon 8180 which has 2.5 base and 3.8 boost is somewhat slower compared to AMD 2990WX which has 3Ghz base and 4.2 boost, all at 1800$. So Intel needs to sell you a 28 core Xeon which runs at least 4.0 Ghz boost to give you equal performance to AMD product and costing in the range of 1800$. I don't this becoming a reality.
They will push the CPU as far as it can get with 250W tdp (I expect 4.5Ghz turbo), so it will be marginally faster than what AMD has, but they will ask top dollars for it - I am guessing 3000$, maybe 2500$ if they have a very good day.

 

[lemans24]

I'm glad Intel has a sturdy plan in place to target an audience of a dozen or so people. AMD was so silly and decided to target the thousands of others with their products. They really missed out on that niche of a niche's niche.

You are so, SO wrong...there are literally many thousands that can use an all-28core 5ghz chip OFFICIALLY sanctioned from Intel, like options traders, high frequency traders, render farms, any engineering workstation that runs algorithms that takes days to run, etc...

Not everyone will use this new chip for gaming...it will be a workstation class chip too

If so, this is another case of Intel being silly.

The ultra core count market can still easily be serviced by Xeon and workstation boards, and focus on high speed + consumer features up to 16 cores to match threadripper. Splitting this into three consumer platforms is absolutely silly.

Why does Porsche have 20 plus different versions of there 911 when they only have 10,000 buyers a year.
Because it allows them to seel another 100,000 cars that are basically rebadged VW's !!!
Porsche is the MOST profitable car company on the planet...AMD is more like Tesla!!!

Come on...this naming convention is easy:
X = extreme (911)
Z = everything NOT extreme

Skylake-X/-SP exists in three die sizes; 10, 18 and 28 cores. If it's true Intel can fit >18 cores on LGA 2066, then they can fit the full 28. If not, 18 will be the maximum.

Currently Intel have these high-end platforms:
Consumer HEDT: (X299): 10 and 18 core dies
Xeon W (C422): 10 and 18 core dies
Xeon Scalable (C621): 10, 18 and 28 core dies

The dies themselves are the same, they are compatible with all of these. Rumors claim Intel will end up with the consumer HEDT spread across two sockets, which is completely unnecessary.

Why does it have to be completely necessary?? Intel seems to think that their are TWO high end markets: overclocking gamers and workstations.
They both want high speed but but in different ways. I like what AMD is doing as it is doing the BEST for the market which...COMPETITION.
I do not not want ONE cpu from ONE company that does EVERYTHING...do you??
I love capitalism because usually competition will bring the best products to the market and more importantly take crappy products OFF the market.
AMD is in fact doing Intel a favor and if you look at Intel's track record over 50 years, they have beaten nearly EVERY competitor on the market with their lousy x86 based architecure.
X86 is a terrible design but it has one great feature: EVERY one buys.

We need competition!!!

 

[WikiFM]

Will any of these two new chipsets be released with Cascade Lake X? Or is Cascade Lake not ready yet?

 

[lemans24]

X599 platform will be monstrously expensive given Intel's track record. But the real wallet breaker will be the CPUs. Intel currently has its 28 core Xeon CPU priced at 8700$ (this is the cheapest one). How much you think they are willing to subtract from that? Before answering, think about this: a 28 core xeon 8180 which has 2.5 base and 3.8 boost is somewhat slower compared to AMD 2990WX which has 3Ghz base and 4.2 boost, all at 1800$. So Intel needs to sell you a 28 core Xeon which runs at least 4.0 Ghz boost to give you equal performance to AMD product and costing in the range of 1800$. I don't this becoming a reality.
They will push the CPU as far as it can get with 250W tdp (I expect 4.5Ghz turbo), so it will be marginally faster than what AMD has, but they will ask top dollars for it - I am guessing 3000$, maybe 2500$ if they have a very good day.

If you look at the sales records for Threadripper 1 and 79xxx, they both TOGETHER seem to be taking about only 5% of the consumer market meaning AMD and Intel are splitting the HEDT market but the
big difference is AMD sells their chip for their threadripper 1 for half the price 79xx on average so Intel has no push to lower its price when they are selling as many 79xx chips as AMD Threadripper 1.

For Threadripper 2990WX, Amd has got smart and is selling them for a little lower than the 7980xe and will definitely sell more as it has way better performance at similar price as the 7980xe.
Intel will be coming out with a 2990wx competitor which will one again be priced way more than the 2990WX and will sell just as many!!!

The list price for the 28-core Xeon includes the ability to be put in compuer with up to 8 sockets which would of course be a very high end computer and therefore Intel can charge a sky-high price.
Since the consumer 28 core is targeted at a different customer, they can price it at whatever price level they want. The 8180 does not cost $8700 but if you want a SINGLE one, you will pay $8700!!!
Believe me, NO ONE who is buying the Xeon 8180 is paying $8700 in lots of 1000 or more and enterprise customers never pay list price anyway.

AMD will be charging $5000( or more) for their high-end Epyc 2 when it comes out, as they are targeting customers who are willing to pay that price for the performance it offers...

 

[efikkan]

Why does it have to be completely necessary?? Intel seems to think that their are TWO high end markets: overclocking gamers and workstations.

Some market segmentation is necessary, but when pushed too far it hurts the selection of motherboards, CPU coolers etc.

I would expect that Intel continue the tradition of making "every possible permutation" of the CPUs within the HEDT range, even though I think a 12-core, 16-core, 20-core, 24-core and 28-core would have been plenty.

Overclocking on HEDT is pointless, there are no longer any substantial gains without extreme cooling and short lifetime for the CPU. Those who needs these HEDT CPUs will not be willing to sacrifice stability, lifetime and warranty for a few percent extra performance. Having overclocking as a hobby is fine, but overclocking HEDT is pretty wasteful.

 

[yeeeeman]

The list price for the 28-core Xeon includes the ability to be put in compuer with up to 8 sockets which would of course be a very high end computer and therefore Intel can charge a sky-high price.
Since the consumer 28 core is targeted at a different customer, they can price it at whatever price level they want. The 8180 does not cost $8700 but if you want a SINGLE one, you will pay $8700!!!

With this I don't agree - partially. The Xeon 8180 might have some features like the 8 sockets and many memory channels, but the matter of the fact is that it has a HUGE die, with very low probability of yield (meaning 100% functional), so they cannot sell it under a specific price, even if they would want, cause they would lose money. Given that it has 8 billion transistors and this is not a GPU (so, transistors are not that tightly packed together), I would suppose it is quite close to the biggest size the 14nm process supports.
Lets take the following example: if the 8700K which is a 6 core CPU sells for 350$ at best, a 28 core version of it would be 28/6 * 350 ~= 2000$. Given the fact that this is a niche product, it doesn't have yields as good as a 8700K chip would have, etc, etc, it won't be 2 grand for sure. If they have amazingly good yields (which they might have by now since 14nm is mature), they could sell it for 3 grand.
But I agree that enthuasiats (which are also targeted by 2990WX) might be willing to pay the extra money just to get the Intel solution since it has better compilers, better support in scientific apps, better latency, etc. In any case, by the time Intel launches its 28 core super HEDT platform, AMD will already have its Rome silicon ready with all the benefits that come (7nm, low power consumption, better IPC, hopefully better memory latency, etc) so, the Intel platform might not do so well as they as planning. But we'll see.

 

[efikkan]

With this I don't agree - partially. The Xeon 8180 might have some features like the 8 sockets and many memory channels, but the matter of the fact is that it has a HUGE die, with very low probability of yield (meaning 100% functional), so they cannot sell it under a specific price, even if they would want, cause they would lose money. Given that it has 8 billion transistors and this is not a GPU (so, transistors are not that tightly packed together), I would suppose it is quite close to the biggest size the 14nm process supports.

Skylake-SP 28-core (698mm²) is still smaller than TU102(754mm²) produced at TSMC "12nm", and considering Intel's 14nm++ is better than TSMC's "12nm" in performance and reliability, they should have no issues at all. It all comes down to production volume. Currently these wafers are only produced in very low volumes, but this will change with X599. Production cost per die will not be a problem for CPUs costing >>$1000.

 

[yeeeeman]

You have to keep in mind that when comparing a CPU die with a GPU die, you compare apples to oranges. The logic inside a GPU usually consists of a standard block which is multiplied over and over. In a CPU, the logic is much more complex, so the design, verification, layout and testing is much more complex, requires more man hours, more money and usually is more prone to yield issues, since you don't have any room for mistakes.
Sure, the real cost of making a Skylake-SP 28-core might not be more than a few hundred bucks for Intel, but given all the other costs that I have mentioned, plus the capacity issue and the risk of production, it might very well cost somewhere in the thousands of dollars to actually fab a CPU like this. They also won't ramp up the production volume for these high count cores, because there won't be so many people who will buy them, so still, cost will be high.

 

[efikkan]

You have to keep in mind that when comparing a CPU die with a GPU die, you compare apples to oranges. The logic inside a GPU usually consists of a standard block which is multiplied over and over. In a CPU, the logic is much more complex, so the design, verification, layout and testing is much more complex, requires more man hours, more money and usually is more prone to yield issues, since you don't have any room for mistakes.

You do know the Skylake-SP die includes 28 duplicate cores and 6 duplicate memory controllers, right?
CPU dies also include more cache, which have higher tolerances for yields.
TSMC makes high volumes of similar TDP and similar sized dies, I don't see why Intel can't with a better node.

Sure, the real cost of making a Skylake-SP 28-core might not be more than a few hundred bucks for Intel, but given all the other costs that I have mentioned, plus the capacity issue and the risk of production, it might very well cost somewhere in the thousands of dollars to actually fab a CPU like this. They also won't ramp up the production volume for these high count cores, because there won't be so many people who will buy them, so still, cost will be high.

I don't see the problem. Intel know very well the cost of producing wafers. Unless they cancel these products, these will sell thousands. And BTW, Intel usually make excess CPUs for commercial customers anyway.