Intel 10th Gen Core X "Cascade Lake-X" Pricing and Specs Detailed

[Vayra86]

It's funny, desktop CPUs and performance has ALWAYS been the metric which computer enthusiasts and nerds have judged either AMD or Intel by. Now that Intel can't compete there or are perceived to be on the back foot in this segment, suddenly all that matters is mobile and servers.

Mobile!! If you told me 5 years ago that mobile CPU performance and sales would be the main argument for Intel fans defending the company I wouldnt believe you! That's how ridiculous that is. With these HEDT CPUs we're talking about high-end desktop, let's not pretend Intel weren't derided for years trying to improve efficiency and PPW over pure performance and more cores, because they were. Maybe some here only started following CPUs when Ryzen came out and hence weren't around for this.

Now that AMD, not Intel, has brought more cores and wins in pure performance, the people that were lambasting Intel for their mobile focus have changed tact and say it's this mobile performance (and servers!) which show the company is superior or in a great position. You couldn't make it up.

Eh... that is just your color of glasses being in the way I suppose.

When it comes to Intel's market share and leadership then yes, the money is, was and has been in server and mobile for quite some time now. The desktop PC market is dwindling year after year and this has never been a secret. So in the context of 'Intel is screwed'... this definitely applies and it did apply the last ten years. Those ten years, AMD couldn't make a dent in either of those markets OR the desktop.

Today, AMD makes a comeback on desktop and will make a killing on server sooner rather than later. Mobile however... they still don't have anything ground breaking and Intel is very competitive. But mobile is not the cutting edge. The cutting edge STILL is desktop and HEDT, because that is where the trickle down for mobile begins. For consumer, desktop is the point of reference, free of thermal and power constraints.

Even today Intel still has a separate design for HEDT and for MSDT, and while this makes them less flexible and efficient wrt yields, they do have a much better optimized mobile portfolio. AMD has work to do here (which they are doing).

Bottom line, judging the facts and numbers is not choosing a camp, its just trying to get a good picture of the market. There's nothing to win here by choosing sides, you only lose because you'll be oblivious to what really happens.

Regardless of how nice Zen is, the numbers don't lie and Intel has the market on lockdown, still, in every volume segment there is.

 

[Patriot]

You should wait for benchmarks of actual workloads, not synthetics like Geekbench etc. which has leaked this far.
The problem for high core count Cascade Lake will be heat. As long as they can stay within similar clocks, they will remain competitive.


Nope, they were released several months ago, but those prices, yuck.

Cinebenchr15 is leaked as well, but yes it tends to not care about latency as much which is where DAW and like workloads will most likely fall in intel's favor.
I brain farted at 2am when I posted that... yes Xeon-w was released at Xeon prices in june but there have been more recent mentions of intel releasing i9 on the socket and there appears to be a 26 and 28c HEDT waiting in the wings to take on threadripper that is not the 3k chip. Competition is good.

 

[MazeFrame]

Will be interesting to see the i9-10920XE (what the hell is that naming scheme?) against the 3900X.
Intel has double the L2 cache, but 1/3rd the L3 cache.
48 PCIe lanes, again? Is the fear of cutting into their server segment so big?
24 PCIe lanes from the PCH -> Bottleneck all your storage.

Just noticed, the 10 core (10900xe) is price wise up against the 3900x. I think the winner from that is clear.

 

[efikkan]

Cinebenchr15 is leaked as well, but yes it tends to not care about latency as much which is where DAW and like workloads will most likely fall in intel's favor.

That kind of illustrates my point. Cinebench is a benchmark of Cinema4D, which is relevant for those running Cinema4D.
But the mistake that many do is to extrapolate "general performance" of various CPUs based on Cinebench, which is very misleading considering Cinebench scales very differently from the CPU's actual "general performance" across a variety of workloads.

 

[Valantar]

Cinebenchr15 is leaked as well, but yes it tends to not care about latency as much which is where DAW and like workloads will most likely fall in intel's favor.
I brain farted at 2am when I posted that... yes Xeon-w was released at Xeon prices in june but there have been more recent mentions of intel releasing i9 on the socket and there appears to be a 26 and 28c HEDT waiting in the wings to take on threadripper that is not the 3k chip. Competition is good.

Intel can't go beyond 18 cores in socket 2011 - there isn't room for a bigger die in that package.

This is the 9980XE, which uses the 18-core HCC die. It is slightly below 500mm2 (around 21.6x22x4mm).

This is the Xeon W-3175X, which uses the 28-core XCC die. It is nearly 700mm2, at 21.6x32.3mm. There is no way that die can fit onto a Socket 2011 substrate - it's simply too large.

Which is why the 28-core is on socket LGA3647. Fitting that die onto a smaller substrate would not only require disabling a significant amount of I/O (due to loss of pins) but also redesigning the IHS and socket retention mechanism entirely. Not something Intel is likely to do.

They might be able to squeeze more cores in on 10nm, but that isn't arriving for chips this size any time soon. Yields would be utter garbage.

 

[Shatun_Bear]

Eh... that is just your color of glasses being in the way I suppose.

When it comes to Intel's market share and leadership then yes, the money is, was and has been in server and mobile for quite some time now. The desktop PC market is dwindling year after year and this has never been a secret. So in the context of 'Intel is screwed'... this definitely applies and it did apply the last ten years. Those ten years, AMD couldn't make a dent in either of those markets OR the desktop.

Oh gee wow, really? Thanks for your great insight and this amazing revelation, I've never heard it before...

This topic is about Intel's HEDT. I've got a simple simon revelation for you now: HEDT means HIGH END DESKTOP, I couldn't care less how many dual-core mobile laptops Intel sells to insurance businesses, in the HEDT space, Intel is screwed so please try to come to terms with it.

 

[efikkan]

<snip>
They might be able to squeeze more cores in on 10nm, but that isn't arriving for chips this size any time soon. Yields would be utter garbage.

The largest known die for Ice Lake-SP so far is 26 cores with 8 memory controllers.
I would not be surprised if larger core configurations would consist of two dies.

 

[Valantar]

The largest known die for Ice Lake-SP so far is 26 cores with 8 memory controllers.
I would not be surprised if larger core configurations would consist of two dies.

8 memory controllers sure doesn't sound like something suited to a 4-channel socket with just 2011 pins. Intel needs 3647 pins for 6 channels, AMD needs more than 4000 for eight, so that die definitely isn't destined for normal HEDT - unless they follow AMD's lead with Threadripper and use a trimmed-down version of their highest end socket for HEDT, of course.

 

[efikkan]

8 memory controllers sure doesn't sound like something suited to a 4-channel socket with just 2011 pins. Intel needs 3647 pins for 6 channels, AMD needs more than 4000 for eight, so that die definitely isn't destined for normal HEDT - unless they follow AMD's lead with Threadripper and use a trimmed-down version of their highest end socket for HEDT, of course.

The socket for Ice Lake-SP has been known to be LGA-4198 for over a year now.
What this means for Ice Lake-X (and -W?) variants remain unknown.

 

[Valantar]

The socket for Ice Lake-SP has been known to be LGA-4198 for over a year now.
What this means for Ice Lake-X (and -W?) variants remain unknown.

That is true, but I don't think it's likely for Intel to make two separate XCC dice, so the only option then would be to use a version with at least half the memory controllers and a heap of PCIe disabled for HEDT. It is possible (that's what AMD has been doing, after all) but I think quite unlikely. That would entail quite the loss of profits compared to selling the same silicon as fully enabled LGA4198.

 

[Vayra86]

Oh gee wow, really? Thanks for your great insight and this amazing revelation, I've never heard it before...

This topic is about Intel's HEDT. I've got a simple simon revelation for you now: HEDT means HIGH END DESKTOP, I couldn't care less how many dual-core mobile laptops Intel sells to insurance businesses, in the HEDT space, Intel is screwed so please try to come to terms with it.

You're way overinflating things. Chill out. As much as Intel has a problem today for a prospective buyer of an HEDT system, most people are still sitting on an Intel rig with little urge to upgrade. When they do, they will likely consider Ryzen or TR. Its a slow crawl to get more market share though. As much as Intel doesn't have the top offering, they still have the largest market share by far, and it will take many years for that to change. Many years in which 'Intel is NOT screwed', they can get by just fine. Many years in which they have time to find something to counter AMD.

Get it now?

 

[efikkan]

That is true, but I don't think it's likely for Intel to make two separate XCC dice, so the only option then would be to use a version with at least half the memory controllers and a heap of PCIe disabled for HEDT. It is possible (that's what AMD has been doing, after all) but I think quite unlikely. That would entail quite the loss of profits compared to selling the same silicon as fully enabled LGA4198.

As you know, for Skylake-SP/X and Cascade Lake-SP/X intel makes three dies; LCC(10 core), HCC (18 core) and XCC(28 core), all of which have 6 memory controllers. From these they make the Xeon scalable platform(Skylake-SP and Cascade Lake-SP) called Bronze, Silver, Gold and Platinum, which support various multi-socket configurations. For the HEDT/workstation market they have three partially overlapping platforms; X299, C422(LGA2066) and C621(LGA3647), but only X299 have a decent motherboard selection.

There is no reason why a HEDT CPU with all memory controllers would compete with Xeon SP. HEDT is optimized for maximum performance and use "normal" ATX style motherboards, while Xeon SP is energy efficiency optimized and usually run on other motherboard form factors suitable for server racks.

I would prefer if Intel made one HEDT/workstation lineup like AMD does today, and enabled as many memory channels and PCIe lanes as possible. If they really have to sell some CPUs without all memory channels or ECC for yield issues, then they can sell these as cheaper CPUs on the same platform.

 

[Patriot]

Intel can't go beyond 18 cores in socket 2011 - there isn't room for a bigger die in that package.

This is the 9980XE, which uses the 18-core HCC die. It is slightly below 500mm2 (around 21.6x22x4mm).
This is the Xeon W-3175X, which uses the 28-core XCC die. It is nearly 700mm2, at 21.6x32.3mm. There is no way that die can fit onto a Socket 2011 substrate - it's simply too large.
Which is why the 28-core is on socket LGA3647. Fitting that die onto a smaller substrate would not only require disabling a significant amount of I/O (due to loss of pins) but also redesigning the IHS and socket retention mechanism entirely. Not something Intel is likely to do.

They might be able to squeeze more cores in on 10nm, but that isn't arriving for chips this size any time soon. Yields would be utter garbage.

I didn't say they were going to put more cores on 2066, btw 2011-3 had 22 cores...
I said Intel plans to replace the 3175X either bringing i9 to 3647 or making another high binned Xeon-W
Guessing they will be using some of the dies they binned for cooperlake to try and stave off Threadripper 3000 total domination at a more reasonable TDP.

But in the June cascade lake refresh they did not release a 3175x replacement.

 

[efikkan]

This information about Tiger Lake have apparently been in circulation for a while but have completely escaped me. So at least they are preparing the graphics driver for potential desktop parts, without any guarantee of when or if they will actually release. Intel have stated that Ice Lake-SP will ship in Q2 2020, so hopefully we can see desktop and HEDT parts later in 2020 too.

 

[notb]

This topic is about Intel's HEDT. I've got a simple simon revelation for you now: HEDT means HIGH END DESKTOP, I couldn't care less how many dual-core mobile laptops Intel sells to insurance businesses, in the HEDT space, Intel is screwed so please try to come to terms with it.

Since we look at 2066 mostly to learn what features could at some point go into mainstream (e.g. AVX-512, DLBoost lately), this is also how we should perceive Intel's troubles here. Does this concern us? Does it have a meaningful impact on the stuff Intel really wants to sell?
And the answer is: not really. AMD's advantage here stems from a bigger socket (SP3 vs 2066) and a smaller node, so they're able to fit more cores. That's it.

Keep in mind LGA2066 is just "high-end consumer" in the Intel's current lineup. They're not selling Xeons on this socket anymore (they're either LGA1151 or LGA3647).

If you just want to discuss HEDT as a separate lineup... there isn't really much to talk about. It's tiny and irrelevant. Even if 2066 just stopped selling today, it wouldn't have any impact worth discussing.

 

[Gungar]

I mean, better pricing then the price gauging going on before (like Nvidia is doing atm) is good yes... but its odd to use the word "impressive" for it imo.
I mean, they are still making a lot of profit on these so why would you be impressed by them asking less money then they were before?
I guess you could be impressed by AMD to actually force Intel to start cutting down their prices?

Not trying to make it a whole debate though, but for example imo it would be impressive if they could do those 14 cores while consuming less power then a 9700k or something, that would be a technological leap, that would be impressive.

anywho, moving on

They forced Intel to go cheaper architecture, i am not impressed by that.

 

[efikkan]

Keep in mind LGA2066 is just "high-end consumer" in the Intel's current lineup. They're not selling Xeons on this socket anymore (they're either LGA1151 or LGA3647).

Actuall, Intel just updated the lineup today.
Xeon W-32XX is LGA3647
Xeon W-22XX is LGA2066
and still Xeon E-22XX is LGA1151
So the confusion continues…

 

[notb]

Actuall, Intel just updated the lineup today.
Xeon W-32XX is LGA3647
Xeon W-22XX is LGA2066
and still Xeon E-22XX is LGA1151
So the confusion continues…

Yes! I did not see that coming. :-D

Actually it takes some confusion away. The principle stands: Xeons are for production use (mission critical systems). Everything else does consumer + business (non-production).

Intel basically said: we can't put as many cores as we want on LGA1151 successor and OEMs told them LGA3647 is just too big.
I bet this wouldn't happen if Intel could put 16 cores on the small socket, but that's at least a year away.

 

[Chrispy_]

Curious to see the TDP on these, and if the TDP is reasonable, how long they can actually spend at boost clocks.

When you start talking about large chips like these, power consumption is the key bottleneck and 14nm just doesn't seem to be cutting it anymore, no matter how many plus symbols you add after the 14.

 

[jayjr1105]

Curious to see the TDP on these, and if the TDP is reasonable, how long they can actually spend at boost clocks.

When you start talking about large chips like these, power consumption is the key bottleneck and 14nm just doesn't seem to be cutting it anymore, no matter how many plus symbols you add after the 14.

Doesn't matter. Intel TDP is useless. A 9900K is listed at 95w but will pull well over 200w on an AVX workload.

 

[Chrispy_]

Doesn't matter. Intel TDP is useless. A 9900K is listed at 95w but will pull well over 200w on an AVX workload.

True, but this is AMD we're talking about. Their TDP is actually both accurate, real-world representative and user-configurable, especially when it comes to laptops with additional sensors and things like STAPM (skin-temperature aware power management).

TDP for laptops really matters, because thermal headroom and cooling is any laptop's biggest performance constraint.

 

[HwGeek]

So if the latest leaks are correct then looks like AMD gonna Troll Intel and announce their new TR lineup just before those 10th gen x299 CPU's reviews go live, it will make them DOA ;-).

AMD won't miss the one time opportunity to step on Intel and continue with their "New Leader-New Rules" ha?.

 

[Super XP]

AMD's Threadripper HEDT platform should be launching just before or during this Intel 10th gen release time frame.
AMD should easily steal the thunder with better products, pricing, speed & performance.

 

[HwGeek]

I hope they will offer cTDP options so many S.I and custom workstation builders could use higher TDP option for better performance w/o voiding warranty.
So you you build a great workstation with great cooling you could bump the cTDP option for 300W~380W and get much higher clock.
This way it will be benched with this high performance and will be offered by System integrator's - same way they are OK with using Intel's MCE on since it's not voiding the warranty.

 

[Slizzo]

Intel increased the PCIe 3.0 lanes by 4 or are they counting in a different way?

Nope, it does look that way.