> Intel has already shipped ... Ice Lake Xeon Scalable processors to ... its high-profile customers, even though the processors have not yet been launched. This is typical for a server processor ...
Haven't they been saying that since way back in 2019? That's NOT typical of a server processor.
> At present, Intel is already sampling ... Ice Lake processors with customers, and claims that it is on track to ship these CPUs in volumes in the first half of next year.
Oof. And we *still* don't even have a launch date for general availability!
Ouch ... That hurts. Just a month after Cascade. No wonder Chipzillah is using their heft in slide shows and 'personalities' ...
At least, Intel bought 2 years' time between the **new Second Generation Enterprise Xeon Scalable processors, known as Cascade Lake** and now. Sadly, now, the Cascades will be orphaned (I guess) ...
"Sampling" suggests extremely low volume, not the 115K+ Intel has not shipped. Still, a 2-year delay between "sampling" and launch (since the paper launch is in April I assume the real launch will be in May) is obviously unprecedented. And since Ice Lake-SP is to be fabbed on Ice Lake's node I expect the clocks will be very low..
It's very typical of server and HPC chips, and has been for a long time.
Consumer chips appear to pop up fully-formed in a puff of marketing, but server dies have a much longer lead-in before announcement (let alone public availability for a handful that are sold through retail channels). Cascade Lake made it explicit, but these large server chips are designed in concert with the customers who order tens to hundreds of thousands of dies. There's an ongoing exchange with customers describing the workloads they're expecting to encounter over the next few years, and chip makers providing early sample dies to test those workloads on. This is why server CPUs 'launch' in production environments running optimised code. All of this occurs under NDA, so you won't find people detailing the process outside of official press releases (as with Cascade Lake). This isn't just so chipmakers can keep new CPUs secret, it's so customers can keep their future workload predictions secret (if Big Data Processor is known to be wanting to optimise for a certain workload, that provides insights into their desired future product targeting).
If it launched a year ago, like they promised in May, 2019, you'd be on the mark. However, this launch is anything but typical (contrary to what this article suggests).
That just sounds weird and so disconnected from any technical merits of their products that it gives the impression Intel has whipped its marketing and PR departments into overdrive. I smell fear.
Exactly, and in poor taste. It's a bit sad seeing how Marketing has taken over, now that technicals are behind. Same story in the other articles. Marketing gimmicks; smoke and mirrors; trying to save the day.
Very unlikely to see any more Mac models released or updated with Intel chips going forward. Apple will release new Macbook Pro models with their own Apple silicon later this year, then after that iMac and maybe maybe Mac Pro.
(In case you didn't know, the current 13" Macbook Pro, Macbook Air, and Mac Mini all now have Apple's custom designed M1 chip which has been very well reviewed.)
That would only ever have been likely if it used the same platform. Given how long they kept the dustbin going, I doubt Apple would release a new Mac Pro platform only to send it to the junk heap in a year or two when they switch that product line over to ARM.
It's strange how stalled they've been at 14nm and how much they've had to iterate 10nm. It will be interesting to see how Ice Lake Xeon performs in tests, and I don't have high expectations for it. With all their struggles with 10nm, I got the impression that it was only the third or fourth iteration of it that was expected to be any good, the SuperFin and Enhanced SuperFin, so Sapphire Rapids or whatever.
I'm not sure that we'll ever get any exciting leaps from Intel again. Something is wrong, and I don't fully understand it. They can't execute on new process nodes anymore. Does anyone believe that they'll deliver 7nm to mainstream products by 2023? And that it will be excellent in terms of performance and power improvements? They don't seem to be able to do it anymore.
I hope they're able to develop the accelerators and I/O on 10nm+++ or whatever. That would be neat. They haven't done a good job of marketing technology like Quick Assist, the compression and encryption accelerator chip that is available on some of their SKUs. Hardly anyone knows that QA exists. People are wasting huge energy and CPU on compression and encryption when they don't have to, but Intel sucks at marketing. If they got clean-sheet chip designs on 10nm for these accelerators that would be a win. They should also push improved I/O instead of PCIe, technology like OpenCAPI which has much lower latency.
Intel's CPU archs are impossible to keep track of.... So Ice Lake is Sunny Cove 10nm ++ something, something, but Tiger Lake is Willow Cove 10nm++ something else?
So..... Sunny Cove is only for Xeon why? I guess I don't get why there isn't a desktop version of this if 10nm now to the point where they can make chips this big at this performance. Like why bother with Rocket Lake, this has to be better?
Sunny Cove was in Ice Lake U (10xxGx mobile) and is now in Ice Lake SP (Xeons and maybe a Core-X variant?); it's on what most press calls Intel's 10nm+ but Intel just calls 10nm and pretends Canon Lake never happened.
Willow Cove is in Tiger Lake U (11xxGx mobile), should be in Tiger Lake H very soon (11xxxH mobile), and I think it's in Sapphire Rapids (Xeons and maybe a Core-X variant?). That's 10nm SuperFin (or 10nm SF) in Intel's terminology but would have been 10nm++ previously.
Cypress Cove is in Rocket Lake desktop chips (11xxx), and is a backport of Sunny Cove to 14nm.
Golden Cove will be in Adler Lake (12th gen, supposedly for mobile and desktop). That'll be 10nm Enhanced SuperFin or 10 nm+++ if you like that better. If the code name for the Xeon based on it has leaked, I haven't heard it.
Meteor Lake (haven't heard the CPU core code name) in Q1 2023 is supposed to be on 7nm, I think.
AVX-512 is their usual go-to, when they want to gin up some good-looking benchmarks. I think it's also responsible for a fair bit of Rocket Lake's claimed IPC improvements.
who want to name a high-temperature gas cooled reactor "Xe" and I can't tell if they are talking about the Xenon-135 radioactive pollutant or the notorious low performance graphics card architecture. Going high end is like Trabant competing in F1.
Best thing that can happen for those two letters is that Dell buys a bunch of OEM-only graphics cards and discreetly trucks them to a landfill in Lousiana like those E.T. cartridges from Atari.
Exactly. Different spelling but same pronunciation: Zeon
And exactly how high do you have to be to pitch the idea of naming a GPU line after an elemental symbol frequently pronounced as "xenon", when one of your main and most venerable CPU brands is Xeon? Likely the same genius who thought it was a good idea to inject precious metal names into their server CPU models... exactly the market that is *most* specs-focused and *least* likely to get fooled by bling!
On numerous occasions, over the years, I've even heard people misread or misremember Xeon as "xenon". Intel is just *begging* for market confusion!
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
28 Comments
Back to Article
mode_13h - Friday, March 19, 2021 - link
> Intel has already shipped ... Ice Lake Xeon Scalable processors to ... its high-profile customers, even though the processors have not yet been launched. This is typical for a server processor ...Haven't they been saying that since way back in 2019? That's NOT typical of a server processor.
mode_13h - Friday, March 19, 2021 - link
Found it!https://www.anandtech.com/show/14314/intel-xeon-up...
> May 9, 2019
> At present, Intel is already sampling ... Ice Lake processors with customers, and claims that it is on track to ship these CPUs in volumes in the first half of next year.
Oof. And we *still* don't even have a launch date for general availability!
Smell This - Saturday, March 20, 2021 - link
Ouch ... That hurts. Just a month after Cascade.
No wonder Chipzillah is using their heft in slide shows and 'personalities' ...
At least, Intel bought 2 years' time between the **new Second Generation Enterprise Xeon Scalable processors, known as Cascade Lake** and now. Sadly, now, the Cascades will be orphaned (I guess) ...
https://www.anandtech.com/show/14146/intel-xeon-sc...
- April 2, 2019
Santoval - Saturday, March 27, 2021 - link
"Sampling" suggests extremely low volume, not the 115K+ Intel has not shipped. Still, a 2-year delay between "sampling" and launch (since the paper launch is in April I assume the real launch will be in May) is obviously unprecedented. And since Ice Lake-SP is to be fabbed on Ice Lake's node I expect the clocks will be very low..Santoval - Saturday, March 27, 2021 - link
edit : "Intel has *now* shipped".edzieba - Monday, March 22, 2021 - link
It's very typical of server and HPC chips, and has been for a long time.Consumer chips appear to pop up fully-formed in a puff of marketing, but server dies have a much longer lead-in before announcement (let alone public availability for a handful that are sold through retail channels). Cascade Lake made it explicit, but these large server chips are designed in concert with the customers who order tens to hundreds of thousands of dies. There's an ongoing exchange with customers describing the workloads they're expecting to encounter over the next few years, and chip makers providing early sample dies to test those workloads on.
This is why server CPUs 'launch' in production environments running optimised code.
All of this occurs under NDA, so you won't find people detailing the process outside of official press releases (as with Cascade Lake). This isn't just so chipmakers can keep new CPUs secret, it's so customers can keep their future workload predictions secret (if Big Data Processor is known to be wanting to optimise for a certain workload, that provides insights into their desired future product targeting).
Spunjji - Monday, March 22, 2021 - link
I think "Haven't they been saying that since way back in 2019?" is the crucial part of mode_13h's post.Server processors do indeed have a long lead-in, but Ice Lake SP's lead-in has been protracted to say the least.
mode_13h - Monday, March 22, 2021 - link
If it launched a year ago, like they promised in May, 2019, you'd be on the mark. However, this launch is anything but typical (contrary to what this article suggests).mode_13h - Friday, March 19, 2021 - link
> "How Wonderful Gets Done"That just sounds weird and so disconnected from any technical merits of their products that it gives the impression Intel has whipped its marketing and PR departments into overdrive. I smell fear.
GeoffreyA - Saturday, March 20, 2021 - link
Exactly, and in poor taste. It's a bit sad seeing how Marketing has taken over, now that technicals are behind. Same story in the other articles. Marketing gimmicks; smoke and mirrors; trying to save the day.Spunjji - Monday, March 22, 2021 - link
Glad I'm not the only one who found that to have an even more turd-like mouthfeel than their usual advertising boilerplate. 🤢I'm especially in favour of burning all marketing output that uses adjectives as if they were nouns, like "find your happy". 🤮
Pinn - Friday, March 19, 2021 - link
So this in a Mac Pro is a non-starter?Tomatotech - Saturday, March 20, 2021 - link
Very unlikely to see any more Mac models released or updated with Intel chips going forward. Apple will release new Macbook Pro models with their own Apple silicon later this year, then after that iMac and maybe maybe Mac Pro.(In case you didn't know, the current 13" Macbook Pro, Macbook Air, and Mac Mini all now have Apple's custom designed M1 chip which has been very well reviewed.)
Spunjji - Monday, March 22, 2021 - link
That would only ever have been likely if it used the same platform. Given how long they kept the dustbin going, I doubt Apple would release a new Mac Pro platform only to send it to the junk heap in a year or two when they switch that product line over to ARM.JoeDuarte - Saturday, March 20, 2021 - link
It's strange how stalled they've been at 14nm and how much they've had to iterate 10nm. It will be interesting to see how Ice Lake Xeon performs in tests, and I don't have high expectations for it. With all their struggles with 10nm, I got the impression that it was only the third or fourth iteration of it that was expected to be any good, the SuperFin and Enhanced SuperFin, so Sapphire Rapids or whatever.I'm not sure that we'll ever get any exciting leaps from Intel again. Something is wrong, and I don't fully understand it. They can't execute on new process nodes anymore. Does anyone believe that they'll deliver 7nm to mainstream products by 2023? And that it will be excellent in terms of performance and power improvements? They don't seem to be able to do it anymore.
I hope they're able to develop the accelerators and I/O on 10nm+++ or whatever. That would be neat. They haven't done a good job of marketing technology like Quick Assist, the compression and encryption accelerator chip that is available on some of their SKUs. Hardly anyone knows that QA exists. People are wasting huge energy and CPU on compression and encryption when they don't have to, but Intel sucks at marketing. If they got clean-sheet chip designs on 10nm for these accelerators that would be a win. They should also push improved I/O instead of PCIe, technology like OpenCAPI which has much lower latency.
mode_13h - Saturday, March 20, 2021 - link
Tiger Lake seems decent. If they can get their 10+ nm yields up, I think a Xeon based on Willow Cove could keep them in the game.As for improved I/O, they're pushing CXL.
scineram - Saturday, March 20, 2021 - link
There is no such thing?drothgery - Monday, March 22, 2021 - link
I think Sapphire Rapids (late this year/early 2022) is based on Willow Cove?EthiaW - Sunday, March 21, 2021 - link
They do have a new low latency socket family named CXL.mode_13h - Sunday, March 21, 2021 - link
CXL = Compute Express LinkIt's had a fair amount of coverage on this site, if you care to search for it.
Operandi - Monday, March 22, 2021 - link
Intel's CPU archs are impossible to keep track of.... So Ice Lake is Sunny Cove 10nm ++ something, something, but Tiger Lake is Willow Cove 10nm++ something else?So..... Sunny Cove is only for Xeon why? I guess I don't get why there isn't a desktop version of this if 10nm now to the point where they can make chips this big at this performance. Like why bother with Rocket Lake, this has to be better?
drothgery - Monday, March 22, 2021 - link
Sunny Cove was in Ice Lake U (10xxGx mobile) and is now in Ice Lake SP (Xeons and maybe a Core-X variant?); it's on what most press calls Intel's 10nm+ but Intel just calls 10nm and pretends Canon Lake never happened.Willow Cove is in Tiger Lake U (11xxGx mobile), should be in Tiger Lake H very soon (11xxxH mobile), and I think it's in Sapphire Rapids (Xeons and maybe a Core-X variant?). That's 10nm SuperFin (or 10nm SF) in Intel's terminology but would have been 10nm++ previously.
Cypress Cove is in Rocket Lake desktop chips (11xxx), and is a backport of Sunny Cove to 14nm.
Golden Cove will be in Adler Lake (12th gen, supposedly for mobile and desktop). That'll be 10nm Enhanced SuperFin or 10 nm+++ if you like that better. If the code name for the Xeon based on it has leaked, I haven't heard it.
Meteor Lake (haven't heard the CPU core code name) in Q1 2023 is supposed to be on 7nm, I think.
Operandi - Monday, March 22, 2021 - link
I read that and the next thing I remember is waking up on the floor with my face in small pool of drool. Not sure how long I was out for.....JayNor - Monday, April 5, 2021 - link
Sapphire Rapids has Golden Cove cores.EthiaW - Sunday, March 21, 2021 - link
Perhaps intel is simply too ashamed to launch these products because they cannot beat AMD Millan even in cherrypicked workloads.mode_13h - Sunday, March 21, 2021 - link
AVX-512 is their usual go-to, when they want to gin up some good-looking benchmarks. I think it's also responsible for a fair bit of Rocket Lake's claimed IPC improvements.PaulHoule - Monday, March 22, 2021 - link
Weren't the Xeons the bad guys in Gundam? Didn't they crash one of O'Neill's Island 3 colonies into NYC or something like that?I have been offering my services as a branding consultant to firms like
https://x-energy.com/
who want to name a high-temperature gas cooled reactor "Xe" and I can't tell if they are talking about the Xenon-135 radioactive pollutant or the notorious low performance graphics card architecture. Going high end is like Trabant competing in F1.
Best thing that can happen for those two letters is that Dell buys a bunch of OEM-only graphics cards and discreetly trucks them to a landfill in Lousiana like those E.T. cartridges from Atari.
mode_13h - Monday, March 22, 2021 - link
Exactly. Different spelling but same pronunciation: ZeonAnd exactly how high do you have to be to pitch the idea of naming a GPU line after an elemental symbol frequently pronounced as "xenon", when one of your main and most venerable CPU brands is Xeon? Likely the same genius who thought it was a good idea to inject precious metal names into their server CPU models... exactly the market that is *most* specs-focused and *least* likely to get fooled by bling!
On numerous occasions, over the years, I've even heard people misread or misremember Xeon as "xenon". Intel is just *begging* for market confusion!