Ice Lake, Baby: Intel's 10nm 3rd Gen Xeon Scalable server processors to arrive at last Intel on Tuesday announced the availability of its "Ice Lake" 3rd Gen Xeon Scalable processors, intended for applications running on servers, high-end workstations, and in data centers. And, yes, this is Intel's first Xeon Scalable processor family on its much-delayed 10nm process node. "Our 3rd Gen Intel Xeon Scalable …
"Our 3rd Gen Intel Xeon Scalable platform is the most flexible and performant in our history"
Surwe, as if you would proudly announce a chip that would be less flexible and performant than what was made before. Hurray!
Also, I note it supports BOATKUWC, i.e. Boatload Of Acronyms To Keep Up With the Competition..
No, I haven't had my coffee yet, why?
I recall the adverts for Ariel/Persil (washing powder for clothes) in the UK in the 80s where they showed how much cleaner "new" Ariel was vs "old" Ariel. I think it was Jasper Carrot who opined that the tag for the version from 10 years previous must have been "cleans some of your clothes some of the time".
"Surwe, as if you would proudly announce a chip that would be less flexible and performant than what was made before. Hurray!"
You might, though. They can sell chips on other benefits. It could be similarly performant as the old one but a lot more energy-efficient. Or as performant per core but capable of more parallelism. Or as performant on standard tasks but with extra acceleration for particular operations. Those are at least possible alternatives, albeit ones that don't often make the headline announcement.
I seem to recall this same kind of thing going on every Xeon CPU launch, where they say the new processor is 2.65x or something faster than some xxx generation but then when you look at the processors that are lined up for comparison, usually the new latest one has a bunch more cores than the old one.
Considering that core count matters greatly in software licencing these days, it might be nice if they could clearly show what it looks like if you compare the different generations of CPUs using exact same core count and exact same clock speed...
I believe the 11900k is quite a bit faster than the 6900k.
But how does it compare to the 10900k?
The most positive review I’ve seen of it described it as “a waste of sand”. I wonder if this Xeon is any better?
Is this a higher frequency 10nm process than the one Intel didn't use for the i9-11900K and its brethren because 10nm was too slow? Or perhaps business users don't need the high frequencies that gamers demand, and so a slower chip that uses less electricity works for them? Oh, I see: they've upped the core count, and database can do with throughput, it's gamers that need low latency.
Highest base clock is 3.6GHz (see table in "Next Platform" article). But recent Xeons have not been screamers in terms of frequency, because it is more power-efficient to run concurrent workloads on an increased number of cores rather than at increased frequency - as long as there is enough work to go round.
GPUs are also an example of that effect - taking it to an extreme, perhaps - and I suspect that many gamers only buy high-end CPUs for the bragging rights. IMO for gaming it is better to select a GPU to match your budget, then get the cheapest CPU that won't bottleneck the GPU. And even then, go for sustained clock frequency rather than CPU core count.
Anandtech has a review (https://www.anandtech.com/show/16594/intel-3rd-gen-xeon-scalable-review) that compares the speed of the new Intel chips and the AMD EPYC (Milan). The EPYC wins in almost all tests apart from AVX-512 and idle power consumption. The idle power consumption of the EPYC is higher due to the chipset being in the package, the extra 64 PCIe 4.0 lanes and 256 MB of L3 cache vs 60 MB .
Much of the performance gain from higher IPC in the new Xeon is lost due to the lower frequencies (3.4GHz vs 4GHz).
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
