Monday, May 3rd 2021

Intel "Sapphire Rapids" Xeon Processor Could Feature Up To 80 Cores: New Leak

Intel's upcoming Xeon "Sapphire Rapids" enterprise processor come come with CPU core-counts as high as 80, according to the latest round of photo-leaks. An earlier article predicted the chip cram up to 56 cores alongside on-package HBM. The processor reportedly features up to 80 cores, spread across four 20-core chiplets. Unlike on the latest AMD EPYC processor, there doesn't appear to be a centralized I/O controller die. This particular processor is based in the LGA4189 package, which features additional pins compared to the LGA4577-X socket from the 56-core leak. The newer socket has additional pins that enable next-gen I/O, which include PCI-Express Gen 5.0, and CXL 1.1 interface.
Source: HotHardware
Add your own comment

36 Comments on Intel "Sapphire Rapids" Xeon Processor Could Feature Up To 80 Cores: New Leak

#26
Jism
phill
I wonder what it'll be like to cool......
Usually 280W to 400W of heat dissapation. Clocks of enterprise / server CPU's generally are lower compared to consumer(s) CPU's. Second; in a datacentre, where these things belong, there is no such thing as silence as datacenter rooms are temperature controlled or even mass hooked to a watercooling system. 400W is'nt that much. With servers your more looking at "how much work can i do for the money" where power consumption alread is taken into account. You pay for current an hour btw.
Posted on Reply
#27
Crackong
Does it support multiple sockets or already saturated all the inter-connects in the gluing process ?
Posted on Reply
#28
phill
I wonder if it's going to fail before it's even been released... Well, will look forward to see what it can do....
Posted on Reply
#29
MentalAcetylide
I wonder how long(if its even possible) before "GPU-type" cards just replace CPUs altogether.
Posted on Reply
#30
davideneco
Its 56 core ....

If i follow your Logic , ICL-SP is 56 core instead 40
And Skylake SP is 36 core not 28 ....
Posted on Reply
#31
dragontamer5788
MentalAcetylide
I wonder how long(if its even possible) before "GPU-type" cards just replace CPUs altogether.
Replace?

GPUs are a completely different model of computation. CPUs are low latency, GPUs are high latency. It takes extraordinary amounts of effort to rewrite code to efficiently work on a GPU, and its unknown if some algorithms could ever be efficient on a GPU.

EDIT: If you mean a 60-core CPU-card that plugs into the PCIe port, well, that was already done. Look up Xeon Phi 5110P, no one seemed to like it.
Posted on Reply
#32
MentalAcetylide
dragontamer5788
Replace?

GPUs are a completely different model of computation. CPUs are low latency, GPUs are high latency. It takes extraordinary amounts of effort to rewrite code to efficiently work on a GPU, and its unknown if some algorithms could ever be efficient on a GPU.

EDIT: If you mean a 60-core CPU-card that plugs into the PCIe port, well, that was already done. Look up Xeon Phi 5110P, no one seemed to like it.
So basically its not possible & would be comparable to wanting a BelAZ 75600 used in stripping pits to perform like a race car?
Posted on Reply
#33
dragontamer5788
MentalAcetylide
So basically its not possible & would be comparable to wanting a BelAZ 75600 used in stripping pits to perform like a race car?
I mean...

It can happen. Just really unlikely. And if it does happen, the end result won't really look like a truck fundamentally anymore, and more like a jet-plane.

----------

Ex: PEZY Accelerators are CPU-like and GPU-like (en.wikichip.org/wiki/pezy/pezy-scx/pezy-sc). But as a hybrid design, they don't look anything like a traditional CPU or GPU: clearly some kind of weird mix between the two. The A64Fx is another hybrid: mostly CPU (ARM architecture), but it uses GPU-RAM instead of traditional CPU-RAM.

I expect future CPUs to grow more GPU-like, and for future GPUs to grow more CPU-like. But at the same time, there's clearly "CPU-only" workloads and "GPU-only" workloads. Some tasks may never make sense on the other architecture.
Posted on Reply
#34
MentalAcetylide
dragontamer5788
I mean...

It can happen. Just really unlikely. And if it does happen, the end result won't really look like a truck fundamentally anymore, and more like a jet-plane.

----------

Ex: PEZY Accelerators are CPU-like and GPU-like (en.wikichip.org/wiki/pezy/pezy-scx/pezy-sc). But as a hybrid design, they don't look anything like a traditional CPU or GPU: clearly some kind of weird mix between the two. The A64Fx is another hybrid: mostly CPU (ARM architecture), but it uses GPU-RAM instead of traditional CPU-RAM.

I expect future CPUs to grow more GPU-like, and for future GPUs to grow more CPU-like. But at the same time, there's clearly "CPU-only" workloads and "GPU-only" workloads. Some tasks may never make sense on the other architecture.
Oh no doubt its within the realm of possibility, but just not economically feasible. Years ago for shits & giggles, me and a friend did some rough calculations to see what it would take to get one of those fully loaded stripping pit dump trucks in our area to do 80mph and came up with a figure close to 100,000 HP. Never mind what it would take to stop the behemoth after reaching that speed. It would be more of an expensive novelty venture.
Posted on Reply
#35
KarymidoN
how they're gonna feed such beast? nuclear reactor integrated in the Mobo? only sell for datacenters in Iceland or Siberia?
Posted on Reply
#36
candle_86
the jokes on all of them, data centers seldom go for these crazy core counts, its cheaper to buy more mainstream core density's and build out the rack more, because you have less heat concentration, something with 400w load won't fit in a 1u rack which is the gold standard, and it will dump to much heat at one spot. Most data centers I've been in recently use servers with between 12-16 cores per server with load balancing it all fits neatly in 1u enclosures, it's easier to cool, and its significantly cheaper.
Posted on Reply
Add your own comment