Intel's Make-Or-Break 18A Process Node Debuts For Data Center With 288-Core Xeon 6+ CPU (tomshardware.com) 40
Intel has formally unveiled its Xeon 6+ "Clearwater Forest" data-center processor with up to 288 cores, built on the company's new Intel 18A process and using Foveros Direct packaging. The chip targets telecom, cloud, and edge-AI workloads with massive parallelism, large caches, and high-bandwidth DDR5-8000 memory. Tom's Hardware reports: Intel's Xeon 6+ processors with up to 288 cores combine 12 compute chiplets containing 24 energy-efficient Darkmont cores per tile that are produced using 18A manufacturing technology, two I/O tiles made on Intel 7 production node, as well as three active base tiles made on Intel 3 fabrication process. The compute tiles are stacked on top of the base dies using Intel's Foveros Direct 3D technology, whereas lateral connections are enabled by Intel's EMIB bridges.
Intel's 'Darkmont' efficiency cores have received rather meaningful microarchitectural upgrades. Each core integrates a 64 KB L1 instruction cache, a broader fetch and decode pipeline, and a deeper out-of-order engine capable of tracking more in-flight operations. The number of execution ports has also been increased in a bid to improve both scalar and vector throughput under heavily threaded server workloads.
From a cache hierarchy standpoint, the design groups cores into four-core blocks that share approximately 4 MB of L2 cache per block. As a result, the aggregate last-level cache across the full package surpasses 1 GB, roughly 1,152 MB in total. This unusually large pool is intended to keep data close to hundreds of active cores and reduce dependence on external memory bandwidth, which in turn is meant to both increase performance and lower power consumption. Platform-wise, the processor remains drop-in compatible with the current Xeon server socket, so the CPU has 12 memory channels that support DDR5-8000, 96 PCIe 5.0 lanes with 64 lanes supporting CXL 2.0.
Intel's 'Darkmont' efficiency cores have received rather meaningful microarchitectural upgrades. Each core integrates a 64 KB L1 instruction cache, a broader fetch and decode pipeline, and a deeper out-of-order engine capable of tracking more in-flight operations. The number of execution ports has also been increased in a bid to improve both scalar and vector throughput under heavily threaded server workloads.
From a cache hierarchy standpoint, the design groups cores into four-core blocks that share approximately 4 MB of L2 cache per block. As a result, the aggregate last-level cache across the full package surpasses 1 GB, roughly 1,152 MB in total. This unusually large pool is intended to keep data close to hundreds of active cores and reduce dependence on external memory bandwidth, which in turn is meant to both increase performance and lower power consumption. Platform-wise, the processor remains drop-in compatible with the current Xeon server socket, so the CPU has 12 memory channels that support DDR5-8000, 96 PCIe 5.0 lanes with 64 lanes supporting CXL 2.0.
Wattage? (Score:2)
Any information on the wattage of this beast?
Re: Wattage? (Score:1)
Re: (Score:2)
Re: (Score:2)
Re:Wattage? (Score:4, Informative)
> Any information on the wattage of this beast?
If you have to ask, you can't afford the electricity... OK, it's not too bad for what it is.
---
Maximum Power (TDP):
The 288-core "Darkmont" E-core flagship is rated at a 450W TDP. The broader platform supports a range up to 500W for the highest-performance configurations. In comparison, it offers an 11% lower TDP than the previous 288-core "Sierra Forest" (500W) while delivering 17% higher performance.
Idle Power:
Official specific idle wattages for Clearwater Forest have not been released yet, but early data for the previous Sierra Forest 144-core architecture showed system idle levels around 315W. Clearwater Forest is expected to significantly improve on this through its Intel 18A process and RibbonFET technology, which target "exceptional energy efficiency" and low leakage. Intel claims the 288-core Xeon 6+ can reduce overall runtime rack power by 38% compared to a dual-socket 288-core Sierra Forest platform.
Key Efficiency Specs:
Process Node: Intel 18A (1.8nm-class).
Architecture: 288 "Darkmont" Efficiency-cores.
Efficiency Gain: Up to 60% higher performance-per-watt compared to predecessor platforms.
Re: (Score:2)
Re: (Score:2)
Re: (Score:1)
Do you ever have anything useful to say beside being a bitter malcontent?
Re: (Score:2)
Re: Lets hope it breaks them (Score:2)
Re: (Score:2)
If there were to be a "transition to AI", probably. But there is not going to be one. The public (and CEO) perception of LLM-type AI is largely a hallucination and very few of the promises made will pan out, because the tech used is not capable of delivering on them, regardless of tweaks and optimizations. It is the wrong tool for that and you cannot make a silk purse from a pig's ear. If your buyers are clueless, you may get away with faking it for a while though.
Re: (Score:2)
While I agree that there should be more competition, Intel is are not the ones that bring that. They are one of the main reasons why there is not enough competition though.
Re: (Score:2)
Intel no longer calls the shots in this sector, and hasn't for a while. If they went under, then that's one major foundry less, and TSMC would have a virtual monopoly. We need Intel's foundry business, at the very least. As far as x86 goes, all the patents must have expired by now, since this platform is now trying to hold its own against Arm, and Microsoft isn't making that easy by continuing to crap on Windows 11. It shouldn't be difficult for semiconductor startups to design new x86 CPUs, although fi
Re: (Score:2)
Yes. But I do not think they can exist in this state. They either have to be top-dog again (very bad) or go away (bad, but not as bad).
Re: (Score:2)
I think they can. The leaders who were obsessed w/ them being top dog, like Andy Grove, are dead, so they now have pretty mediocre leaders, just happy to survive. During the Intel of Andy Grove or Craig Barrett, it was inconceivable that Intel would need investments from Nvidia or Uncle Sam. In fact, Intel shone wrt TSMC: while the latter was getting subsidies from the Taiwanese government, Intel was rivaling them in process technology w/o any help from anyone. It was an entity of pride, even for people
Re: (Score:2)
Precisely! If nothing else, I want Intel to continue to survive as a US-based foundry company, even if TSMC is building fabs in Arizona and Ohio. We need more companies in the foundry business: maybe also have Global Semiconductors and a few others there as well. And we need more players in the sub 10nm process range
As far as x86 goes, there too I don't want either Intel or AMD to have a monopoly. Incidentally, aren't all the patents related to x86 expired by now, given that they've been around for mo
Re: (Score:1)
I seem some stupid people got mod-points again. I recommend looking up "Stockholm Syndrome".
Not for you (Score:5, Interesting)
As big as these devices are, they're essentially embedded CPUs. They're intended for baseband signal processing in cellular networks. They live just behind the O-RAN layer (cellular transceivers that do the RF your devices see,) and process the baseband signal with FFT hardware offload (Intel VRAN Boost.) Hybrid SDR, essentially. After baseband processing and error correction, the signals are authenticated, metered, etc. A large number of low power X86 cores then run all the proprietary operator code for the network.
The customers for these are well-heeled wireless network operators, and they don't care about prevailing prices for 1-2TB of the DDR5-8000 they need to feed each core 2-4GB of high performance RAM: the cost is a fraction of what they pay for the RF transceiver hardware and everything else it takes to operate a wireless network. So they're paying full retail fresh out of the foundry, and Intel's massive investments in new nodes and incredibly sophisticated integration (3 different nodes stacked in a 3D package...) pay off handsomely.
Re: (Score:2)
This makes the cache layout make a lot more sense. It would be limiting for a lot of workloads, but not for that.
Re: (Score:3)
They're intended for cloud compute. The fact that they're targeting a different market ought to tell you something...
Re: (Score:2)
cloud compute
Abject linguistic nebulosity.
Modulating and processing enormous amounts of cellular traffic can be shoehorned under the caption "cloud compute" any time such machinations are found necessary.
Re: (Score:2)
... no.
Intel has a separate list of SKUs for signal processing, and they're usually low core count Atom-derivatives. Examples from the recent past include Snow Ridge:
https://www.intel.com/content/... [intel.com]
There will probably be Darkmont-based products for use in base stations, but they will not be 288c 500W+ monsters in a cloud rack. Those CPUs are intended to be successors to Sierra Forest:
https://www.intel.com/content/... [intel.com]
Sierra Forest was meant to compete with AMD's Bergamo CPUs (albeit a year late). Sierra
Re: (Score:2)
We're talking here about Xeons. They are anything but embedded: they are for the other end of the computing spectrum - workstations, servers and supercomputers. Intel had actually sold its wireless chipset division to Apple a while back now, so they're no longer in that business
Re: (Score:2)
Intel had actually sold
Intel sold its XMM modem chipset business to Apple. It did not exit the wireless market. As you can see here [intc.com], (from 2023, four years after the Apple sale,) Intel has been and continues developing new products for wireless operators, and is now, with this new Xeon, on its third generation of such devices. Overall, Intel exited the low margin, highly competitive "client" end of the wireless world and focused on the high margin wireless network operator end, where their VLSI capabilities face less competiti
But... (Score:2)
Will it run Crysis?
Re: (Score:2)
That meme is as obsolete as Beowulf clusters. Crysis is not praticularily demanding compared to games released 10 years after it.
Crazy! (Score:2)
That's some crazy density. And at 500 watts TDP, I suspect this will need to be water cooled.
I could not find what the speed was. Hopefully better than that 1.8Ghz lameness of most of today's Xeons.
If you're unfortunate enough to have to license Windows on these 288 cores, Windows Server 2025 Datacenter Edition will run you over $100,000
Re: (Score:2)
Not your standard server CPU... (Score:2)
Interesting, this does not seem to be the standard server CPU with Intel calling them "efficiency" cores. I wonder if they feel they have fallen behind the performance game to AMD and trying shift to "efficiency". Their Sapphire Rapids was disappointing, the next one Emerald Rapids was also not great (ok performance if you did not use all cores, which did not make a good solution for cloud providers), but Granite Rapids seemed to be a meaningful improvement - just AMD's Turin came out and was so much faster
Re: (Score:2)
The low end of this workstation line has 6 cores. It would actually be ideal for a workstation running a hypervisor - be it Proxmox, HyperV or VMware - and then, on top of that, VMs of one's choice. Given all the issues around Windows 11 and the fact that macOS on Intel support is fading, if not gone, I'd have a bunch of VMs there - FreeBSD, OpenBSD and a bunch of OSs not internet connected - Windows 7, macOS Mojave, OS/2 Warp or ArcaOS and Kali Linux. Then I would be able to run whatever I need in an ap
Unpaywalled write-up of some of the new tech (Score:3)
It's made of glass! Well, bits of it anyhow. And 1.8nm process. Pretty neat, right?
Re: (Score:2)
That nm process is marketing technically. A18A achieves the same transistor density as TSMC's N3 (3nm node size) but it does slightly edge out Samsung's SF2 process.
But the important part is that there's progress.
Re: (Score:1)
"Later this year" (Score:2)
There are leaks suggesting it's more of a Q1 2027 product. It remains to be seen whether it can compete with Turin-dense competently. Venice will annihilate it. Intel may replace/displace it with Diamond Rapids-AP. There's a lot of questions about the future of this product. LBT was asked directly about the health of Clearwater Forest during the Q4 earnings call and he ignored the question.