News: 1664476207

  ARM Give a man a fire and he's warm for a day, but set fire to him and he's warm for the rest of his life (Terry Pratchett, Jingo)

Intel accidentally leaked its 34-core Raptor Lake chip. What do the dies tell us?

(2022/09/29)


Analysis At this week's launch of Intel's 13th-gen Core series, it appears staff accidentally left out on display a wafer of previously undisclosed 34-core Raptor Lake processor dies.

As spotted by [1]Tom’s Hardware , the wafer bore a sticker reading “Raptor Lake-S, 34 core.” A close inspection of each die revealed a mesh of 34 distinct CPU cores, suggesting these are all performance cores; there's no sign of four-core clusters indicative of Intel’s efficiency cores. The die also appears to depict eight memory controllers — far more than you’d expect to see on a regular desktop chip — as well as UPI blocks, which are only necessary for multi-CPU systems.

The discovery seems to indicate Intel has a Raptor-Lake-family workstation part in development to challenge AMD’s Threadripper processors. Threadripper tops out at 64 cores spread across eight dies. This “Raptor Lake S” die appears to pack just over half that into a single monolithic die.

[2]

Of course, there’s no guarantee this thing even works, if all 34 cores will be activated or usable, or if it will ever see commercial release.

[3]

[4]

Chipmakers show off wafers at events, like its Innovation 2022 conference in Silicon Valley this week, to impress journalists, analysts, and customers. The dies are usually already publicly announced components.

The cynical among us will think this wafer was strategically and quietly left out on display at the event for someone to find, but from what we can tell, this was a mistake by Intel. Then again, it is Chipzilla.

A fire breathing raptor?

While the dies may depict 34 cores each, don’t expect performance to scale linearly compared to Intel’s desktop chips. While it might have more than 4x the performance cores as the just-launched 13th-gen Core processors, it is still going to be confined by power and thermal limits. Remember, with eight performance cores and eight efficiency cores, Intel’s [5]Core i7 13700K is rated to consume 253W of power.

Most workstation and server processors max out at 280W per socket, which doesn’t leave much headroom for all those extra performance cores. That is unless Intel decides to push that limit. It wouldn’t be surprising to see Intel go down that route. As Intel has demonstrated over the past two generations of Core-series parts and its growing lineup of GPUs and accelerators, it’s not afraid of a little heat. Intel’s Guadi-2 AI accelerators and upcoming [6]Ponte Vecchio GPUs both have 600W TDPs.

[7]AMD's Ryzen V3000 goes head to head with Intel's embedded chips on power, oomph

[8]Chipmakers still shoveling cash into new fabs as demand slows

[9]AMD was right about chiplets, Intel's Gelsinger all but says

[10]Nope, still no Intel Sapphire Rapids Xeon processors for you right now

Given enough 12-volt EPS connectors, some beefy voltage regulation modules, and a socket capable of delivering all that power to the chip, it’s not inconceivable to think Intel might push the power-consumption envelope well beyond the 280W we’ve come to expect from datacenter and workstation chips. However, doing so in a workstation class chip would almost certainly require liquid cooling at a minimum.

More likely, Intel will continue the longstanding tradition of trading frequency for core count. It essentially boils down this: for a fixed power budget the more cores you have, the fewer volts you have to pass around. As a result, the only way to boost clocks while also increasing core count is to ether to make the architecture more efficient or increase the power budget.

Probably not an HEDT part

While it’s been almost three years since Intel last launched a high-end desktop (HEDT) processor — its last being the 18-core, 36-thread Core i9 10980XE — this 34-core Raptor Lake die probably isn’t its successor.

In many regards, Intel’s 12th and now 13th-gen Core-series processors have filled the hole left by the 10980XE, with its 13th-gen Core i9 13900K now offering 24 cores — eight performance and 16 efficiency.

[11]

The presence of eight memory controllers and UPI blocks – Ultra Path Interconnect – would seem to back that conclusion up, and suggests instead that this die may be destined for Intel’s Xeon W platform.

As workstation chips go, Intel has historically offered a diverse lineup of chips, ranging from its Core-series parts on the low-end to HEDT, Xeon W, and more recently Xeon Scalable parts for thread-sensitive workloads.

A 34-core chip is by no means revolutionary for Intel, either. The chipmaker’s Xeon W-3375 processor is based on its Ice Lake Xeon Scalable platform and boasts up to 38 cores. Meanwhile, Intel’s Xeon Scalable chips top out at 40. But since we’ve only seen one die, there’s no reason to think Intel won’t have higher core-count SKUs.

[12]

Regardless of whether this chip makes it to commercial release, Intel’s apparent oversight offers a glimpse at how its thinking about workstation parts even as it follows AMD and Apple down the chiplet rabbit hole with [13]Sapphire Rapids . ®

Get our [14]Tech Resources



[1] https://www.tomshardware.com/news/intels-unannounced-34-core-raptor-lake-cpus-displayed-on-wafer

[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2YzYVe8MZ2lv7-5NLPvRKhQAAANA&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0

[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YzYVe8MZ2lv7-5NLPvRKhQAAANA&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33YzYVe8MZ2lv7-5NLPvRKhQAAANA&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[5] https://www.theregister.com/2022/09/27/intel_13thgen_cpu_innovation/

[6] https://www.theregister.com/2022/08/23/intel_ponte_vecchio/

[7] https://www.theregister.com/2022/09/28/amd_ryzen_embedded_chips/

[8] https://www.theregister.com/2022/09/28/chip_fabrication_investment_semi/

[9] https://www.theregister.com/2022/09/28/intel_chiplets_advanced_packaging/

[10] https://www.theregister.com/2022/09/27/intel_sapphire_rapids_xeon_processors/

[11] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YzYVe8MZ2lv7-5NLPvRKhQAAANA&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[12] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33YzYVe8MZ2lv7-5NLPvRKhQAAANA&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[13] https://www.theregister.com/2022/09/27/intel_sapphire_rapids_xeon_processors/

[14] https://whitepapers.theregister.com/



Yeah (scratches chin like used to in school)

steviebuk

"accidentally"

coughPRstuntcough

Re: Yeah (scratches chin like used to in school)

Anonymous Coward

Yeah, everyone knows what this is, and the author, like the rest of the trade press will happily play along. Months of breathless followup stories and click bait.

Zero chance a wafer for an unreleased chip to "accidentally" get out of a fab in another part of the county/world and land in a trade show case. Zero. That would require a chain of career ending F-ups that would cross multiple departments.

So in the context of a staged leak, this is a play to buy time. Specifically time to market for their new designs, but also a play to protect their share prices in the face of rising interest rates and high inflation, neither of which plays well for a company who's 5 year plan is to set mountains of cash on fire to close the engineering gap with it's rivals.

This processor may be a fun and handy distraction, but it probably won't see production in any form. And it's not like they are saying it will. So no shareholder suit when it fails to materialize. This is a shot across the bow of the other players in the CPU game, not so much a warning shot as a reminder they have a few cannon balls. That and helping them keep the system integrators from developing a wandering eye.

For us the fun will be watching how long it takes to bring an equivalent product to market. Clearly they are heavily implying that something big is right around the corner, but what they showed off is going against the current design and architecture roadmap they have just laid out. This isn't a chiplet based design. So if this ships at all it would be an evolutionary dead end. A last gloriously over the top capstone of an obsolete architecture.

In that light I feel like people are going wow over a wafer of 64 core itaniums, or Netburst cores.

Re: Yeah (scratches chin like used to in school)

cornetman

> This isn't a chiplet based design. So if this ships at all it would be an evolutionary dead end. A last gloriously over the top capstone of an obsolete architecture.

Not only that, their yields would be awful. It would be an *expensive* CPU.

Re: Yeah (scratches chin like used to in school)

Michael Duke

So just like the rest of Intel's "Workstation" class chips then.

Mankind is poised midway between the gods and the beasts.
-- Plotinus