Intel’s Xeon 600 Brings Granite Rapids Muscle To Excessive-Finish Desktops

Some of us act as if Intel and AMD have killed off the “high-end desktop” or HEDT market, however that is by no means actually been the case. As an alternative, what occurred is that mainstream desktops went from a most of 4 cores to 16  – 24 cores mainly in a single day, propelling HEDT programs out of attain for many fans. In order for you a processor with extra cores than you may shake a stick at, although—solely P-cores, too—in addition to a humongous pile of PCI Categorical connectivity and reminiscence bandwidth, Intel and AMD will gladly promote you one.

What Make Intel’s Xeon 600 Processors Distinctive?

Certainly, that is precisely what Intel is asserting at present: the Xeon 600 Processors for Consumer Workstations, which exchange each the extant Xeon W-2500 and Xeon W-3500 sequence CPUs. As Intel says, it is doing a little bit of product stack compression with this launch, however we’ll get into that in a second. The purpose is, these are new Xeon processors meant for workstations, based mostly on the Granite Rapids structure.

Intel Xeon 600 sequence chips are structured in fairly a special method from the earlier era; if you wish to learn extra about that, try our earlier protection of Granite Rapids. Intel says that these components are fabricated on Intel 3, identical to the Granite Rapids server processors, and that they’re based mostly on the Redwood Cove core structure which debuted in Meteor Lake within the shopper section again in December of 2023. Be aware {that a} portion of the 61% multi-threading efficiency soar right here comes from evaluating a 60-core processor to an 86-core CPU, however that is nonetheless a substantial per-core benefit for the brand new CPUs.

Huge adjustments this time round embody a humongous soar in L3 cache, however we’ll get to that in a second. Certainly, most of the different adjustments are instantly targeted on the AI market, in fact. As earlier than, all 86 cores assist AVX-512 and Intel’s AMX extensions, though these chips are the primary from Intel to assist the FP16 knowledge sort. Intel says these chips are “the AI developer’s dream,” as they’re based mostly on the identical structure as Xeon server processors, and thus present a constant improvement platform.

Intel Xeon 600: Server Processor With A Consumer Chipset

As Intel says, HEDT is admittedly “server CPUs with a shopper chipset,” and that is what this slide is illustrating. The platform combines the Xeon 600 CPUs, sporting as much as 86 CPU cores, 128 PCI Categorical 5.0 lanes, new CXL 2.0 for CXL Reminiscence assist, with the W890 chipset that newly contains Wi-Fi 7—the form of factor that is necessary for a shopper workstation. Other than that, the chipset platform does not change for this era.

Identical to the Granite Rapids server chips, the workstation model helps Multiplexed Rank DIMMs for improved reminiscence bandwidth. We have reported on this earlier than and have an honest explainer, however the brief model is that it is mainly like doing dual-channel accesses to every particular person reminiscence module. This improves bandwidth—as much as 8000 MT/s switch price on these components—however it does have a draw back, which is decreased max capability. Intel did not say what the max capability is with MRDIMMs, however it will be decrease as a result of controller limits.

The Intel Xeon 600 Sequence Processor Line-Up

Here is the slide you’ve got all been ready for, which is the desk of processor SKUs. As you may see, the higher-end configurations do certainly get an enormous soar in L3 cache from previous-generation components. Curiously, Intel’s presentation credited the transfer to Redwood Cove for the large bump in L3 cache though that is actually extra of a platform design choice than something to do with the processor core itself.

Solely a number of the SKUs shall be getting boxed variations, and Intel restricts each overclocking options and MRDIMM assist to the higher-end fashions within the household. The latter makes excellent sense; MRDIMMs’ greater bandwidth is not actually mandatory on the chips with fewer CPU cores. It does not provide decreased latency, solely greater bandwidth, and that is useful while you’re doing a number of crunchy AVX math, however the 24-core and under fashions merely do not have the core rely to want the upper bandwidth.

On this slide Intel explains its choice to lock off overclocking solely to the higher-end fashions. The corporate says that it is actually meant as an possibility for system integrators and VARs who wish to provide tuned configurations to their clients, and are prepared to tackle the elevated assist prices. That is completely legitimate; these tuned configurations are sometimes much less about what we usually consider as “overclocking” and extra about energy and platform tuning. Even nonetheless, this market is small in comparison with the market of grownup fans who can be prepared to pay maybe $999 for a giant fats CPU that they will then tune for greater clock charges to enhance efficiency in shopper workloads. Sadly, the best way Intel has segmented it, the most affordable chip for this use case is the $1699 Xeon 658X.

Whereas that mannequin retains the total 128 PCI Categorical 5.0 lanes and 8-channel reminiscence assist, it “solely” has 24 CPU cores, which we should level out is identical quantity because the Core Extremely 9 285K, a chip which runs almost a full 1 GHz sooner than this Xeon half. Certain, these are a mixture of P-cores and E-cores the place this chip is all P-cores, however as an fanatic myself, it is onerous to justify, except the elevated bandwidth and PCIe connectivity that include an HEDT system are an absolute necessity.

How Do Intel’s New Xeon 600 Sequence Processor Carry out?

We can’t spend a number of time on this, however Intel shared some benchmarks exhibiting off the efficiency of the brand new chips. The primary slide is exhibiting SPEC Workstation 4.0 pace, providing a variety of efficiency outcomes from +/-3% on the low finish all the best way as much as 61% sooner in Finance Companies when evaluating in opposition to the earlier era Xeon w9-3595X. On the second slide, Intel claims efficiency beneficial properties 27% and better in three skilled benchmarks.

In these two slides, Intel emphasizes its rendering efficiency, the place it claims to be 74% sooner than the earlier era in Blender CPU, and 29% sooner in Topaz Labs AI video upscaling. The corporate celebrates the discharge of Intel Open Picture Denoise 2.4, which is a bit like NVIDIA’s DLSS Ray Reconstruction for offline rendering. It is accelerated on the AMX models of the brand new CPUs using the added FP16 performance. Fairly cool, though Intel did not give any efficiency numbers or different comparisons.

Lastly, Intel says the brand new chips are sooner at AI versus the earlier era. Xeon 698X is outwardly 24% sooner at linear algebra, 18% sooner at large-set knowledge evaluation, and 16% sooner at CPU-based AI inference versus the earlier era. We would like to see benchmarks versus AMD’s newest EPYC CPUs, that are additionally excellent at AI.

Now, this half is a really fascinating transfer from Intel. These new chips completely do have their place, however as Intel itself factors out, it is not as clear-cut as “desktop versus workstation” anymore, as a result of trendy desktop CPUs are so quick. Intel recommends a “desktop workstation” with a Core Extremely 200 processor and a W880 chipset for “latency-sensitive interactive & moderately-threaded workloads,” whereas suggesting that the Intel Xeon 600 CPUs provide higher efficiency in “highly-threaded, high-memory bandwidth, massive memory-capacity, & excessive PCIe expandability workloads,” with examples of every.

Intel presents these figures, evaluating a Core Extremely 5 245K in opposition to a Xeon 636. The comparability is not notably flattering contemplating that the Core Extremely half is all of one-third the price of the Xeon machine, earlier than factoring in the price of motherboards for the monster socket and reminiscence to fill the 4 channels, however that is precisely the purpose. The Xeon 636, regardless of shedding in some workloads, can merely do issues that the Core Extremely cannot. It is a candid and clear declaration from Intel’s advertising and marketing and we approve.

That just about wraps it up. The brand new Xeon 600 processors include as much as 86 modified Redwood Cove cores, fats L3 caches, MRDIMM assist, as much as 128 PCIe 5.0 lanes, CXL 2.0 assist, all of the vPRO administration options IT departments know and love, and radically simplified naming versus the earlier era. Hopefully, that intuition for simplified naming carries over to the corporate’s shopper CPU division quickly too. Intel says the Xeon 600 Processors for Consumer Workstations shall be accessible beginning in late March of this yr.