As much as 56 Cores and 112 PCIe 5.0 Lanes

For the entire singular focus that Intel has positioned on its client Core desktop CPU elements in the previous few years, you would be forgiven for pondering that Intel has forgotten about their Xeon premium processor lineups for workstations. Between the de facto retirement of Intel’s desktop-grade Xeon W-1×00-series lineup, and the repeated delays of Intel’s current-generation huge silicon elements for servers, the Sapphire Rapids-based 4^th Era Xeon Scalable collection, there hasn’t been a lot noise from Intel within the workstation area in the previous few years. However now that Sapphire Rapids for servers has lastly launched, the logjam in Intel’s product roadmap has finally cleared out, and Intel is lastly ready to renew cascading their newest silicon into new workstation elements.

This morning Intel is asserting their first top-to-bottom refresh of workstation elements, the Xeon W-3400 and Xeon W-2400 collection. Geared toward what Intel is broadly classifying because the Skilled Workstation and Mainstream Workstation markets respectively, these chip lineups are meant to be used in high-performance desktop workstation setups, significantly those who require extra CPU cores, extra PCIe lanes, extra reminiscence bandwidth, or a mix of all three components. Based mostly on the identical Sapphire Rapids silicon as Intel’s recently-launched server elements, the brand new Xeon W SKUs will carry down many (however not all) of the options which have come to outline Intel’s modern server silicon, together with a brand new chipset (W790) and motherboards which can be extra appropriate to be used in high-performance workstations.

As with the brand new Xeon Scalable Elements, the massive three additions listed here are the shift to Intel’s Golden Cove CPU structure – with all of the IPC and clockspeed advantages that brings – together with the addition of help for DDR5 reminiscence and PCIe 5 for I/O connectivity. All of which is a major improve over the combination of Cascade Lake and Ice Lake elements that make up Intel’s earlier product stack. In the meantime in comparison with Intel’s current desktop processor lineup, these are all options that had been pioneered on Alder Lake (12^th Gen Core) again in late 2021, the workstation-focused Xeon W elements are going to be constructing issues out to a a lot bigger diploma.

Beginning on the high, the Xeon W-3400 collection (Sapphire Rapids-112L) will fluctuate from 12 to 56 cores, and all will embrace 112 PCIe 5.0 lanes, help for as much as 4 TB of DDR5-4800 reminiscence throughout eight reminiscence channels, ECC reminiscence (RDIMM-only), Intel vPro, and Intel Customary Manageability (ISM). 4 of the seven W-3400 SKUs (X-series) profit from unlocked multipliers and, as such, formally help overclocking. In the meantime a step down from, the Xeon W-2400 collection (Sapphire Rapids-64L), will provide between 6 and 24 CPU cores paired with a pared-down 64 lanes of PCIe 5.0 connectivity, help for as much as 2 TB of DDR5-4800 reminiscence throughout 4 reminiscence channels, and the entire remainder of the Xeon W trimmings similar to ECC reminiscence.

The brand new Xeon W elements will likely be changing a mish-mash of various Xeon generations from Intel. Whereas Intel did launch some Ice Lake-based Xeon elements in 2021 – the Xeon W-3300 household – these elements had been a supplemental replace of types for Intel’s Xeon lineup, for particular prospects that wanted the additional CPU cores or PCIe bandwidth. For everybody else, the outgoing Xeon W product stack, the circa 2019 W-3200 and W-2200 households, have been primarily based on Intel’s Cascade Lake silicon – which itself was a modest replace to Intel’s Skylake elements. So the significance of the launch of the Xeon W-3400/2400 collection to Intel’s workstation lineup is tough to overstate: this can be a main overhaul and improve of Intel’s Xeon workstation stack.

The brand new Xeon W elements, in flip, will likely be going up towards AMD’s Threadripper Professional 5000 WX elements, that are primarily based on AMD’s Zen 3 structure. The latest Threadripper Professional elements launched final spring, and AMD has primarily had the run of the market by way of CPU efficiency since then, because of a major benefit in core counts and IPC. Even with their new elements, Intel technically nonetheless isn’t fully closing that core depend hole, however the increase in IPC, core counts, and clockspeeds ought to assist to degree the enjoying area by way of general CPU efficiency – although by how a lot stays to be seen.

Intel Xeon W-3400 Collection: ‘Skilled’ Platform with As much as 56 Cores, 112 PCIe 5.0 lanes, and 8-Channel Reminiscence

Intel’s Xeon W-3400 and W-2400 collection workstation processors are primarily based on Intel’s Golden Cove CPU structure, the identical structure as Intel’s Alder Lake (twelfth Gen) desktop processors. Representing the premier line-up from Intel’s 4th Gen Xeon Scalable Sapphire Rapids premium workstation choices, the W-3400 household has seven SKUs in whole. The Xeon W-3400 ranges from a modest 12-core/24-thread half (w5-3425) to a extremely anticipated 56-core/112-thread half, the flagship w9-3495X.

For the Xeon W-3400 collection specifically, these elements are primarily based on Intel’s Sapphire Rapids Excessive Core Depend (XCC) silicon, which is presently utilized in Intel’s higher-end Xeon server elements. The XCC silicon depends on 4 compute tiles, certain collectively utilizing Intel’s newest EMIB interconnect – a primary for a Xeon workstation processor.

The person tiles for a Sapphire Rapids XCC chip are all similar/symmetrical, so every tile gives 1 / 4 of the CPU cores, I/O, and reminiscence channels of the complete chip. As such, every tile can present as much as a most of 32 PCIe 5.0 lanes (112 whole on the w9-3495X), whereas every tile additionally contains as much as two reminiscence controllers offering eight-channel reminiscence throughout the W-3400 collection.

Specializing in the top-end SKU of the Xeon W-3400 household, the Intel Xeon w9-3495X, it has related vibes to Intel’s earlier behemoth Xeon W-3175X, which was launched in 2019 and got here with official help for overclocking. Just like the Skylake-based Xeon W-3175X, the newest Xeon w9-3495X additionally has an unlocked multiplier for overclocking.

The Intel Xeon w9-3495X has 56 cores (for 112 threads), and in contrast to Intel’s desktop elements, each final one among these is a Efficiency (P) core. Additionally current is a complete of 105 MB of Intel’s Sensible L3 Cache, with official help for eight-channels of DDR5-4800 ECC RDIMM reminiscence, with a most capability of as much as 4 TB.

Just like the server half it is primarily based on, w9-3495X has a reasonably toasty TDP score, coming in at 350 Watts. And in apply, peak energy consumption is more likely to be a lot increased underneath full load with Intel’s Turbo Enhance and Turbo Enhance Max 3.0 applied sciences enabled, particularly on 56-unlocked cores. Though it has a base frequency on the 56 Golden Cove cores of 1.9 GHz, it has a turbo frequency of as much as 4.6 GHz, and due to Turbo Enhance Max 3.0 (Intel’s favored core know-how), a handful of cores can increase additional to 4.8 GHz.

The opposite SKUs from the Xeon W-3400 household vary from 36-cores all the way down to 12-core choices, such because the w9-3475X (36C/72T) and the w5-3425 (12C/24T).  In the end, the entire Xeon W-3400 elements provide the identical variety of DDR5 reminiscence channels and PCIe lanes, so what separates the completely different SKUs is CPU core counts, max reminiscence clockspeeds, L3 cache, and naturally, value.

In the meantime, as beforehand famous, 4 of the Xeon W-3400 SKUs – the w9-3495X, w9-3475X, w7-3465X, and the w5-3435X – are all “unlocked” processors. That is one thing Intel hasn’t supplied on a Xeon W half in a number of years and comes with some fascinating ramifications. Apart from probably the most fundamental skill to change the clockspeed multipliers for the CPU, unlocked processors may have their AVX and AMX offsets adjusted to maintain the processors from dropping fairly as a lot underneath heavy SIMD masses. Lastly, all of those elements additionally provide some tuning choices for his or her mesh interconnects, although Intel hasn’t stated what exactly might be tweaked right here.

Costs on the Intel Xeon W-3400 household begin at $1189, with Intel offering pricing on a 1K per unit pricing (tray) and never individually bought retail SKUs. The Xeon w9-3495X has a 1KPU value of $5889, which makes the highest SKU and every subsequent W-3400 SKU dearer than the earlier era of Xeon W-3300 chips, however they do include increased core counts, sooner turbo frequencies, extra L3 cache, and help for DDR5-4800. 

It’s price mentioning that every one of Intel’s W-3400 SKUs function help for as much as 4TB of eight-channel DDR5-4800 ECC reminiscence, even the underside SKU, the w5-3425 (12C/24T). So there are alternatives within the Xeon product stack for techniques that want a complete lot of DRAM, however not essentially a ton of CPU cores. Do be aware, nonetheless, that truly hitting 4TB requires utilizing 2 DIMMs per channel (DPC), which requires backing off to DDR5-4400 reminiscence speeds.

With 112 PCIe 5 lanes obtainable from the CPU (and but extra from the chipset), the Xeon W-3400 chips can help a reasonably large variety of I/O gadgets. This works out to seven discrete x16 graphics playing cards, or as much as 28 x4 high-speed storage gadgets. This, together with core counts and reminiscence channels, is likely one of the main differentiators from the lower-tier Xeon W-2400 collection – and ought to be a welcome improvement for Intel platform customers who had been caught with a fraction of the I/O bandwidth on Intel’s earlier Xeon W elements.

Curiously, 112 PCIe 5 lanes is definitely greater than Intel gives in its Sapphire Rapids server elements. The Xeon Scalable lineup tops out at simply 80 lanes.  This discrepancy comes from the truth that Intel solely enabled 5 of the 7 root ports for his or her server elements, leaving an extra 2 ports (32 lanes) unused. Nevertheless because the workstation Sapphire Rapids elements don’t have to allocate any pins to supporting Intel’s multi-socket UPI hyperlinks, it could appear that Intel has as a substitute allotted these pins to carrying the extra PCIe lanes for the workstation elements. It is price noting that Intel is utilizing the identical socket for each server and workstation chips right here – LGA 4677 – however with the pin adjustments I would not count on them to be suitable.

In the meantime, in one other first for Intel, the corporate has stated that they will help DDR5 XMP 3.0 reminiscence overclocking profiles for RDIMMs. The main points on this announcement are very scant, however at a excessive degree it will give unlocked processor house owners working on W790 the choice of making an attempt to squeeze extra out of their reminiscence if they’ll. Typically talking, reminiscence overclocking and the rock-solid stability of RDIMMs are diametrically opposing targets, so will probably be fascinating to see how this performs out available in the market. The DRAM could have the opportunity clock increased than simply DDR5-4800, however can the registered clock drivers (RCDs)?

As an apart, all of this speak explicitly round RDIMMs is intentional: in a giant change from earlier Xeon W platforms, the Sapphire Rapids Xeon workstation platforms is not going to help UDIMMs. This can be a limitation of the DDR5 specification, which calls for various voltages for UDIMMs and RDIMMs respectively. Whereas UDIMMs take 5 volts, RDIMMs take 12 volts, rendering them incompatible. For those who’ve ever had the possibility to see an DDR5 RDIMM in particular person, you could have observed that they’re even keyed in another way from UDIMMs, so they’re each bodily and electrically incompatible.

In the end, which means that customers will must pair these processors and W790 motherboards with dearer, albeit higher-quality ECC-enabled DDR5 RDIMMs. For dyed within the wool workstation customers that is unlikely to be a difficulty (or perhaps a distinction that will get observed), however anybody hoping to construct an HEDT-style system or low-end workstation on a budget goes to search out that the ultimate price ticket for a Xeon W system goes to be increased than what you would pull off with the W-3200/2200 collection.

Accelerated Computing: AMX and CXL Make the Reduce, Most Area-Particular Accelerators Do Not

For his or her Sapphire Rapids Xeon silicon and ensuing server elements, Intel launched a slew of various acceleration blocks and different accelerator-related options. Between matrix extensions (AMX), varied area particular {hardware} acceleration blocks, and help for Compute eXpress Hyperlink (CXL) for exterior accelerators, Intel ended up devoting a good bit of silicon to non-CPU duties. This has meant that for his or her Xeon Scalable server elements specifically, Intel has opted (if not wanted) to lean on these accelerator options, with one DSA engine enabled in the entire chips. Nonetheless, QAT, DLB, and IAA should not supported. This in lieu of simply uncooked x86 CPU efficiency for differentiating the {hardware} from its predecessors and its competitors.

However for his or her workstation elements, issues are a little bit extra easy, for higher and for worse. In brief, not all of Intel’s accelerated computing options are being made obtainable within the Xeon W-3400/2400 households. So let’s do a fast rundown of which of Sapphire Rapids extra esoteric options made the minimize for Xeon W.

Maybe most critically of all, Intel’s Superior Matrix Extensions (AMX) did make the minimize, and help for them is totally current and enabled on the Xeon W-3400/2400 household. AMX is Intel’s matrix math execution block, and just like tensor cores and different sorts of matrix accelerators, these are ultra-high-density blocks for effectively executing matrix math. AMX isn’t a devoted accelerator, reasonably it’s part of the CPU cores, with every core getting a block, which permits AMX code to be blended with x86 (and AVX) code, and can also be why Sapphire Rapids has damaging clockspeed offsets for utilizing the ultra-dense code.

AMX is Intel’s play for the deep studying market, going above and past the throughput they’ll obtain in the present day with AVX-512 through the use of even denser knowledge buildings. Whereas Intel has AMX-enabled GPUs (Intel Knowledge Middle Max GPU Collection) that transcend even this, for Sapphire Rapids Intel is trying to handle the client section that wants AI inference happening very near CPU cores, reasonably than in a much less versatile, extra devoted accelerator. The brand new AMX models additionally help Bfloat16, making certain that each tier of Intel’s accelerated computing blocks (AVX and AMX) help this frequent mid-precision floating level format for deep studying.

One in every of Sapphire Rapids’ new domain-specific {hardware} accelerator blocks, the Knowledge Streaming Accelerator (DSA), additionally made the minimize. This block is for offloading/accelerating sure operations, similar to knowledge copies and easy computations similar to calculating CRC32s. The DSA block is offered throughout the entire Xeon W SKUs.

Nevertheless you will not discover point out of the remainder of Intel’s accelerator blocks, similar to Intel Dynamic Load Balancer (DLB), Intel In-Reminiscence Analytics Accelerator (IAA), and Intel QuickAssist Know-how (QAT). This even if these accelerators are all a part of the identical purposeful block on the Sapphire Rapids silicon. These different accelerator blocks are all primarily aimed toward servers, so it isn’t stunning to not see their inclusion, however it does imply anybody prototyping code for servers might want to take a look at on an precise Xeon Scalable in the event that they’re utilizing their options.

Lastly, whereas CXL help is absent from Intel’s Xeon W spec sheets, Intel has confirmed to us that it’s in truth supported on each households. The built-on-top-of-PCIe customary for host-to-device connectivity has been within the wings for a number of years now, and Sapphire Rapids is the primary Intel CPU platform to help the know-how. Like a few of these different options, it’s primarily meant for servers, so there’s much less of an impetus to carry it to workstations. Nonetheless, Intel has enabled it for customers trying to leverage its performance.

Intel Xeon W-2400 Collection: As much as 24-Cores, 64 PCIe 5.0 lanes, For Mainstream Workstations

Dropping down a tier, we’ve the Xeon W-2400 collection (Sapphire Rapids-64L), which is designed as a ‘Mainstream’ workstation platform. Xeon W-2400 gives a bit greater than half as many PCIe lanes because the W-3400 SKUs, with 64 PCIe 5.0 lanes obtainable, and the variety of reminiscence channels is minimize in half as effectively to 4 channels. As such, this implies costs are decrease on the W-2400 collection than its beefier W-3400 counterparts, going as little as $359 for the entry-level Xeon w3-2423.

General, the Xeon W-2400 collection will vary from 6 cores as much as 24 cores. Intel is utilizing their Sapphire Rapids Medium Core Depend (MCC) silicon right here, which not like the XCC silicon, is a standard monolithic die. This implies no fancy EMIB packaging is required to construct the chip – as a substitute, Intel solely has to fab one reasonably massive die.

On the top-end of the Xeon W-2400 lineup is the w7-2495X, which options 24-cores/48-threads, 45 MB of Intel Sensible L3 cache, and a TDP of 225 Watts.  Intel additionally has three w5 collection SKUs, and eventually the trio of w3 SKUs.

Like its expert-tier counterpart, the Xeon W-2400 collection gives a constant reminiscence and I/O configuration throughout the complete lineup. This implies 64 lanes of PCIe 5 coming from the CPU, and 4 channels of DDR5 reminiscence, permitting for a most of two TB of reminiscence general. It it additionally price mentioning that solely the w5 and w7 SKUs provide full DDR-4800 reminiscence speeds; the w3 elements are all capped at DDR4-4400. The silver lining? All SKUs drop to this velocity in a 2 DPC configuration, so for those who had been trying to construct a 2 TB system for no matter purpose, you will not get penalized.

Just like the Xeon W-3400 collection, the W-2400 household additionally has a number of unlocked X SKUs in its arsenal, together with the top-tier w7-2495X. Different SKUs with unlocked multipliers embrace the w7-2475X with 20 cores and 37.5 MB of L3 cache, and two w5 SKUs (w5-2465X 16C/32T and w5-2455X 12C/24T). You will not discover any unlocked w3 elements, nonetheless, as all three entry-level w3 SKUs are totally locked down.

Intel W790 Chipset: Helps each Xeon W-3400 and W-2400 Platforms

All of Intel’s Xeon W-3400 and W-2400 collection SKUs profit from Intel vPro and Intel’s Customary Manageability (ISM) applied sciences. Each the Xeon W-2400 and W-3400 households are supported by the related W790 chipset, though CPU-specific options such because the variety of reminiscence channels and PCIe lanes obtainable rely on the processor itself.

A number of the most important options of the W790 chipset embrace a Direct Media Interface (DMI) 4.0 x8 hyperlink between the processor and the chipset itself, in addition to as much as 16 PCIe 4.0 lanes and help for as much as eight SATA 3.0 ports. W790 additionally helps as much as 5 USB 3.2 Gen2x2 (20Gbps) ports, contains an Intel Wi-Fi 6E PHY, and may help 2.5 GbE controllers natively.

Though there is no point out of latest motherboards, there are anticipated to be Intel W790 motherboards from distributors similar to ASUS, GIGABYTE, Supermicro, and ASRock. System integrators similar to Dell, Lenovo, and Supermicro are anticipated to take priority first in delivering options and techniques earlier than DIY builders can get their arms on them.

The ASRock W790 WS motherboard

ASRock emailed us simply earlier than the launch to stipulate its W790 WS mannequin, with a 20+2-phase energy supply, twin 10 GbE controllers, and help for as much as 2 TB of DDR5-4800 ECC RDIMMs throughout eight slots. Though this board helps each Xeon W-3400 and W-2400 processors, this board is just enabled for quad-channel reminiscence.

One thing price mentioning regarding the newest era of motherboards is that W790 boards are more likely to value greater than the C621A-based boards that had been used to help the Xeon W-3300 collection (Ice Lake). It’s because W790 boards have 4 extra lanes of DDR5 reminiscence and 48 extra PCIe 5 lanes to account for. Whereas we count on to see completely different ranges of board designs with completely different slots and I/O configurations obtainable sooner or later, Intel hasn’t specified if a few of these motherboards will help each households, or if distributors will design particular boards across the particular person Xeon W-3400 and W-2400 collection.

Intel’s Xeon W-3400 and W-2400 processors can be found to pre-order from business companions, whereas techniques deployments are anticipated someday in early March. Intel’s anticipated and really useful pricing begins at $359 for the Xeon w3-2423 and goes as much as $5889 for the Xeon w9-3495X.