Utilizing AI to Construct Higher Processors: Google Was Simply the Begin, Says Synopsys

With the launch earlier this yr of Intel’s Ice Lake Xeon Scalable platform, the primary Intel enterprise platform based mostly on 10nm, we had been at all times questioning to what extent this silicon would penetrate different markets. In the present day Intel is answering that query by launching the Xeon W-3300 household, utilizing the identical Ice Lake Xeon silicon however in a extra targeted single socket platform. Xeon W-3300 will supply processors from 12 core as much as 38 cores, peaking at 270 W TDP, however providing help for as much as Four TB of reminiscence.

The Xeon W-3300 Household

That is the third technology of Intel’s Xeon W processor household, initially highlighted by the overclockable 28-core Xeon W-3175X based mostly on the Skylake household of enterprise CPU architectures. Intel made an enormous effort with the launch of that processor, and adopted it shortly with the Xeon W-3200 collection based mostly on Cascade Lake, providing extra PCIe lanes and a rise in frequencies. The brand new Ice Lake Xeon W household makes use of Intel’s Ice Lake structure, as we highlighted in our preliminary Xeon Scalable overview of the platform, affording a sizeable gen-on-gen bump efficiency enchancment in opposition to the earlier Intel processors. In that overview we’ve highlighted that Ice Lake supplied +19% IPC enhance, and the achieve in core depend and TDP supplied an uplift regardless of some regressions in frequency in comparison with Cascade Lake.

Whereas the Xeon Scalable line presents as much as 40 core variants, Xeon W-3300 will go so far as 38 cores, however Intel is focusing totally on the 32-core model as its lead product. The brand new Xeon W-3365 with 32 cores is designed to be aggressive in opposition to AMD’s Threadripper Professional 3975WX, which we not too long ago reviewed, providing the identical variety of cores, the identical DRAM help, and has some tradeoffs.

Intel Xeon W-3300 Ice Lake SKUs
AnandTech	Cores	Base Freq	1T Turbo	nT Turbo	L3 (MB)	TDP	Value (1ku)
W-3375	38 / 76	2500	4000	3300	57	270 W	$4499
W-3365	32 / 64	2700	4000	3500	48	270 W	$3499
W-3345	24 / 48	3000	4000	3700	36	250 W	$2499
W-3335	16 / 32	3400	4000	3700	24	250 W	$1299
W-3323	12 / 24	3500	3900	3700	21	220 W	$949

The highest processor is the Xeon W-3375 with 38 cores and 76 threads. The bottom frequency of two.5 GHz is offset with a 4.Zero GHz single core turbo and a 3.Three GHz all-core turbo, enabled by the 270 W TDP, which will increase the TDP of the highest Xeon W processor from 255 W within the earlier technology. 

All of the processors will help 64 lanes of PCIe 4.0, 8-channel DDR4-3200 reminiscence (up from 6-channel), and with 256 GB LRDIMMs as much as Four TB per socket (16 modules). The usual Ice Lake enterprise core presents 1.5 MB of L3 cache per core, and for all of the processors right here that customary stays the identical, nonetheless the 12-core model makes use of 1.75 MB of L3 cache per core. Intel has not disclosed the die configurations of the decrease core depend components, nonetheless the 28+ core components are nearly actually utilizing the 40 bodily core configurations.

These new processors use the identical LGA4189 socket because the mainline Xeon Scalable platform, however are enabled by a brand new technology of C621A chipset which permits for Wi-Fi 6E and Thunderbolt Four growth on motherboards ought to the board OEM constructed it in. Additional details about the C621A has not been detailed presently.

The opposite angle new to the Xeon W household is the addition of recent directions for cryptography – Ice Lake cores now help SHA extensions, have doubled AES throughput, and VPMADD52 instruction units. On high of this, Intel is pushing its newest technology of DLBoost know-how to hurry up AI workloads by environment friendly use of the on-board AVX512 items. Though not disclosed, we're underneath the impression that the Xeon W-3300 components have twin AVX512 FMAs per core.

Some Xeon W-3365 Benchmarks

Forward of this launch, we had been contacted by a system integrator who shipped us a system that arrived a few days in the past with the W-3365 put in. Sadly the system didn’t survive the transatlantic cargo totally intact, and we’re nonetheless debugging precisely why. However we had been capable of strip out the board and CPU to do some preliminary testing. We’re going to be doing a full overview hopefully subsequent week.

• Processor: Intel Xeon W-3365 (32-core Ice Lake, 270W)


• Cooler: Noctua NH-U14S LGA4189


• Motherboard: Supermicro X12SPA-TF


• Reminiscence: Eight x 16 GB of DDR4-3200 RDIMM (128 GB)


• Graphics: NVIDIA RTX 3090 FE


• Storage: Intel Optane P5800X 800GB + 2x1 TB Samsung 980 Professional


• Energy Provide: 2x Corsair AX860i


• Working System: Home windows 10 20H2

We’re evaluating this processor to the 28-core W-3175X and the 32-core TR Professional 3975WX in these preliminary exams. Intel is focusing on this processor very a lot at workstation-dedicated workflow. Additionally, the advantage of all these processors listed being 32-core or beneath means they match inside Home windows’ 64-thread masks restrict, which may restrict some exams to 64 threads.

On the subject of energy, we're measuring a peak 292 W throughout AVX-512 workloads and 286 W in AVX2 workloads, nonetheless that is preliminary information from a couple of exams we have been capable of run up to now.

We are able to see that the Ice Lake Xeon W-3365 already pushes forward of the Skylake W-3175X in nearly all our benchmark suite, save for a few exams that appear to be cache/frequency-dependent. Towards the Zen 2-based 3975WX it's extra benchmark dependent. We're ready for AMD to launch its Zen 3-based Threadripper and TR Professional over the subsequent few months.

Xeon W-3175X vs Xeon W-3365 vs TR Professional 3975WX
Intel Xeon W-3175X	Intel Xeon W-3365	AnandTech	AMD TR Professional 3975X
$2999	$3499	Price	$2750
Skylake Xeon	Ice Lake Xeon	Microarchitecture	Zen 2
28 / 56	32 / 64	Cores / Threads	32 / 64
3100 MHz	2700 MHz	Base Frequency	3500 MHz
4300 MHz	4000 MHz	Turbo Frequency	4200 MHz
1024 KB	1280 KB	L2 Cache / core	512 KB
38.5 MB	48 MB	L3 Cache	128 MB
255 W	270 W	TDP	280 W
6 x DDR4-2933	Eight x DDR4-3200	Reminiscence	Eight x DDR4-3200
512 GB	Four TB	Reminiscence Capability	Four TB
PCIe 3.Zero x48	PCIe 4.Zero x64	PCIe	PCIe 4.Zero x128
PCIe 3.Zero x4	PCIe 3.Zero x8	Chipset Hyperlink	PCIe 4.Zero x8
Sure	Sure	AVX-512	No

We’re hoping to place extra time into testing this platform, together with the remainder of the system despatched to us; workstations hardly ever focus solely on the CPU, and rely very a lot on the put in graphics, storage, and reminiscence relying on the completely different bottlenecks of the workload. Keep tuned for that overview.