Monday, November 28th 2022
Intel Sapphire Rapids "Fishhawk Falls" Unveil Scheduled for February with Availability Following in April
Intel's Sapphire Rapids CPUs are designed to represent the company's finest work for the server/enterprise and high-end desktop processor (HEDT) sector. According to Wccftech, we have an idea of the launch date and availability of the HEDT Sapphire Rapids Workstation counterparts, codenamed Fishhawk Falls. The Intel Sapphire Rapids-WS CPU and W790 motherboard unveil should happen on the 7th week of 2023, which means the 12th Feb - 18th Feb launch window. For availability, April of 2023 is scheduled to meet the needs of the upcoming HEDT clients. With a declining HEDT market, Intel is in no rush to deliver the CPUs, with priority being the server Sapphire Rapids designs.
As a reminder, the HEDT models should come with up to 56 cores and 112 threads, 105 MB of L3 cache, 350 W TDP, and 112 PCIe Gen 5.0 lanes. For memory, the platform support 8-channel DDR5-4400 (1DPC) / DDR5-4800 (2DPC) configurations with a capacity of up to 4 TB.
Source:
Wccftech
As a reminder, the HEDT models should come with up to 56 cores and 112 threads, 105 MB of L3 cache, 350 W TDP, and 112 PCIe Gen 5.0 lanes. For memory, the platform support 8-channel DDR5-4400 (1DPC) / DDR5-4800 (2DPC) configurations with a capacity of up to 4 TB.
15 Comments on Intel Sapphire Rapids "Fishhawk Falls" Unveil Scheduled for February with Availability Following in April
Is this octa channel in the normal sense or in the DDR5 sense?
I was about to comment on it being essentially 12th gen in 2023, but since server variant Golden Cove already has 2MB L2, it's basically 13th gen.
96c, 384MB L3 cache, 12 channels DDR5, 128 PCIE-5
SR is pretty much a DOA product for general computing.
The onboard accelerators are their last bet.
This is not a server part. It is a workstation part.
I'm curious about the MCC and HCC SR chips without all the accelerators. Those will be interesting.
Perhaps there are other industries where workstations need more cores, but I'm not involved with any of them. Simulation/rendering on workstations directly is already a dying breed as we've been farming that out to dedicated renderfarms for over a decade. Even that model is getting replaced by both GPU and CPU cloud rendering where you rent several thousand cores for short periods through AWS or similar.
What is the primary HEDT use for more cores and more bandwidth? I'm struggling to think of any applications that need more than the 32T and 128GB most consumer platforms offer. Even with massive AEC city masterplan models, we're not currently hitting limitations in RAM, and the individual workstations have no capability to open the entire model, they stream that from a server running in the Autodesk cloud - the biggest problem isn't the RAM on the workstation, it's the internal address limitations of the software we're using. A workstation with 2TB of RAM would be pointless because the software sort of shits the bed before you hit 100GB. What actually happens if you open and work on segments of the massive central model and this is synced to the central model by a server. Not only is the central model huge, it's too big or complex for any one person to actually work on anyway - we have small teams of 2-10 people working on individual sections.
I bought a 3960x about a year after launch when I finally found it at list price, and a good portion of that decision was based on AMD's assurance that the TRX4s socket were a necessary change that was designed for the long term. Yeah, right.