Sapphire Rapids is officially out

[c-o-pr]

A module with a 10,000 pin connector seems like a gross design cut, but since computers are now designing computers, maybe it's all beyond human scale.

A phone today is a design with upwards of 5 trillion device nexus. But its modular edge is order-10 connections.

What is it about servers that the modular edge needs to be order 10,000?

Obviously, to allow mixing and matching of capability variations, but this assumes there's something that isn't known about application config the must be deferred to a downstream integrator. With a device so complex, why is it not known?

It's like Intel is saying "we still don't know for sure what this is good for" about the most complex devices ever built.

I guess there's some sort of metaphysical hope in this.

Another way to see it is that such a vast edge affords fine-grained manipulation of rent based on precise application of capabilities, i.e., withholding capacity from customers.

Put a different way, when you can create an infinitely powerful device, profit attends tactical withholding of the application of that power.

The more I think about this, the more strategic Apple's choice to move away from these designs looks to me: they didn't want to end up in a rent trap for the core (npi) of their products.

 

[ckrueger99]

 

[c-o-pr]

I'm happy to see some SR HW, but all I came away with was the desktop has finally stagnated because even bloated SW can't waste all that power.

HW has truly and finally outclassed every desktop PC use-case.

OTOH it's still about 10 orders of magnitude to weak to truly simulate a brain, so it looks like we're stuck for about the next 20 years with 4K Clippy.

 

[sash11ua]

Got me Supermicro X13SRA-TF with w5-3425. Still in the initial stages of trying to boot. Stuck at kernel issue
[EB|#LOG:EXITBS:START]
MSR register is unlocked out of the box.

Attaching ACPI and Opencore debug log.

 

[etorix]

Going that far is already a good start. But with an unsopported GPU on top of the unsupported motherboard and CPU you've not chosen an easy path… I hope this is not your first Hackintosh!

 

[sash11ua]

Going that far is already a good start. But with an unsopported GPU on top of the unsupported motherboard an CPU you've not chosen an easy path… I hope this is not your first Hackintosh!

Definitely not the first one, but so far it's been harder than my dell laptop)
Got a lot of info from windows installation, so I'm able to spoof GPU. The PCI path is weird on this board.
ssdtTime could not find a CPU in dsdt. I had to manually do the PLUG table. No EC device either, so fake EC table used.

 

[etorix]

ssdtTime could not find a CPU in dsdt. I had to manually do the PLUG table.

Oldstyle SSDT-PLUG is actually not needed for the last versions of macOS. CPUs are there in DSDT, and the issue for macOS is known.

                Device (C000)
                {
                    Name (_HID, "ACPI0007" /* Processor Device */)  // _HID: Hardware ID

But your results puzzle me: Without SSDT-PLUG-ALT (or SSDT-CPU-WRAP) the boot process should halt just after loading ACPI tables and not even reach the stage of launching the kernel! Has macOS learned to use ACPI processor device declarations?

 

[sash11ua]

Trying to troubleshoot kernel boot issues. Attaching a serial output log. Not sure where to move from here...

 

[ramazarusx]

Make sure the extended APIC is disabled, and patch registers regardless if they're unlocked. @EVILGENIUS and I hit the same problem a while ago and if memory serves me right it was something in the bios settings. Not sure how is it on the Supermicro but on the Asus and ASRock the ACPI tables are a complete mess. I won't be able to do anything for few more days test wise since I'm still waiting on Asus to return my RMA'd board. (Second board with POST issues after clearing BIOS, great QC on Asus part).

 

[c-o-pr]

Intel Xeon MAX 9480 Deep-Dive 64GB HBM2e Onboard Like a GPU or AI Accelerator

We deep-dive into the Intel Xeon MAX 9480 and see several surprises when combining Xeon cores and HBM2e memory (like a GPU uses)

www.servethehome.com