Discussion Intel current and future Lakes & Rapids thread

[eek2121]

This so much. This was actually the main thing I hoped Gelsinger would change. Apparently he restarted the traditional Intel practice of sweeping layoffs right after bringing back his old fellow comrades in arms. It's not like there's a lesson to learn in all this after all. 🤷

Tis the season for tech layoffs. Not really specific to Intel. When interest rates jump, companies make a big show of focusing on profitability, primarily via cost cutting.

Due to labor laws in the US and elsewhere, companies usually do not pick and choose who gets cut. Instead, a uniform, well understood policy is usually laid out. That is to say that you could be the most talented engineer at a company and still get cut.

I understand why Intel did/is doing layoffs, but it will only hurt them in the long run.

 

[Exist50]

Granite Rapids AP has been pictured and seems to be sampling to customers. As long as it's not nearly as buggy as SPR was, I can see it coming out on time.

GNR had plenty of its own issues, but it was staffed around when the reality of ICX/SPR was hitting Intel, so I expect it to be in far better shape.

Just read that Sapphire Rapids started getting designed in 2015. If the CPU was originally scheduled for release in late 2021, does that mean Intel's DC development for a CPU starts ~6 years before launch?

That might well be the case, but there's nothing typical about SPR, even by Intel's historically poor standards. It was originally supposed to use WLC and PCIe 4.0. Could probably add years just on redefinitions.

maybe he knowns and thinks that high end architectures are becoming more commodity and the real value is semi manufacturing of which there is only them and TSMC left?

I doubt that's the case, even in his head. My read is that he promised investors billions of cost reduction to get them to go along with his spending plan for the fabs, and now he's stuck with the reality of trying to make good on that promise. Maybe he's hoping that once the fabs are stabilized, he can make up the lost ground in design, but I'm not really convinced that he's fully internalized the consequences. Especially since the AXG cuts mean risking any chance of Intel competing in AI.

 

[Geddagod]

Intel Xeon Max 9480/9468 Show Significant Uplift In HPC & AI Workloads With HBM2e Review - Phoronix

www.phoronix.com

SPR-HBM review for anyone interested.

 

[nicalandia]

500 Watts....

 

[H433x0n]

500 Watts....

View attachment 82384

Things are so bad at Intel that Granite Rapids is somehow worse than Sapphire Rapids! They went backwards with GNR using 42% more power to power 60C compared to SPR.

That or there's something lost in translation.

 

[H433x0n]

Due to labor laws in the US and elsewhere, companies usually do not pick and choose who gets cut. Instead, a uniform, well understood policy is usually laid out. That is to say that you could be the most talented engineer at a company and still get cut.

I don't think that's the case. That hasn't been how things work at my employer and I doubt they're breaking US labor laws.

 

[Hitman928]

I don’t know what else to say but ugh. . .

Bug Forces Intel to Halt Some Xeon Sapphire Rapids Shipments

Sapphire Rapids hits another roadblock, the resumption date is unknown.

www.tomshardware.com

 

[Thunder 57]

I don’t know what else to say but ugh. . .

Bug Forces Intel to Halt Some Xeon Sapphire Rapids Shipments

Sapphire Rapids hits another roadblock, the resumption date is unknown.

www.tomshardware.com

That must have been a kick in the nuts to the author, who happens to love Intel.

 

[Geddagod]

That must have been a kick in the nuts to the author, who happens to love Intel.

Don't think so

 

[Thunder 57]

Don't think so

I can show you why but it will have to wait until tomorrow. It's bed time.

 

[Geddagod]

I don’t know what else to say but ugh. . .

Bug Forces Intel to Halt Some Xeon Sapphire Rapids Shipments

Sapphire Rapids hits another roadblock, the resumption date is unknown.

www.tomshardware.com

Lmao SPR is still buggy. I find it a little surprising how it's only for the MCC models though, I thought the XCC was always the buggier variant.

 

[Geddagod]

Things are so bad at Intel that Granite Rapids is somehow worse than Sapphire Rapids! They went backwards with GNR using 42% more power to power 60C compared to SPR.

That or there's something lost in translation.

Or they just increased frequency?
SPR does the same thing- they have 16C variants rated at 270 watts - 185 watts.

Also he also posted this:

Which is just like, what???

 

[H433x0n]

Or they just increased frequency?
SPR does the same thing- they have 16C variants rated at 270 watts - 185 watts.

Also he also posted this:
View attachment 82389
Which is just like, what???

I was mocking the doomer-ism of this forum and being sarcastic. I admit the sarcasm doesn’t come out well in text format.

GNR-AP is going to have much more than 60 cores. Whomever tweeted that doesn’t have the full context / info.

 

[Exist50]

500 Watts....

View attachment 82384

I don't think it's worth paying any attention to those numbers. Those don't even seem like AP core counts, except for maybe the 80c one on the lower end of the spectrum. Regardless, they're firmly in the ES stage. More realistic SKUs should start showing up beginning of next year, if they want a mid-ish year ship date.

 

[Exist50]

Also he also posted this:

Which is just like, what???

What about that?

 

[Geddagod]

GNR X3 is PRQ already? > A year before launch?

 

[lightisgood]

GNR X3 is PRQ already? > A year before launch?

EMRs, planed late this year, was PRQ in April.
GNR-AP could be released in Jan-Feb/24.

 

[Geddagod]

EMRs, planed late this year, was PRQ in April.

Really? Did not know that.
Are server chips PRQ ready 6-8 months before launch?

GNR-AP could be released Jan-Feb/24.

Intel claimed SRF will launch 1H 2024, and Granite Rapids shortly after. While I suppose that could meant SRF 1Q and GNR 2Q, Intel's reluctance to claim GNR is also coming 1H 2024 sounds like a silent admission that it's 2H, though I would expect Q3 since they did specify 'shortly after'.

 

[Exist50]

GNR X3 is PRQ already? > A year before launch?

I believe that's talking about die configs. They have a single compute die package for power on activities, but the shipping version will have 3. I don't think that's supposed to mean that GNR has PRQ'd already, which it certainly has not.

EMRs, planed late this year, was PRQ in April.

It has not. Intel tends to announce in their earnings when things PRQ. And a launch event usually follows a quarter or so later.

GNR-AP could be released in Jan-Feb/24.

Unlikely. Very optimistically, they PRQ in Q2'24, and have an official launch event maybe September/October.

 

[Geddagod]

I believe that's talking about die configs. They have a single compute die package for power on activities, but the shipping version will have 3. I don't think that's supposed to mean that GNR has PRQ'd already, which it certainly has not.

This makes much more sense lol.

That might well be the case, but there's nothing typical about SPR, even by Intel's historically poor standards. It was originally supposed to use WLC and PCIe 4.0. Could probably add years just on redefinitions.

Would this have meant ICL would have used Palm Cove?

 

[Geddagod]

Also this is a bit unrelated to this thread, but this chart (power breakdown of Intel Haswell) in Chips and Cheese did intrigue me

Is Core Static essentially the core 'leakage' while Cache dynamic, Micro-op Execution, Instruction Decoders, etc etc is all the core dynamic power loss (where parts of the CPU are actually being used)?
And has the Core static ratio increased with newer cores as they use newer nodes, and newer nodes have more leakage power? (I'm assuming better power gating can only do so much).

 

[Exist50]

Would this have meant ICL would have used Palm Cove?

Nah, though they did kill Ice Lake server, and then bring it back from the dead as SPR started to shift out. Was a disaster in its own right, only overshadowed by the even bigger disaster that was SPR.

 

[Exist50]

This so much. This was actually the main thing I hoped Gelsinger would change. Apparently he restarted the traditional Intel practice of sweeping layoffs right after bringing back his old fellow comrades in arms. It's not like there's a lesson to learn in all this after all. 🤷

What I'm struggling to reconcile here is the optimism I've heard from so many seasoned engineers vs the seemingly dismal reality I see as a 3rd party. Maybe people's opinions have changed over the past few months (not like I send out surveys, lol), or maybe my sample is biased (very likely), but something isn't quite sitting right with me. I guess it's ultimately on Gelsinger to prove that he's not just more of the same, but it'll take many years to see the end result.

 

[Exist50]

Is Core Static essentially the core 'leakage' while Cache dynamic, Micro-op Execution, Instruction Decoders, etc etc is all the core dynamic power loss (where parts of the CPU are actually being used)?

Sounds about right. As for dynamic vs leakage power, I think FinFET helped keep leakage from getting out of hand, but I can't claim to have seen a breakdown recently. TSMC's 20nm node was awful though, if that's their baseline.

 

[moinmoin]

What I'm struggling to reconcile here is the optimism I've heard from so many seasoned engineers vs the seemingly dismal reality I see as a 3rd party. Maybe people's opinions have changed over the past few months (not like I send out surveys, lol), or maybe my sample is biased (very likely), but something isn't quite sitting right with me. I guess it's ultimately on Gelsinger to prove that he's not just more of the same, but it'll take many years to see the end result.

If you are mostly in contact with people in R&D that seeming disconnect is actually pretty normal. Essentially you wouldn't want to work on something you are pessimistic about to begin with. Productization of R&D work is the crucial part though, the industry is full of cases where promising developments were either never brought to market or positioning (like missing the right time to launch, Intel is getting good at that...) was so mismatched that they were pretty much DOA.