Discussion Intel current and future Lakes & Rapids thread

[aigomorla]

Ahh the good ol 6000 series..
We also had GT in that line too if i remember correctly...

They wanted to get all car like, and call it GT like grand tourismo, because GT was a thing back then before Fast and Furious started to make it into a meme.

 

[Geddagod]

Would it be correct to assume that with SPR, Intel is running their cores at ~3-4 watts per core?
Using TDP of 350, 60 cores, 1/2 - 2/3 of TDP going to cores while the rest goes to IO/interconnect (using Anandtech's Milan numbers)? I suppose SPR's interconnect might require marginally more energy.
Wouldn't this be equivalent in per-core power of a ~20 watt, 4 core mobile part from Intel? Or am I grossly underestimating IO power consumption in these low power mobile skus?

 

[aigomorla]

Would it be correct to assume that with SPR, Intel is running their cores at ~3-4 watts per core?

don't think so.
Then each cpu would have different TDP based on core count which is not happening.
And when they boost clock, you got a whole different set of values which does matter on core count.

 

[Geddagod]

don't think so.
Then each cpu would have different TDP based on core count which is not happening.
And when they boost clock, you got a whole different set of values which does matter on core count.

Srry, Should have meant top end SPR, as in the 60 core one. 350TDP. Lower core count skus boost higher.
Boost clock matters on what workload, but the AFAIK core power only deviates depending on how cache heavy or IO heavy the workload is, SPEC INT vs FP was 1/2-2/3, which is why I gave the range I did. There's also prob a 'base' IO power, no matter how well the code fits into private L2 and doesn't dip into L3 or RAM, there's still going to be some baseline of power for mesh that's not given to the core itself to boost higher.

 

[eek2121]

Srry, Should have meant top end SPR, as in the 60 core one. 350TDP. Lower core count skus boost higher.
Boost clock matters on what workload, but the AFAIK core power only deviates depending on how cache heavy or IO heavy the workload is, SPEC INT vs FP was 1/2-2/3, which is why I gave the range I did. There's also prob a 'base' IO power, no matter how well the code fits into private L2 and doesn't dip into L3 or RAM, there's still going to be some baseline of power for mesh that's not given to the core itself to boost higher.

Depends on how many cores are active, the workload, etc.

Worst case scenario sees the chip running at 1.9ghz, so the 3-4w/core might be valid.

 

[A///]

I assumed that Ultra was the Cove+Mont hybrid higher end SKUs or just Cove SKUs perhaps for the Core 9 Ultra SKUs.

On the other hand Core regular would be just Mont at the low end and Cove+Mont mainstream SKUs toward the higher end of that range at Core 7.

🤔

that would fall in line with what I said would it not? i said that under the guise of intel having raptor lake refresh laptop skus too or it'll go into effect next year or if not the regular core series would be raptor lake mobile as it is now. this is all very confusing for this dinosaur.

 

[Exist50]

Apparently Intel's doing yet another round of layoffs, this time hitting the engineering teams in particular. Heard an additional ~7% of the Xeon engineering team is being laid off. That's on top of a few percent from previous rounds. Other teams are reportedly seeing similar numbers.

 

[itsmydamnation]

Apparently Intel's doing yet another round of layoffs, this time hitting the engineering teams in particular. Heard an additional ~7% of the Xeon engineering team is being laid off. That's on top of a few percent from previous rounds. Other teams are reportedly seeing similar numbers.

Jesus , any hint on the cause ?

 

[Geddagod]

Well... on the flip side of that news almost...

https://twitter.com/x/status/1673647482050600961

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.

 

[CakeMonster]

I'm kind of tired of the stock/management discussion that people tend to feel strongly about while lacking insight, but considering the layoffs of the last few years in addition to seeing the same short term policies ruining the future of companies me and my acquaintances work for, I do worry a bit.

 

[Thunder 57]

I'm kind of tired of the stock/management discussion that people tend to feel strongly about while lacking insight, but considering the layoffs of the last few years in addition to seeing the same short term policies ruining the future of companies me and my acquaintances work for, I do worry a bit.

And here I was thinking you were about to say "I'm kind of tired of the hentai girls these leakers use".

 

[CakeMonster]

And here I was thinking you were about to say "I'm kind of tired of the hentai girls these leakers use".

I guess there's no point in fighting internet battles you can't possibly win...

 

[A///]

while intel did layoff engineers they rehired a good bit. intel could trim the fat in non critical roles, like their marketing department who clearly seem to do lines of coke when they come up with such crap as ultra. with layoffs come job listings.

the way i see it is you're passing around engineers back and forth between fabbing and fabless companies like one big swingers party. a little good and bad each round.

 

[Geddagod]

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?
This sounds completely unbelievable, unless they are also talking about developing the 'next gen' core architecture? As in designing GLC and designing SPR took ~6 years.
I always thought it was 1-2 years for a new arch + 3 years for product definition to product launch, so maybe?

Would be glad if anyone can help point out any discrepancies in this chain of reasoning.

 

[Exist50]

Jesus , any hint on the cause ?

Nothing special. They're trying to shore up their falling financials. See Gelsinger's claims of $8-10 billion in savings by 2025? Who do you think is paying that price?

while intel did layoff engineers they rehired a good bit. intel could trim the fat in non critical roles, like their marketing department who clearly seem to do lines of coke when they come up with such crap as ultra. with layoffs come job listings.

These layoffs go hand in hand with an extended hiring freeze, so add in a few more percent for attrition.

 

[igor_kavinski]

These layoffs go hand in hand with an extended hiring freeze, so add in a few more percent for attrition.

Hopefully, someone really capable isn't impacted. It would suck for Intel if they let someone go who then went on to AMD or Apple and created something great. What a way to get back at Intel, that would be.

By the way, I think Jim Keller hired Raja Koduri coz both feel that Intel didn't give them a fair chance, though in Raja's case, he underestimated how much dough Intel was prepared to lose for his "vision".

 

[Exist50]

Hopefully, someone really capable isn't impacted.

Those people left / are leaving in droves because of the pay cuts. But that's nothing new. There are entire companies staffed with the engineers Intel's driven away over the years. Ampere was basically founded by Intel's former Oregon Core team. It's this shortsightedness that keeps me pessimistic about Intel's future. The last time they thought they could lay a bunch of people off without consequence gave us SPR. I think Gelsinger has clearly spent too long in management.

 

[A///]

These layoffs go hand in hand with an extended hiring freeze, so add in a few more percent for attrition.

I wasn't going by the recent layoffs but ones going back to under krzanich.

 

[itsmydamnation]

Those people left / are leaving in droves because of the pay cuts. But that's nothing new. There are entire companies staffed with the engineers Intel's driven away over the years. Ampere was basically founded by Intel's former Oregon Core team. It's this shortsightedness that keeps me pessimistic about Intel's future. The last time they thought they could lay a bunch of people off without consequence gave us SPR. I think Gelsinger has clearly spent too long in management.

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?

So he is trying to not crater products/ revenue to badly while at the same time cut costs and thus not really caring about long term talent retention in product/design teams ?

 

[A///]

Those people left / are leaving in droves because of the pay cuts. But that's nothing new. There are entire companies staffed with the engineers Intel's driven away over the years. Ampere was basically founded by Intel's former Oregon Core team. It's this shortsightedness that keeps me pessimistic about Intel's future. The last time they thought they could lay a bunch of people off without consequence gave us SPR. I think Gelsinger has clearly spent too long in management.

only seeing this now my mind wandered reading gedda's post on time tables. this is an issue which has been plaguing the entire industry for years. stagnated wages and too much fussy footing by management. they don't give a damn if their staff leave for other companies because they know some other dumbass will take the next available low paying job when they could make potentially make more elsewhere. if your claim is correct and not bs intel is yet not in the find out stage of that infamous line.

 

[A///]

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?
This sounds completely unbelievable, unless they are also talking about developing the 'next gen' core architecture? As in designing GLC and designing SPR took ~6 years.
I always thought it was 1-2 years for a new arch + 3 years for product definition to product launch, so maybe?

Would be glad if anyone can help point out any discrepancies in this chain of reasoning.

if a process node was late...

 

[Geddagod]

if a process node was late...

I'm using the date it was originally supposed to launch- 2021, though. If you included the node delays and when it actually launched - 2023, you would be seeing ~8 years.

Going back, the only way I see this working is this-
Broadwell- 2016
Skylake- 2017
Icelake- 2018
Sapphire Rapids- 2019

Which would mean SPR was delayed for ~4 years total, not the reported ~3. It would give SPR design to planned launch date ~4 years, which is much more reasonable imo.

 

[A///]

I'm using the date it was originally supposed to launch- 2021, though. If you included the node delays and when it actually launched - 2023, you would be seeing ~8 years.

Going back, the only way I see this working is this-
Broadwell- 2016
Skylake- 2017
Icelake- 2018
Sapphire Rapids- 2019

Which would mean SPR was delayed for ~4 years total, not the reported ~3. It would give SPR design to planned launch date ~4 years, which is much more reasonable imo.

you need to consider everything else that was using up fab time. what you're focusing on is a line of products that is a part of their internal issues. intel canceling certain products in loop while others were being placed into the loop that were better is what i would expect from a company that's been tripping on its own feet for a few years now. it's why a while back i was having fun picking on some people here for taking the sapphire rapids results which weren't great when it's an ancient product that intel is now getting out. of course none of this works in intels favour and this is by no means a party trick liking snapping a bunny out of someone's top hat.

at the end of the day intel's in a awful situation where they are bleeding but existing contractsd and their ability to manufacture and ship more skus than amd who will forever remain supply ocnstrained when peaked out due to their reliance on tsmc will be able to ship and thus remain in people's good graces to some level. amd has been making a sweeping comeback over six years now but there are still old school people my age or older who sit in management positions who are policy pushers who would argue for buying intel products over amd when hardware refresh periods roll around. it's a mix of not keeping up with the times or thinking this is temporary for amd or they read some bull online about amd consumer level hardware having issues with the pro, tr and epyc will hav ebeen more thoroughly tested. seeing hyper scalers buy into amd doesn't sway these people. I have friends in this field my age or not much older like doug's age who are real old school and continue to ignore AMD's wins and keep buying intel. there's more reason than being dumb and old of course.

 

[moinmoin]

It's this shortsightedness that keeps me pessimistic about Intel's future. (...) I think Gelsinger has clearly spent too long in management.

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. 🤷

 

[A///]

if the new ones weren't producing you try the old. but im also not in management. at least gunslinger isn't starting every day with a prayer at work.