Discussion Intel current and future Lakes & Rapids thread

[igor_kavinski]

Trying to be funny clown

I try to please

 

[Hulk]

Apparently the i7 sku gets 12 little cores, up from 8. They disable 1 of the e-core clusters rather than both of them.

The enigma that was Raptor Lake Refresh is beginning to take shape.
Better yields can result in higher clocks up and down the stack while upping the i7 to 3 E clusters could provide enough differentiation from 13 gen from Intel's point of view. Of course if pricing is right they may get away with it. For a while.

 

[A///]

ignoring the circus here but i found jhh's comments about intel fab for future nodes very interesting. i assume he was asked once he came down from his high. i do not think this was meant to make tsmc offer better pricing because tsmc could drop nvidia and get that customer space back filled quickly. jensen's words are straight shooting that intel idf 2.0 is working out as planned and Intel is getting success on their future nodes.

Call me a crazy old man but I have a feeling intel will outpace tsmc within the next 2-3 years.

Where do you get 20% performance uplift? I would be surprised to see a 20% uplift even in multi threading.

20% uplift is what I expect to see from Raptor lake to Arrowhead lake. mtl on desktop being canceled paves hte way for more fab time towards laptops which are undoubtably an intel strong point even now because amd can't churn out the volume required by odms.

raptor on lga1700 is a nice twist and intel is technically supporting 3 gens of processors on one chipset. the new updated boards from some partners allow for 5 gigabit ethernet and wifi7, only worth the money if you haven't got the chipset yet. dunno who makes the 5 gbe chip for these, does intel have one? there's a lot of bull crap swirling around like a cofee cake with swirled cheese depending on how advanced intel had gone with the refresh to tame power use only to push the chip further and use the same power as the normal launch did or a little more with more headroom.

 

[aigomorla]

I dont see how Intel is going to win this, looking at SPR.
It feels they are so far behind now, its kinda depressing.

I guess this is how bulldozer felt.

But consumer cpu's is like whatever....
Its the HEDT and Enterprise market which impact vol the most.
And Intel's Enterprise lineup is not competitive, or an alternative, it more feels like a "last resort" kind of thing.

 

[Exist50]

Did Sunny Cove/Cypress Cove, architecturally, only have an ~10% gain in perf/watt over Skylake? This seems pretty low compared to Zen 3<Zen 2 which had closer to a 15-20% gain in perf/watt.

(power draw at the wall Kitguru, CB 20)
~170 watts, 6 zen 3 cores had ~17% gain in perf/watt
~200 watts, 16 zen 3 cores had ~21% gain in perf/watt
~220 watts, 12 zen 3 cores had ~21% gain in perf/watt
Meanwhile Cypress Cove
6 cypress cove cores ~100 watts (CPU only) had ~9% gain in perf/watt (hardware luxx CB 20)
8 cypress cove cores ~180 watts (CPU only) had ~7% gain in perf/watt (hardware luxx) .

Also the way I ended up doing perf/watt was by matching up similar power consumption among same core count models.

Sunny Cove is a horrible incremental design, made all the worse because of how long it took after Skylake. Willow Cove and Golden Cove tried to plaster over that mistake, but with mixed results.

 

[Exist50]

On another topic, I just learned two disturbing new details about Intel yesterday.

1) They lost their newly-appointed head of CPU cores (Boyd Phelps). He was an Intel lifer, so that's surprising. No clear answer as to why. Some suspect politics, but not sure.

2) The budget cuts they've been making are actually retroactive. As in, they don't have to cut 10% or whatever from the rest of the year's budget, but rather 10% of the total year's budget through spending reduction in the remainder. That's going to be an insane number of layoffs.

 

[H433x0n]

I dont see how Intel is going to win this, looking at SPR.
It feels they are so far behind now, its kinda depressing.

I guess this is how bulldozer felt.

But consumer cpu's is like whatever....
Its the HEDT and Enterprise market which impact vol the most.
And Intel's Enterprise lineup is not competitive, or an alternative, it more feels like a "last resort" kind of thing.

SPR is basically a time capsule from Q3 2021. I think a lot of criticism for SPR is fair since it is a disappointment from a performance standpoint but I also think most of it is misdirected. SPR was already in motion before Pat Gelsinger, so saying SPR is indicative of the current direction I feel is inaccurate. SPR is essentially a microcosm of everything that went wrong with Intel in that era. SPR was designed to be Intel's first tiled server product as well as the first to incorporate DDR5, PCIe5 and CXL. That's an extremely heavy lift.

 

[dullard]

2) The budget cuts they've been making are actually retroactive. As in, they don't have to cut 10% or whatever from the rest of the year's budget, but rather 10% of the total year's budget through spending reduction in the remainder. That's going to be an insane number of layoffs.

It was widely reported that the layoffs might be 20%. But I could see how people focus on just the 10% number and not the full effect of that.

Intel Comments on New Layoffs, Budget Cuts in Client CPU and Data Center Groups

Intel is reducing its workforce.

www.tomshardware.com

Intel planned to reduce its budget by 10%, resulting in "as much as" 20% layoffs in the impacted groups (CCG and DCG)

 

[mderbarimdiger]

Call me a crazy old man but I have a feeling intel will outpace tsmc within the next 2-3 years.

OK, will do whatever your asking for, so, let me call you: crazy old man.

Seriously, intel is officially talking about their aspirations to be the "second" biggest fab by 2030, and intel is not a company that is shy or afraid to hype, yet your feelings dare to think ISF will "outpace" TSMC by 2026?

 

[Exist50]

It was widely reported that the layoffs might be 20%.

Not widely reported. That number comes from a rather disreputable blogger, based on simple estimates of opex expenses. No one seems to know what the real percent is, and Intel's apparently not saying even internally. Which of course makes people assume the worst.

 

[clemsyn]

Call me a crazy old man but I have a feeling intel will outpace tsmc within the next 2-3 years.

I hope this will be true but I highly doubt it. We really do not know what Intel has under their sleeve so I would not even consider calling you a crazy old man. If you turn out to be correct then I will be the crazy old man

 

[A///]

I hope this will be true but I highly doubt it. We really do not know what Intel has under their sleeve so I would not even consider calling you a crazy old man. If you turn out to be correct then I will be the crazy old man

On the basis of implementing bspd and gaafet intel will be first.there's a lot of hullaballoo of dlvr working on desktop and being improved through revisions allowing more headroom at the same or similar power envelopes. IFS cutting wafer pricing low enough but still making a small profit due would slowly leach customers from tsmc. If there's a small performance difference between them and TSMC the customers will pick intel on costs basis alone. they can maintain their end customer msrps while paying less giving them larger margins. tsmc's future node development lies mostly with their government and their large customers like apple. remember gelsinger said he'd love to get apple back, but at ifs because he knows Apple won't bother going x86 again. If apple bails most of their wafers from tsmc intended for desktop computers where apple may need to increase power/ lose efficiency to keep an edge over x86 and keep efficiency sensitive platforms at tsmc like the ipad, cpus destined for the mba and the iphone then tsmc will need to ask their government for more assistance to keep the same level of research going. That's not cheap.

 

[msj10]

I know it's MLID so most of the stuff is probably wrong but it's still interesting.
I guess these "Extra big cores" are the rumored Royal Core.
also cougar cove having such a big performance uplift while coming only one year after LNC seems like total BS

 

[Exist50]

also cougar cove having such a big performance uplift while coming only one year after LNC seems like total BS

Congrats, you have more common sense, or at least more morals, than MLID does. Of course those numbers are BS. Though this is very funny in context.

 

[Henry swagger]

View attachment 81284

I know it's MLID so most of the stuff is probably wrong but it's still interesting.
I guess these "Extra big cores" are the rumored Royal Core.
also cougar cove having such a big performance uplift while coming only one year after LNC seems like total BS

Tsmc 3nm must be hitting 7ghz 😁

 

[Glo.]

Congrats, you have more common sense, or at least more morals, than MLID does. Of course those numbers are BS. Though this is very funny in context.

Its quite possibly his worst "leak" yet.

 

[Exist50]

Its quite possibly his worst "leak" yet.

I wouldn't say that, not that it's a high bar or anything. It's his typical 95%+ complete garbage, but there are a few kernels of truth. I'm amused that he actually found the Beast Lake name, for example.

 

[igor_kavinski]

I'm amused that he actually found the Beast Lake name, for example.

Which 5% is the truth? Please do tell!

 

[Glo.]

I wouldn't say that, not that it's a high bar or anything. It's his typical 95%+ complete garbage, but there are a few kernels of truth. I'm amused that he actually found the Beast Lake name, for example.

I was talking about the main topic: Arrow Lake.

 

[dullard]

I was talking about the main topic: Arrow Lake.

Which parts are off the mark in your view:
* His very high confidence rumor that ARL stands for Arrow Lake
* His high confidence rumors (Q4 2024, 8+32, 30%-40% ST over Raptor Lake, 40% over Meteor Lake in MT)
* His lower confidence rumors in white?

 

[DrMrLordX]

Which 5% is the truth? Please do tell!

"And", "the"

 

[Exist50]

Which 5% is the truth? Please do tell!

I have no desire to play the leak game. But the Beast Lake name is real. That detail, at least, I will confirm. He's also got something with the ARL SKUs, but doesn't appear to have any idea how to interpret the information he's given. Though of course, a heaping helping of BS throughout.

 

[Geddagod]

Lmao when videocardz quotes that MLID video, Raichu said a lot of it was wrong. When videocardz replied asking specifically which part, Raichu basically said everything.
Question- when companies scrap previous architectures, do they start work on a new architecture or do they grab a simultaneous architecture in the works? Do companies like Intel develop multiple architectures during the path finding phase and then run them through sims to find the most feasible one?
Because a similar thing with Panther Cove happened with Ocean Cove, where they canned the architecture and started work on a new one. And if that precedent with Ocean Cove continues, the new 'cove', cougar cove, would look to be just a minor update of lion cove, maybe on a new node.

 

[A///]

I guess this is how bulldozer felt.

???????????????

I listen to some of his video rants every now and again when I'm half in the bag. I think he's very delightful to listen to and I appreciate it when young people, i assume he's young, talk about hardware in such great detail and enthusiasm. he's quite the overclocker too. i don't recall him ever being so sad about intel like you have. there needs to be more bulldozers on youtube and less mlids or similar freaks of nature. honest goodness discussion of hardware.

 

[A///]

Lmao when videocardz quotes that MLID video, Raichu said a lot of it was wrong. When videocardz replied asking specifically which part, Raichu basically said everything.
Question- when companies scrap previous architectures, do they start work on a new architecture or do they grab a simultaneous architecture in the works? Do companies like Intel develop multiple architectures during the path finding phase and then run them through sims to find the most feasible one?
Because a similar thing with Panther Cove happened with Ocean Cove, where they canned the architecture and started work on a new one. And if that precedent with Ocean Cove continues, the new 'cove', cougar cove, would look to be just a minor update of lion cove, maybe on a new node.

It depends how badly you've managed to f*** up the design and whether anything is salvageable. This was a regular occurrance at one company i worked at because there was no good communication and good ground work laid out. always a rush to get something out the door. they've historically had many lay offs due to poor performance periods. i was victim to one of those many years ago. i hold no grudge, karmas doing them right in. not x86 less complexity