Discussion Intel current and future Lakes & Rapids thread

[Geddagod]

https://twitter.com/x/status/1687624549565857792

Between Alder Lake and Raptor Lake max turbo. i7 Raptor Lake goes up to 5.2 Ghz and i7 Alder Lake up to 4.8 Ghz.

I kid you not, I was just typing this out.
I heard RWC is heavily synthesized, which might actually be good news abt the health of Intel 4- laying out by hand usually allows for better frequency, albeit it's way more work. Even with that, it's ST Freq is higher than Intel 7 initially.
I really do hope ARL-U is buffed RWC rather than just the old MTL dies. With some 'RPC' level work, they might be able to add some extra perf, and I'm curious how good Intel's Phydes team is.

 

[A///]

albeit it's way more work.

understatement of the century there gedda.

If it wasn't for you I'd not have figured he was talking about 14th gen meteor lake.

 

[Exist50]

I kid you not, I was just typing this out.
I heard RWC is heavily synthesized, which might actually be good news abt the health of Intel 4- laying out by hand usually allows for better frequency, albeit it's way more work. Even with that, it's ST Freq is higher than Intel 7 initially.
I really do hope ARL-U is buffed RWC rather than just the old MTL dies. With some 'RPC' level work, they might be able to add some extra perf, and I'm curious how good Intel's Phydes team is.

I don't think RWC has notably different design methodology from GLC/RPC. LNC is where they should enter the 21st century.

 

[Geddagod]

I don't think RWC has notably different design methodology from GLC/RPC. LNC is where they should enter the 21st century.

Srry, went back and checked, ye RWC is apparently the last 'heavy' hand layout core, not the core they started the majority of automatic synthesis on, though apparently they started moving away from heavy hand layouts by GLC (from looking at the die shot of the core).
Honestly rumors of LNC being underperforming isn't exactly the best sign of the transition to standard EDA tools though....

 

[A///]

Srry, went back and checked, ye RWC is apparently the last 'heavy' hand layout core, not the core they started the majority of automatic synthesis on, though apparently they started moving away from heavy hand layouts by GLC (from looking at the die shot of the core).
Honestly rumors of LNC being underperforming isn't exactly the best sign of the transition to standard EDA tools though....

I've had a few glasses of wine this evening but am I reading this correct? Up until now Intel was still handlining circuits on their processors? source?

 

[Geddagod]

I've had a few glasses of wine this evening but am I reading this correct? Up until now Intel was still handlining circuits on their processors? source?

Literally just people on discord lol. Apparently a lot of the macro layout is by hand though.

 

[A///]

Literally just people on discord lol. Apparently a lot of the macro layout is by hand though.

I don't know what that is, but thanks for the heads up. I remember a lot of huah being mentioned when dr su said amd had been using ai for tracing circuits and thought that was strange because hand work had largely been a privilege smaller companies were still doing or those with larger budgets and time. It's not always perfect and some manual cleanup needs to be done but it's more efficient than a human doing it. there's a lovely pic of old school intel engineers in the 1970s I think it is doing the handwork on the floor whilst laying down on the floor. a lot of old school craft when the way of the dinos. cheers for the info.

 

[A///]

oh nm @Geddagod it's that stupid chat thing you explained to me a while back. I knew I'd seen the name before.

 

[Exist50]

oh nm @Geddagod it's that stupid chat thing you explained to me a while back. I knew I'd seen the name before.

It's not worth it. Half memes, half people pretending to know what they're talking about, even worse than here, if that's possible.

 

[Geddagod]

It's not worth it. Half memes, half people pretending to know what they're talking about, even worse than here, if that's possible.

It's fun : )

 

[A///]

It's not worth it. Half memes, half people pretending to know what they're talking about, even worse than here, if that's possible.

I can hardly figure out how to set a reminder on my iphone and you think I can understand that thing. mark tried to explain it to me about 4x before giving up and a video didn't help me understand much. much of the internet is too much for older people like you or myself. used to be simpler and good back in the day

It's fun : )

so was lsd in the 80s and early 90s.

 

[A///]

none of this explains what's so amazing about the meteorlake speeds. throw us a bone here.

 

[Exist50]

mark tried to explain it to me about 4x before giving up and a video didn't help me understand much

You'd probably be better off putting him (and some others) on mute. Best not to feed the trolls. But regarding Discord, I don't think you'd like the forum equivalents there much. Way too fast paced and meme heavy. Gives me a headache.

none of this explains what's so amazing about the meteorlake speeds. throw us a bone here.

I don't think there is anything amazing about them? If anything, low 5GHz range is kinda disappointing. A better question would be why some people got it into their heads that Intel 4 was destined to crater clock speeds even for mobile. I really hope no one was actually taking bepo seriously...




You cannot call out members who are not even participating in the thread, much less call them trolls.
You really aren't paying attention to what we are telling you.


esquared
Anandtech Forum Director

 

[SpudLobby]

https://twitter.com/x/status/1687624549565857792

Between Alder Lake and Raptor Lake max turbo. i7 Raptor Lake goes up to 5.2 Ghz and i7 Alder Lake up to 4.8 Ghz.

Honestly, this is decent. It's not good in the sense of matching the previous generation but given Intel's foundry setup where this essentially *is* the major pipe cleaning die, and we don't expect the initial node improvements to apply over the entire frequency curve particularly the top end that Intel (or AMD) are so reliant on, this is decent, and for mobile parts, this isn't really even a downgrade. Ideally, we see energy/power improvement in the modest clock ranges which the PL1 figures leaked seem to suggest.

Part of my calculus here is relative because people were too cynical about Meteor Lake frequencies and Intel 4, but even aside from that I think this is about a dead-on "B- to B+" outcome for anyone who had a realistic expectation about Intel 4. A grade would've been the same frequencies or a bit higher but actually having the part commercially ready in Q1/Q2 which diehard Intel fans assured me was the case (lol). MTL has other issues of course but I am primarily talking about the part's availability and CPU tile performance as it relates to Intel 4.

 

[Thunder 57]

Honestly, this is decent. It's not good in the sense of matching the previous generation but given Intel's foundry setup where this essentially *is* the major pipe cleaning die, and we don't expect the initial node improvements to apply over the entire frequency curve particularly the top end that Intel (or AMD) are so reliant on, this is decent, and for mobile parts, this isn't really even a downgrade. Ideally, we see energy/power improvement in the modest clock ranges which the PL1 figures leaked seem to suggest.

Part of my calculus here is relative because people were too cynical about Meteor Lake frequencies and Intel 4, but even aside from that I think this is about a dead-on "B- to B+" outcome for anyone who had a realistic expectation about Intel 4. A grade would've been the same frequencies or a bit higher but actually having the part commercially ready in Q1/Q2 which diehard Intel fans assured me was the case (lol). MTL has other issues of course but I am primarily talking about the part's availability and CPU tile performance as it relates to Intel 4.

I expect MTL to be better recieved than ARL. Pat has even said that MTL will be their next "Centrino" moment.

 

[SpudLobby]

I expect MTL to be better recieved than ARL. Pat has even said that MTL will be their next "Centrino" moment.

Gelsinger said that because of AI though, and while I think the VPU will be useful I suspect it's still not remotely the main focal point of concern, but yeah it could be received for other reasons. I think the bar is lower though coming off RPL/ADL to MTL so it depends on how well they execute on idle and device runtimes partially.

MTL has some other problems though per Exist50 on the tile structure so IDK. I think MTL and ARL will get them into business for mobile again which should be the goal, but LNL will be their real moment to shine on something AMD just won't have.

 

[eek2121]

You'd probably be better off putting him (and some others) on mute. Best not to feed the trolls. But regarding Discord, I don't think you'd like the forum equivalents there much. Way too fast paced and meme heavy. Gives me a headache.

I don't think there is anything amazing about them? If anything, low 5GHz range is kinda disappointing. A better question would be why some people got it into their heads that Intel 4 was destined to crater clock speeds even for mobile. I really hope no one was actually taking bepo seriously...

Honestly, this is decent. It's not good in the sense of matching the previous generation but given Intel's foundry setup where this essentially *is* the major pipe cleaning die, and we don't expect the initial node improvements to apply over the entire frequency curve particularly the top end that Intel (or AMD) are so reliant on, this is decent, and for mobile parts, this isn't really even a downgrade. Ideally, we see energy/power improvement in the modest clock ranges which the PL1 figures leaked seem to suggest.

Part of my calculus here is relative because people were too cynical about Meteor Lake frequencies and Intel 4, but even aside from that I think this is about a dead-on "B- to B+" outcome for anyone who had a realistic expectation about Intel 4. A grade would've been the same frequencies or a bit higher but actually having the part commercially ready in Q1/Q2 which diehard Intel fans assured me was the case (lol). MTL has other issues of course but I am primarily talking about the part's availability and CPU tile performance as it relates to Intel 4.

You guys should stop comparing this chip to desktop. It is a mobile “P” die that performs well enough that Intel is using it for multiple segments. The top Intel mobile chips have a 5.4 ghz clock and most are less. Same with AMD. At least one SKU looks to absolutely crush the 7940hs in terms of performance. No idea how it will perform against the 7945hx, but that is a 55-75W chip, not something a “P” chip would compete with.

 

[SpudLobby]

You guys should stop comparing this chip to desktop. It is a mobile “P” die that performs well enough that Intel is using it for multiple segments. The top Intel mobile chips have a 5.4 ghz clock and most are less. Same with AMD. At least one SKU looks to absolutely crush the 7940hs in terms of performance. No idea how it will perform against the 7945hx, but that is a 55-75W chip, not something a “P” chip would compete with.

Where do you see me comparing it to desktop chips? Don't include me in this sect, I've consistently said on this site:
A) I don't care - or care very, very little - about desktops.
B) MTL peak clocks aren't as big of a deal as is idle power draw or power draw at modest clock ranges e.g. PL1.
C) I don't think I've ever even brought up the 7945HX which is a POS desktop chip with a crappy IO die.

That said, if you actually read, you'll note the reason people were antsy about Intel 4's clocks is partially about the fact that IPC boosts aren't really happening with RWC in MTL, so regressions in clocks might affect the maximum ST even on mobile, and everyone has PTSD from Intel 10NM both in terms of yields and clocks etc. Personally, I wouldn't care even if it didn't reach 5GHz, but this will temper some of the insane "15-25W ST in a mobile product is ok" crowd.

At any rate, I think MTL despite problems reported (Exist50 can detail) has a real opportunity to take Intel back to "decent" and "a real alternative" for mobile products if they improve the average use case efficiency and perf which Pl1 improvements suggest. There's also the iGPU which is reportedly great and on N5 (which means the power draw and energy efficiency should also be solid), and the fact that the legendary RDNA3 iGPU in Phoenix was both not remotely as wide/big as the rumors predicted, and then AMD wet the bed on the physical design. Meaning Intel's MTL iGPU has a real shot at being at least as good as Rembrandt's and possibly with better low power operation than Phoenix.


How Intel does vs the 7945HX junk is not irrelevant of course, but IMO what matters more is Intel being able to tide themselves over with the most competitive product since Ice Lake/TGL in the 15-28W range where their idle and crappy energy efficiency just ruin current products. They need something to last until Lunar Lake and ARL.

 

[A///]

Centrino....... that's a head scratcher for a reference. I hope gunslinger was referencing that centrino began on pentium m and pentium m was very good as I remember it being back in the day.

either mtl is a bit of a dark horse no one knows much about as a final product or it's all hooey and pat here is polishing knobs for no reason other than to sell investors the idea intel's making major progress.

 

[mikk]

I don't think there is anything amazing about them? If anything, low 5GHz range is kinda disappointing. A better question would be why some people got it into their heads that Intel 4 was destined to crater clock speeds even for mobile. I really hope no one was actually taking bepo seriously...

This is the highest clock speed for an Intel non + node ever. Compare it to Ivy Bridge 22nm mobile, Broadwell 14nm mobile or Icelake 10+ mobile. It bodes well for Intel 3 because Intel can usually squeeze-out much more for the first + refresh of a new node. Not that long ago people compared Intel Meteor Lake to Cannonlake. 100-200 Mhz behind Raptor Lake 7 Ultra/10ESF+ and Phoenix TSMC N5 is a respectable result.

 

[A///]

It was the chinese leakers that said mtl was suffering frequency regressions that would make it a failed processor or that was arrowlake. may have been both. it's very hard to keep track of all these little changes. the bastardised English of these leakers doesn't help either.

 

[mikk]

It was the chinese leakers that said mtl was suffering frequency regressions that would make it a failed processor or that was arrowlake. may have been both. it's very hard to keep track of all these little changes. the bastardised English of these leakers doesn't help either.

Just 4 months ago some people in this forum claimed Intel 4 is in almost the same shape as 10nm back in 2016

I'm saying Intel 4 is in almost the same shape as 10nm back in 2016. Intel 3 is a derivative node. It took Intel maybe two years to get 10nm+ out the door in commercial products. Even if Intel 4 shows up in good order by the end of 2023 in Meteor Lake, it's completely unreasonable to expect Intel 3 to show up less than 6 months later in viable quantities. Unless we believe "it's not the process it's the design" is the reason for every Intel 4 product to be scotched up to this point, which seems ridiculous.

Maybe Intel can set wafers on fire for huge losses and push out a few Sierra Forest units to key customers by March 2024 (end of Q1). Anything more than that? Nah, not buying it.

 

[SpudLobby]

This is the highest clock speed for an Intel non + node ever. Compare it to Ivy Bridge 22nm mobile, Broadwell 14nm mobile or Icelake 10+ mobile. It bodes well for Intel 3 because Intel can usually squeeze-out much more for the first + refresh of a new node. Not that long ago people compared Intel Meteor Lake to Cannonlake. 100-200 Mhz behind Raptor Lake 7 Ultra/10ESF+ and Phoenix TSMC N5 is a respectable result.

To be fair this is totally expected, e.g. you'd expect newer nodes to improve, but in practice the bar you want to measure RE: improvement is relative to the products on the first iteration, which I think is where you're also getting to with the non + mention or pointing out that, in fact, Meteor Lake and Intel 4 don't look like Cannon Lake and Intel 10NM. Which, in practice Cannon Lake was hardly a real product, Ice Lake is the right comparison, and I don't think Meteor Lake and Intel 4 look like Ice Lake and Intel 10NM so far at all. Frankly, the node is just leaps and bounds ahead of 10NM/7 and I think in the long term (1-5 years from now) the cost structure of Intel 4/3 will be much better than Intel 10NM ever was in part thanks to the shrink and EUV, or getting HD libraries right (hopefully) and not screwing around with Cobalt^1.


I4 does have a Cobalt-coated Copper metal/interconnect layer but the performance and electrostatics seem to be better than with Copper or Cobalt alone per Intel so.

 

[A///]

Just 4 months ago some people in this forum claimed Intel 4 is in almost the same shape as 10nm back in 2016

tbf @DrMrLordX is a pessimist of both amd and intel. he and I argue pages when it comes to a zen launch because how little or slow AMD is in forming a product to go up against intel, and you'd think intel's funeral is in a week with how gravely he feels about Intel now.

 

[mikk]

On bilibili there is one user who is telling Redwood Cove improves IPC by 8-9% in a mixed Int+FP workload over Golden Cove. Another one said according to Dell MTL IPC impoves by about 10%.