Discussion Intel current and future Lakes & Rapids thread

[jpiniero]

Yeah, I think MTL-S coming later than -P/-M was expected for the most part. More importantly though, is the Intel Evo specification for Meteor Lake being listed for 1H 23. IMO that's probably when we can expect them (-P/-M) to launch.

That would be pretty aggressive timeline unless Raptor Lake is only for S.

 

[uzzi38]

That would be pretty aggressive timeline unless Raptor Lake is only for S.

I'm expecting another CML/ICL type split at first tbh.

 

[DrMrLordX]

So really the only reason we have Rocketlake-S rather than Tigerlake-S is because 2021 is year of 10nm and it just started ramping and they had to prioritize on mobile and server?

Intel produces Ice Lake-SP on 10nm+, not 10SF which is used for TigerLake or 10SFE which is used for Alder Lake. Can they switch up fab capacity between 10nm+ and 10SF/SFE effortlessly? I'm not sure that the two processes compete for fab capacity in that fashion.

Tigerlake-S might have been only few % faster per clock but would have ended up being far lower power. So Rocketlake offers at most 10% more performance at much higher power.

It would have been a much stronger product, but we still don't know if Intel could yield enough 8c dice on 10SF to meet the needs of the market.

If you believe certain people here, Intel apparently only gets 1-2 chips per wafer, because they can’t possibly have good yields. 🤣

Amusing exagerration, but I don't think Intel's yields were even that bad for Cannonlake. Fact is that Intel is still playing with their cards close to their chest wrt yields on 10nm+, 10SF, and 10SFE. I've read that they've shipped 40 million TigerLake 2c/4c so far, but that exists outside of any context. How many wafers did it take to accomplish that feat? I think it is clear that 10nm+ still isn't yielding well since Intel has allegeldy shipped only ~100k Ice Lake-SP units.

That would be pretty aggressive timeline unless Raptor Lake is only for S.

May/June 2023 isn't that ambitious unless you take into account that 7nm has already had public announcements of delays (with unknown internal delays). Assuming we're talkingabout 1H 2023 for Meteor Lake? Meanwhile what does that leave for Intel on the desktop until then? Alder Lake-S? When do we expect Raptor Lake-S, and what exactly is Raptor Lake-S anyway?

 

[jpiniero]

When do we expect Raptor Lake-S, and what exactly is Raptor Lake-S anyway?

A year after Alder Lake-S ships.

https://twitter.com/x/status/1375466906212130817

Raichu seems to think Raptor Lake-S will have 24 cores. The most realistic combination if that's true is 8+16 so it might just be tacking on 2 small clusters and frequency/TDP tweaks from Alder Lake.

Edit: 12+12 would be more competitive marketing wise but I would think it would necessitate ditching the Ring for something else. Don't know if they want to do the work. Plus that would start to get a little big.

 

[DrMrLordX]

A year after Alder Lake-S ships.

Hmm. No dates mentioned in the screenshot but it would be very Intel-like to have an Alder Lake refresh one year later.

Raichu seems to think Raptor Lake-S will have 24 cores. The most realistic combination if that's true is 8+16 so it might just be tacking on 2 small clusters and frequency/TDP tweaks from Alder Lake.

Edit: 12+12 would be more competitive marketing wise but I would think it would necessitate ditching the Ring for something else. Don't know if they want to do the work. Plus that would start to get a little big.

Eventually Intel has to be thinking about adapting the mesh to consumer chips. The downside there is that mesh can degrade performance unless mesh/uncore speed is kept high, which in turn is bad for power. I don't know if mesh will ever work for mobile chips either. If they want the same chip layouts for -M/-P and -S CPUs then mesh is maybe a no-go.

 

[jpiniero]

It doesn't have to be the mesh. Just not the ring.

 

[IntelUser2000]

I'm expecting another CML/ICL type split at first tbh.

There was a twitter post that talks about how Raptor Lake might not come if Meteor Lake goes well or something?

So maybe it ends up being like Rocketlake.

It does look good (compared to the 5900H, for example). But I wonder about that base clock rate. Geekbench doesn't take that long to run. When both CPUs are thermally limited it seems (on paper, anyway) the 5900H would have a large lead. I look forward to someone investigating this.

Part of the reason is the 5900H is close to the top SKU for AMD while 11800H on that Geekbench comparison is quite far down.

The 11900H is at 2.5GHz for Base.

Also since Icelake the Base doesn't matter a whole lot. Icelake mobile parts are in the low 1GHz. You'll be hard pressed to have it reach that low.

 

[IntelUser2000]

Intel UHD 750 benchmark

Based on Time Spy it's 38% faster than UHD 630 but Xe architecture has better optimizations so it'll probably end up being 50-60% faster in games.

Compared to the mobile 96 EU it's at about 40-45% performance. Desktops aren't limited by thermals/power and the memory is usually better. The test system uses DDR4-4200.

As always, most sources are wrong on the TMU/ROP information for Intel GPUs. The 32 EU version should have 8 ROPs and 16 TMUs, exactly a third of 96EU version.

Overclocking from 1.3GHz to 1.5GHz results in 14% gains which means it's nearly all compute bound almost not at all memory bound.

 

[jpiniero]

There was a twitter post that talks about how Raptor Lake might not come if Meteor Lake goes well or something?

Sounds risky.

 

[repoman27]

According to this website, the largest die variant (XCC) is 640 mm2. It's got 42 cores in total, but with 2 disabled for now at the highest SKU.
Regardless, I agree that it is incredibly ambitious for 10nm+.

That's just a floorplan mockup generated by SkyJuice based on leaked specs with the die size being extrapolated from photos of the LCC die.

I was counting the cores/dies I could actually see in the wafer photo. I went back and looked again really closely, and I can't for the life of me make out those two cores in the top row. I think it's actually a 40C design, and Intel may be selling fully enabled parts. Does anyone else see 42 cores in those photos?

My size estimate was just a ballpark based on how the dies fit on a 300 mm wafer, but it's probably pretty close. I wasn't counting pixels or anything, seeing as the image only shows a partial wafer at an oblique angle.

 

[jpiniero]

I thought I saw the SKU list of Ice Lake Server and indeed Plat goes up to 40 cores. I think the absolute max you will actually be able to buy is 32 though. Intel could always do things like price it way higher than Milan and then play the discount game based upon how many chips they actually yield.

 

[lobz]

If you believe certain people here, Intel apparently only gets 1-2 chips per wafer, because they can’t possibly have good yields. 🤣

A really provocative and deliberately misleading exaggeration of people's words that is somehow supposed to be funny. Although I did not laugh, here's hoping at least some did, so you're not outing yourself as a fool for nothing, because I don't wish ill for anyone.

 

[Asterox]

Intel UHD 750 benchmark

Based on Time Spy it's 38% faster than UHD 630 but Xe architecture has better optimizations so it'll probably end up being 50-60% faster in games.

Compared to the mobile 96 EU it's at about 40-45% performance. Desktops aren't limited by thermals/power and the memory is usually better. The test system uses DDR4-4200.

As always, most sources are wrong on the TMU/ROP information for Intel GPUs. The 32 EU version should have 8 ROPs and 16 TMUs, exactly a third of 96EU version.

Overclocking from 1.3GHz to 1.5GHz results in 14% gains which means it's nearly all compute bound almost not at all memory bound.

If we look only at Time Spy scores, 4200mhz ddr4+UHD750 iGPU comparison vs stock 1900mhz Vega 7 CU+3800mhz ddr4 memory.

stock iGPU 1300mhz, 664
iGPU 1500mhz, 758

AMD Ryzen 5 PRO 4650G Renoir APU review

There is a new hybrid processor in town, meet the AMD Ryzen 5 PRO 4650G. The uniqueness of this event is that the processor series announced and released a few weeks ago should not have got into the ... Test Results Time Spy

www.guru3d.com

 

[SAAA]

A year after Alder Lake-S ships.

https://twitter.com/x/status/1375466906212130817

Raichu seems to think Raptor Lake-S will have 24 cores. The most realistic combination if that's true is 8+16 so it might just be tacking on 2 small clusters and frequency/TDP tweaks from Alder Lake.

Edit: 12+12 would be more competitive marketing wise but I would think it would necessitate ditching the Ring for something else. Don't know if they want to do the work. Plus that would start to get a little big.

I don't think Intel will bother adding cores to Raptor Lake at all, as you point out maybe it will be a frequency jump, similar to Skylake -> Kabylake keeping the die pretty much identical.

There are rumours of some cache or memory change to favour gaming? Not too sure how to take those, I think frequency is better honestly: if they decrease all latencies and eat some wattage into the much lower power consumption Alder will have vs Rocket lake that could benefit all workloads, not just "gaming".

There was a twitter post that talks about how Raptor Lake might not come if Meteor Lake goes well or something?

So maybe it ends up being like Rocketlake.

The chance they might cancel it makes me more confident they won't put too many resources here, cores other than small clusters would quickly enlarge dies and the only reasonable step-up would be 10 or 12, a not indifferent area increase. But I think they reserve that for 7nm, maybe dual die with 8+8 big/small cores each, also a new architecture.

 

[coercitiv]

There are rumours of some cache or memory change to favour gaming? Not too sure how to take those, I think frequency is better honestly

If the slide was true, Intel mentioned two things about Raptor Lake:

• CPU core changes for improved performance
• improved CPU cache for gaming

 

[jpiniero]

I don't think Intel will bother adding cores to Raptor Lake at all, as you point out maybe it will be a frequency jump, similar to Skylake -> Kabylake keeping the die pretty much identical.

I could see marketing insisting on adding cores, especially if the die size addition of the small cores isn't much. Maybe 8 small cores in one cluster is possible.

improved CPU cache for gaming

Higher uncore speeds possibly?

 

[mikk]

Intel UHD 750 benchmark

Based on Time Spy it's 38% faster than UHD 630 but Xe architecture has better optimizations so it'll probably end up being 50-60% faster in games.

Compared to the mobile 96 EU it's at about 40-45% performance. Desktops aren't limited by thermals/power and the memory is usually better. The test system uses DDR4-4200.

As always, most sources are wrong on the TMU/ROP information for Intel GPUs. The 32 EU version should have 8 ROPs and 16 TMUs, exactly a third of 96EU version.

Overclocking from 1.3GHz to 1.5GHz results in 14% gains which means it's nearly all compute bound almost not at all memory bound.

Interestingly the improvement in Firestrike GPU score is a lot bigger, it's an 64% uplift there.

 

[eek2121]

A really provocative and deliberately misleading exaggeration of people's words that is somehow supposed to be funny. Although I did not laugh, here's hoping at least some did, so you're not outing yourself as a fool for nothing, because I don't wish ill for anyone.

Lighten up man, this is an internet forum... SMH

 

[IntelUser2000]

If the slide was true, Intel mentioned two things about Raptor Lake:

• CPU core changes for improved performance
• improved CPU cache for gaming

Well, specifically "Hybrid CPU core changes for improved performance".

Interestingly the improvement in Firestrike GPU score is a lot bigger, it's an 64% uplift there.

Good catch. Firestrike might be more representative of the differences then, at least between those two GPUs.

I wish they bring something significant with "Gen 12.2" on Alderlake, because it seems like after Xe they are coasting in graphics. RDNA2 on Remembrandt is going to slaughter it even with 8 CUs.

That's been the historical Intel though. One "Conroe" and coast. One "Xe" and coast.

 

[lobz]

I am lightened up, thanks for your concern!

I still stand by my comment.

 

[Hulk]

That's been the historical Intel though. One "Conroe" and coast. One "Xe" and coast.

True. But they have never had this much pressure from competitors as they do today. They can rest on their laurels at their peril.

 

[coercitiv]

One "Xe" and coast.

Coasting with dGPU ambitions? That would be a short story indeed. They better tock after each tick.

 

[mikk]

I wish they bring something significant with "Gen 12.2" on Alderlake, because it seems like after Xe they are coasting in graphics. RDNA2 on Remembrandt is going to slaughter it even with 8 CUs.

That's been the historical Intel though. One "Conroe" and coast. One "Xe" and coast.

Iris Xe 96EU 1300-1350 Mhz is bandwidth starved, there are single->dual rank 3200 gains of 10% and LPDDR4 4266 adds another 20% in some cases. They can increase the GPU clock speed to 1.5-1.6 Ghz and with DDR5/LPDDR5 we should get a nice uplift. iGPU performance for desktop models doesn't matter really, display and media feature set is more important. Quicksync encoding quality is a lot better than on Gen9.

 

[uzzi38]

Coasting with dGPU ambitions? That would be a short story indeed. They better tock after each tick.

Considering Gen13 is only rumoured to appear alongside Meteor Lake, coasting is probably the right word. Progress is just too slow. I know that's partially down to 7nm delays, but still.

 

[IntelUser2000]

iGPU performance for desktop models doesn't matter really, display and media feature set is more important. Quicksync encoding quality is a lot better than on Gen9.

Yes, but it matters a lot on mobile. Trifecta of battery life, graphics, and CPU wins coming with Remembrandt. Alderlake may be a big advancement, but will face stiff competition at the very least.

Intel is also slow on driver updates even with Xe.

Can we have more consistent execution please? It's impressive how AMD is doing that on both areas. It feels like AMD executing well is better looking at the big picture.