Discussion Intel current and future Lakes & Rapids thread

[DrMrLordX]

And if they did, TSMC could not provide the volume.

TSMC is allegedly providing 40 kwpm of N3 to Intel. Meanwhile, Intel will (supposedly) have half that capacity for Intel 4. Intel 3 is an unknown, but it is likely that any Intel 3 wafers will come at the expense of Intel 4 wafers.

 

[Exist50]

Wasn't there an internal plan at Intel to make Meteor Lake 100% N3 until management pushed an Intel 4-based design instead? It's not like they couldn't revert to that.

That was more of a proposal than a plan, and long ago. They're way too far down the current path to realistically change course for Meteor Lake now. Though for Arrow Lake, I wouldn't be surprised to see them dual sourcing N3 and Intel 20A for compute dies.

 

[Henry swagger]

Meteor lake and arrow look so exciting. Wonder how powerful is the ai chip in meteor lake is ? 🤔

 

[Doug S]

But TSMC N3 is so bad TSMC already announced a replacement for it, some time down the line.

According to whom?

How is N3E any different than N5P or N4, or N7+ or N6? They always refine/improve processes, which "replaces" original process for new designs.

 

[Markfw]

Meteor lake and arrow look so exciting. Wonder how powerful is the ai chip in meteor lake is ? 🤔

Why ?

 

[IntelUser2000]

Putting it on 20A would explain the delay from late 2023 to 2024.

Is that really a delay for Arrowlake?

I always questioned the H1 launch for Meteorlake, and with them basically saying it's not taped out yet, Q2 2023 is basically the earliest we can hope for.

Even if we assume say April for MTL, that's an awful short time until Arrowlake. Perhaps the expectations were wrong.

 

[dullard]

Why ?

Did you mean: (A) Why would someone be excited by Meteor Lake and Arrow Lake? Or (B) Why would someone wonder how powerful the AI capabilities are?

For (A), Intel is claiming decent performance per watt gains between now and then. Taking Intel's slides as being true, Intel 7 -> Intel 4 is 20% gain, Intel 4 -> Intel 3 is 18%, and Intel 3 to Intel 20A is 15% gains in performance per watt. That would be a 62% improvement in a couple years. Of course, always take manufacturer predictions with a grain of salt, but that seems to be a significant gain. Also having the rumored 32 E cores with Arrow Lake will end a lot of the Alder Lake performance issues with multithreading and issues with foreground applications. Finally, ARC Battlemage graphics should be the first really game playable integrated graphics for many people.

For (B), AI could have some nice applications if it is powerful enough. I have well over 100k photos. Searching for the right photo can be time consuming even with a decent organization scheme. Instead with decent AI, I could just asking the computer with speech to show me pictures of me at age 5 holding the fish. Security cameras could tell the difference between a pet and a burglar. Email spam filters could work properly instead of the constant whack-a-mole game we play. Real time language translation could be no longer hit-and-miss. I could take a few photos of food I like and ask the computer to give me a new recipe that includes ingredients that I like. AI could take pressure off the currently limited GPUs. Etc. And I'm not even very creative in what AI could do. But it all depends on how powerful it is.

For you in particular, I would think you'd be interested in AI for protein folding. It seems far more accurate than other methods: https://www.nature.com/articles/d41586-020-03348-4

 

[nicalandia]

Intel Sierra Forest will continue the Gracemont Performance/Area dominance over any currently known CPU uArchs. With up to 128C/128T

 

[Joe NYC]

According to whom?

How is N3E any different than N5P or N4, or N7+ or N6? They always refine/improve processes, which "replaces" original process for new designs.

My (simplistic) understanding is that N3 radically increases the number of processing steps, substantially increasing the time between wafer in and out. N3E brings the number of these steps back to sane levels.

 

[Joe NYC]

Oh God... I have to admit, I have JUST watched the rearview mirror video itself from Pat Gelsinger right now.

This is nothing short of terrifying if you hold Intel shares, and I also think, he's gonna do more harm to the company in the next 5 years than the last 3 CEOs together.

Edit: he's gonna achieve that in the next 5 years, DESPITE the fact that technologically Intel will be in a 2 to 3 times better position the whole time than before.

Will NVidia buy Intel to end the misery? For NVidia to buy Intel in all stock transaction, it would be similar market cap ratio as AMD buying Xilinx.

 

[dullard]

Is that really a delay for Arrowlake?

I always questioned the H1 launch for Meteorlake, and with them basically saying it's not taped out yet, Q2 2023 is basically the earliest we can hope for.

Even if we assume say April for MTL, that's an awful short time until Arrowlake. Perhaps the expectations were wrong.

Delay might be the wrong word, but I don't know of a better word. Expectations in rumors can always be wrong, and this might be one case of that. Here is one example where the rumor was Q2 2023 for Meteor Lake and Q4 2023 for Arrow Lake (meaning the rumors were for a very short reign for Meteor Lake):

Intel Next-Gen Arrow Lake, Lunar Lake, Nova Lake CPUs Rumored To Succeed Meteor Lake

Intel's next-gen CPUs succeeding its Meteor Lake 2023 lineup have allegedly leaked out which include Arrow Lake, Lunar Lake & Nova Lake.

wccftech.com

The rumors could have been wrong instead of Arrow Lake having a delay.

 

[Joe NYC]

Intel Sierra Forest will continue the Gracemont Performance/Area dominance over any currently known CPU uArchs. With up to 128C/128T

View attachment 57617

Is this slide from some old presentation? It seems out of date.

And BTW, it will not "continue" performance / area dominance. It is taking a 2 year nap to 2024.

 

[Joe NYC]

Why ?

I think the main reason why Meteor lake looks exciting is that it is Intel's first serious entry in the chiplet era, and Intel's take of how the chiplet based designs can be architected.

I am curious about the chiplet that Intel labels SOC. What it has, what kind of functionality.

It could house the various AI components. But it could also house System Level Cache, which would be quite exciting.

 

[nicalandia]

Is this slide from some old presentation? It seems out of date.

And BTW, it will not "continue" performance / area dominance. It is taking a 2 year nap to 2024.

Well Alder Lake has been released already so is currently the Performance/Area Champion. That is what Intel intends, on the Data Center where niche instructions are not really need it but Multi-Threaded Grunt is favorable.

 

[IntelUser2000]

Intel Sierra Forest will continue the Gracemont Performance/Area dominance over any currently known CPU uArchs. With up to 128C/128T

View attachment 57617

That's Granite Rapids not Sierra Forest, and also from an older roadmap.

@Joe NYC Was Sierra Forest ever planned for 2022? Early rumors had it for end of 2023.

It won't be using Gracemont cores anymore but "next"mont.

Apparently it was originally to be on TSMC N3. This actually bodes positive for Intel's future nodes.

Here is one example where the rumor was Q2 2023 for Meteor Lake and Q4 2023 for Arrow Lake (meaning the rumors were for a very short reign for Meteor Lake):

That rumor seems to contain half truths. Anyways 2 quarter gap is pretty unrealistic.

 

[Joe NYC]

Well Alder Lake has been released already so is currently the Performance/Area Champion. That is what Intel intends, on the Data Center where niche instructions are not really need it but Multi-Threaded Grunt is favorable.

What I meant was that the slide you posted has been superseded in yesterday presentation by this slide:


Which splits the server roadmap, pushes Granite Rapids and Sierra Forrest back to wait for Intel 3 process.

 

[moinmoin]

Performance/Area Champion

The highest possible performance with the biggest possible area champion?

 

[Joe NYC]

@Joe NYC Was Sierra Forest ever planned for 2022? Early rumors had it for end of 2023.

It won't be using Gracemont cores anymore but "next"mont.

Apparently it was originally to be on TSMC N3.

I don't recall the first mentions of Sierra Forrest.

What I don't understand is if the "Mont" architecture is already so performance / area efficient, then why not release it ASAP on Intel 7? What's the point of waiting for Intel 3 or TSMC N3?

Why give AMD 2 years to lap the "Mont" cores with Zen 4c? Something is not adding up here...

 

[dullard]

Few miscellaneous things I've noticed.
...
Between this depiction of Arrow Lake as identical to Meteor Lake.
...
- it looks like they're using actual products as at least the basis for these images.

Looking at the Arrow Lake depiction, it shows 16 x 21 = 336 blocks for the GPU tile. If we assume each block is an execution unit, then that matches fairly well with the rumored 320 execution units for Arrow Lake. That gives them the possibility for a few bad EUs.

But my main question is about the CPU images. That looks like 4 P cores (brown) and 4 E cores (blue) which seems like an awfully low core count for Arrow Lake. Is it instead 4 P cores (blue) and 4 blocks of E cores (brown)? If so, then the E core blocks they seem too large in comparison to the P cores. Are they adding something to the E core die size? Or are they having more than just 32 E cores? I'm just speculating here, the image could be non-representative of the actual products.

 

[nicalandia]

The highest possible performance with the biggest possible area champion?

Yes, at least with available benchmarks(SPEC2006 1T - 453 povray) compared to Apple Firestorm/Icestorm, Intel Golden Cove, Zen3

 

[Joe NYC]

Looking at the Arrow Lake depiction, it shows 16 x 21 = 336 blocks for the GPU tile. If we assume each block is an execution unit, then that matches fairly well with the rumored 320 execution units for Arrow Lake. That gives them the possibility for a few bad EUs.

But my main question is about the CPU images. That looks like 4 P cores (brown) and 4 E cores (blue) which seems like an awfully low core count for Arrow Lake. Is it instead 4 P cores (blue) and 4 blocks of E cores (brown)? If so, then the E core blocks they seem too large in comparison to the P cores. Are they adding something to the E core die size? Or are they having more than just 32 E cores? I'm just speculating here, the image could be non-representative of the actual products.
View attachment 57620

The point of the replaceable tile architecture is that you can replace this tile with another, higher or lower performance tile, with more or less of each of the core.

 

[dullard]

The point of the replaceable tile architecture is that you can replace this tile with another, higher or lower performance tile, with more or less of each of the core.

Yes, that is easy to understand. But what are they trying to represent with this specific tile? is it Arrow Lake-U where they increased P core count and lowered E core count from Alder Lake-U? Or something else?

 

[Joe NYC]

Yes, that is easy to understand. But what are they trying to represent with this specific tile? is it Arrow Lake-U where they increased P core count and lowered E core count from Alder Lake-U? Or something else?

Maybe they just picked that configuration because it looks pretty. I would not be reading too much into it.

What I would be curious about is how the next AMD mobile APU will be structured. Some rumors say that it will be chiplet based,

If it is not, the Intel could actually lap AMD, and offer multiple chiplet combinations to target markets from lowest to highest. And Meteor Lake is, by most estimates, coming in H2 2023.

AMD typically launches APUs in H1, so it will be interesting competition, if Intel is able to ship Meteor Lake on time.

 

[uzzi38]

Apparently it was originally to be on TSMC N3. This actually bodes positive for Intel's future nodes.

Or that N3 is craptastic, like I've been saying for a bit now.

Intel are clearly trying to shift products away from TSMC. Arrow Lake was previously rumoured to be only N3. Now it's looking like N3 is for gfx only. Meteor Lake still probably uses N3 for gfx. Sierra Forrest is now pushed back a year and on i3 instead.

Apple are skipping the base N3 for N3E. And AMD likely won't be using base N3 either, but I don't know about Strix Point's node. It's not a good node.

 

[Doug S]

Apple are skipping the base N3 for N3E. And AMD likely won't be using base N3 either, but I don't know about Strix Point's node. It's not a good node.

There's no evidence of that. Apple is using N4 for A16 because N3 won't be ready in time. N3E won't be ready in time for A17, so they can either use N3 or do something they've never done and stay with the same process for two straight years with N4.