Discussion Intel current and future Lakes & Rapids thread

[itsmydamnation]

But they brought into Zen 2. I guess we just have to disagree.

Zen 3 was completely redone execution and load store system. Thats like 2/3 of the core ....

 

[lightisgood]

I'm boring about ZEN topics.

 

[SpudLobby]

I don't think that's what Bionic is saying. He seems to be implying ARL-U is neither N3 nor 20A. Which could square with some previous rumors of it being essentially a MTL-U refresh on Intel 3.

Makes sense, I didn't see that RE: MTL-U refresh on Intel 3 but that's great, and pretty much what i wanted to see anyways - thought it would be a waste for Intel to *not* use Intel 3 for a client product. I3 have much better leakage, density characteristics owing to the HD libraries and apparently with it some other enhancements. Here's to hoping.

 

[SpudLobby]

Exist, could you point me to those rumors on MTL with i3?

 

[coercitiv]

Exist, could you point me to those rumors on MTL with i3?

See this for example:

Page 830 - Discussion - Intel current and future Lakes & Rapids thread

Page 830 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Only ARL-S gets the new extensions like SHA512 or AVX-VNNI-INT16 according to Intels programming reference pdf as well as the latest GCC 14 compiler release. The other ARL doesn't get the new ones which could suggest it is based on an older architecture like Redwood Cove+Crestmont.

 

[eek2121]

Makes sense, I didn't see that RE: MTL-U refresh on Intel 3 but that's great, and pretty much what i wanted to see anyways - thought it would be a waste for Intel to *not* use Intel 3 for a client product. I3 have much better leakage, density characteristics owing to the HD libraries and apparently with it some other enhancements. Here's to hoping.

I am willing to bet he is wrong. Arrow Lake compute tiles (desktop, mobile) will be fabbed on an Intel process. Everything else will come from internal/external fabs.

 

[Geddagod]

Exist, could you point me to those rumors on MTL with i3?

The original, MTL-R is ARL-U rumor was from Raichu on Twitter. Bionic just seems to be confirming it.

I will say though, I haven't seen anywhere that MTL-R will be on Intel 3. All Raichu said was that it would be on "N-1" and "based on the 1276 series". Seems a bit ambiguous, but it seems like MTL-R would be on Intel 4 rather than Intel 3. Perhaps if it is on Intel 3, however, Intel could be looking to just move their core IP from GNR/SRF over to MTL-R, with some tweaks.

 

[Geddagod]

Isn't the biggest news here "running on its first stepping"?

When's the last time that's been true for any Intel CPU? In the past they never did the level of simulation that e.g. Apple and AMD do, because when you control your fabs you can push hot lots through as often as you like, since there are no other customers to upset. So why waste all that effort on simulation when you can run on real silicon if you're willing to wait a bit? Perhaps they are changing their ways and adapting to the times.

Intel claimed they will be treating their internal design teams as external customers during their foundry seminar a month or two ago IIRC.

 

[Doug S]

I sincerely doubt they're going to be shipping on A-step. It's just what they're testing with for now. Honestly, if LNL ships on B-step, that would be an accomplishment by Intel standards.

Of course not, but reportedly Intel's first stepping often can't even boot.

I guess what is meant by "run" could mean anything between "it can boot if you halve the frequency, double the power, and make a few code tweaks to work around some bugs" to "it passes all tests and operates at the expected frequency" so maybe I was overrating what it meant.

 

[SiliconFly]

Of course not, but reportedly Intel's first stepping often can't even boot.

I guess what is meant by "run" could mean anything between "it can boot if you halve the frequency, double the power, and make a few code tweaks to work around some bugs" to "it passes all tests and operates at the expected frequency" so maybe I was overrating what it meant.

Well, early silicon tends to be just that always.

 

[SiliconFly]

This one keeps bothering me. Pat said: "Arrow Lake is currently running its first stepping in the fab".

Arrow Lake first stepping directly means A0.

If they already have A0 running, it means they also have the other corresponding ARL tiles in the cpu package too. Meaning the tGPU, SoC tile & IO tile. Does that mean, the corresponding Battlemage tGPU (early silicon) is already up & running? Kinda hard to believe.

 

[Geddagod]

Well, early silicon tends to be just that always.

AFAIK other companies (Nvidia, AMD) have much better A0 silicon than what Intel usually has. From what I heard, a problem Intel faces is 'race to tape in' where they would try getting real silicon as soon as possible and then do bug testing from there, while AMD/Nvidia try getting all the bugs out of their designs first before moving on to real silicon. SPR is prob the best example of that. Intel, by treating their design teams as an external partner, would reduce this problem by forcing the design team to have less steppings overall, starting with ARL/LNL.

This one keeps bothering me. Pat said: "Arrow Lake is currently running its first stepping in the fab".

Arrow Lake first stepping directly means A0.

If they already have A0 running, it means they also have the other corresponding ARL tiles in the cpu package too. Meaning the tGPU, SoC tile & IO tile. Does that mean, the corresponding Battlemage tGPU (early silicon) is already up & running? Kinda hard to believe.

ARL doesn't use BMG afaik. LNL does. A bit odd considering ARL should come out at roughly the same time as LNL, perhaps it's because they wanted to derisk ARL. They prob do have real silicon of BMG "up and running" tho, on the LNL compute tile.

 

[lightisgood]

Of course not, but reportedly Intel's first stepping often can't even boot.

I guess what is meant by "run" could mean anything between "it can boot if you halve the frequency, double the power, and make a few code tweaks to work around some bugs" to "it passes all tests and operates at the expected frequency" so maybe I was overrating what it meant.

Intel Says All Products Are On or Ahead of Schedule: Arrow Lake Running In Fabs, Client-Segment Recovers

Intel stated during its recent earnings that all products and programs are on or ahead of schedule with Arrow Lake already running in Fabs.

wccftech.com

> On Intel 20A, our first process node using both RibbonFET and PowerVia, Arrow Lake, our volume client product is currently running its first stepping in the fab.


Pat didn't say "in the lab" but "in the fab".
I guess that ARL A0 silicon is simply being in manufacturing process.

BTW, I have an very optimistic imagination about quality of ARL A0 silicon.
IMO, Intel 20A is sure to equip tolerant design rules (c.f. Intel 7). Furthermore, disaggregated architecture probably bring only 0.1x bugs to compute tile.

 

[Ajay]

AFAIK other companies (Nvidia, AMD) have much better A0 silicon than what Intel usually has. From what I heard, a problem Intel faces is 'race to tape in' where they would try getting real silicon as soon as possible and then do bug testing from there, while AMD/Nvidia try getting all the bugs out of their designs first before moving on to real silicon. SPR is prob the best example of that. Intel, by treating their design teams as an external partner, would reduce this problem by forcing the design team to have less steppings overall, starting with ARL/LNL.


ARL doesn't use BMG afaik. LNL does. A bit odd considering ARL should come out at roughly the same time as LNL, perhaps it's because they wanted to derisk ARL. They prob do have real silicon of BMG "up and running" tho, on the LNL compute tile.

Well, AMD, at least in their CPU division, puts whatever they can on an FPGA and runs that before going to the fab (on top of simulations). As @Doug S pointed out, for a long time Intel had the luxury of running the silicon through as often as they wanted to. They also kept tweaking things (like the famous 'G0' Q6600) - that's allot of steppings.

 

[Dayman1225]

This one keeps bothering me. Pat said: "Arrow Lake is currently running its first stepping in the fab".

Arrow Lake first stepping directly means A0.

If they already have A0 running, it means they also have the other corresponding ARL tiles in the cpu package too. Meaning the tGPU, SoC tile & IO tile. Does that mean, the corresponding Battlemage tGPU (early silicon) is already up & running? Kinda hard to believe.

Unsure about tGPU but I/O and SoC are reused from MTL so not surprising there.

 

[SiliconFly]

Unsure about tGPU but I/O and SoC are reused from MTL so not surprising there.

Oh god. If they're gonna use Alchemist yet again in ARL like @Geddagod says, it's tragic. It'll be the third iteration of alchemist (or Alchemist++). These idiots have to stop doing the +++++ thing. It's infuriating. Looks like Intel in 2024 is still disappointing.

 

[H433x0n]

Oh god. If they're gonna use Alchemist yet again in ARL like @Geddagod says, it's tragic. It'll be the third iteration of alchemist (or Alchemist++). These idiots have to stop doing the +++++ thing. It's infuriating. Looks like Intel in 2024 is still disappointing.

You're overreacting a bit. The release schedule for 2024 is still the best Intel has had in many years (at least for as long as I've been following tech). The releases of SRF, GNR & ARL will help Intel's competitiveness compared to where they are today. This will be the first time since Zen+ that they will have node parity.

 

[Khato]

Oh god. If they're gonna use Alchemist yet again in ARL like @Geddagod says, it's tragic. It'll be the third iteration of alchemist (or Alchemist++). These idiots have to stop doing the +++++ thing. It's infuriating. Looks like Intel in 2024 is still disappointing.

Well, compared to the many years of Gen9 (Skylake through Cometlake) followed by the outlier of single generation Gen11 with Icelake and then a few years again with Gen12LP (Tigerlake through Raptorlake) having two generations of integrated Alchemist doesn't sound too bad, no?

 

[Dayman1225]

Oh god. If they're gonna use Alchemist yet again in ARL like @Geddagod says, it's tragic. It'll be the third iteration of alchemist (or Alchemist++). These idiots have to stop doing the +++++ thing. It's infuriating. Looks like Intel in 2024 is still disappointing.

GitHub repo leaks point to it being Xe LPG+ so I imagine it’s some variation/beefed up alchemist

 

[ondma]

You're overreacting a bit. The release schedule for 2024 is still the best Intel has had in many years (at least for as long as I've been following tech). The releases of SRF, GNR & ARL will help Intel's competitiveness compared to where they are today. This will be the first time since Zen+ that they will have node parity.

I dont follow the server side, but on client, it looks like Intel will be the *least* competitive in the last few years if ARL rumors are true. ALD and RPL competed very well with Zen 3 and Zen 4 (all be it at higher power), but it looks like ARL will be crushed by Zen 5, sadly. Intel is back to the "wait for the next release, we will have great advances". Raptor Lake refresh, very minor gain, that is OK, wait for Arrow Lake. Oops, Arrow Lake has serious problems, Lunar Lake is for mobile, who knows when the next desktop release will be, and if it will still be gimped.

 

[ondma]

Oh god. If they're gonna use Alchemist yet again in ARL like @Geddagod says, it's tragic. It'll be the third iteration of alchemist (or Alchemist++). These idiots have to stop doing the +++++ thing. It's infuriating. Looks like Intel in 2024 is still disappointing.

Nobody ever said Arrow Lake would have Battlemage. For the desktop, who really cares? Lunar Lake will probably have it, and that is where the integrated graphics matter more. Lunar Lake is likely going to be a dud, too, though unless they can get hyperthreading or the new Register thing working. Of course, they could make up for the lack of HT by adding more e cores, but that doesnt appear to be happening either.

 

[lightisgood]

I dont follow the server side, but on client, it looks like Intel will be the *least* competitive in the last few years if ARL rumors are true. ALD and RPL competed very well with Zen 3 and Zen 4 (all be it at higher power), but it looks like ARL will be crushed by Zen 5, sadly. Intel is back to the "wait for the next release, we will have great advances". Raptor Lake refresh, very minor gain, that is OK, wait for Arrow Lake. Oops, Arrow Lake has serious problems, Lunar Lake is for mobile, who knows when the next desktop release will be, and if it will still be gimped.

Disagree.
Alder Lake and Raptor Lake were good player, however, they beat competitor at ~250W power consmption.
Without saying, 250W SKU is small and small market in the real world.

Especially, RPL relatively lacks competitiveness at ~35W power consmption, gpu computing and so on.
35W SKU is mainstream. So I consider MTL & ARL as hopeful successors.

Lastly, this is a topic that has already been discussed.
You should read old messages over again.

 

[Henry swagger]

I dont follow the server side, but on client, it looks like Intel will be the *least* competitive in the last few years if ARL rumors are true. ALD and RPL competed very well with Zen 3 and Zen 4 (all be it at higher power), but it looks like ARL will be crushed by Zen 5, sadly. Intel is back to the "wait for the next release, we will have great advances". Raptor Lake refresh, very minor gain, that is OK, wait for Arrow Lake. Oops, Arrow Lake has serious problems, Lunar Lake is for mobile, who knows when the next desktop release will be, and if it will still be gimped.

In your dreams zen 5 crush arrow lake

 

[mikk]

Oh god. If they're gonna use Alchemist yet again in ARL like @Geddagod says, it's tragic. It'll be the third iteration of alchemist (or Alchemist++). These idiots have to stop doing the +++++ thing. It's infuriating. Looks like Intel in 2024 is still disappointing.

Without the mediocre GDDR6 memory controller, PCIe overhead and rBAR struggle we will finaly see the real IPC gains compared to Xe LP. Alchemist(+) tGPU should do much better than the dGPU version unless bandwidth limitation on DDR5/LPDDR5 is a big issue.

 

[Abwx]

Disagree.
Alder Lake and Raptor Lake were good player, however, they beat competitor at ~250W power consmption.
Without saying, 250W SKU is small and small market in the real world.

At 253W RPL is 2% ahead of the competition at 142W, and is below even if pushed at 300W if the latter is at around 200W.

These were good players for whom had no knowledge of the real numbers and was relying on Intel s marketing instead, i guess that urban legends helped a lot to forge this kind of "opinion"...

Core i9-13900K, i7-13700K & i5-13600K: Gaming-Könige im Test: Leistungsaufnahme und Effizienz

Intel Raptor Lake im Test: Leistungsaufnahme und Effizienz / Leistungsaufnahme in Anwendungen ab Werk / Der Verbrauch steigt

www.computerbase.de