Discussion Intel current and future Lakes & Rapids thread

[Khato]

Intel servers has been hanging on traditional enterprise sales where they are undercutting on price to the point where they don’t make any money. AMD is over 50% in cloud and enterprise is finally starting to open up too. The next releases from Intel will certainly help them but I’m not sure it will be enough to stop AMD from taking market share.

No disagreement that Intel has been selling pretty much at cost in order to keep the fab utilization up and minimize market share loss. But keep in mind that such is in large part a result of their products up until SRF still being on the Intel 7 node. Even if Intel sells a 64 core Emerald Rapids at the same price as a 64 core Genoa that's 2x750mm^2 dies versus 8x72mm^2 compute + 400mm^2 IO. The massive die size on Intel's side is financially viable if it could be sold around the $10k suggested retail price, but that's where the 128C Bergamo and 96C Genoa reside. So instead Intel's flagship has to sell closer to the next tier down around the $5k mark. (Note that these are by no means accurate absolute numbers, but should be reasonably close in relation to one another.)

One question though is what's the source on AMD being over 50% in cloud?

 

[Hitman928]

One question though is what's the source on AMD being over 50% in cloud?

Lisa Su announced it a quarter or two ago, I believe during an earnings call. I should note, though, that AMD typically talks in terms of revenue share, though not always, so they were probably referencing revenue share but I'm not positive about that.

 

[Henry swagger]

Intel servers has been hanging on traditional enterprise sales where they are undercutting on price to the point where they don’t make any money. AMD is over 50% in cloud and enterprise is finally starting to open up too. The next releases from Intel will certainly help them but I’m not sure it will be enough to stop AMD from taking market share.

Amd will never habe 50% market share.. granite rapids will crush turin

 

[jpiniero]

The massive die size on Intel's side is financially viable if it could be sold around the $10k suggested retail price

Even during the time Intel had 90% server market share, they never sold processors anywhere near the MSRP.

They clearly aren't getting enough profit off of each server chip to fund future nodes, that's for sure.

 

[Hitman928]

Amd will never habe 50% market share.. granite rapids will crush turin

Cool story.

 

[H433x0n]

Though we'll of course have to see where Turin ends up in actual benchmarks once it shows up to market - the lack of any Genoa/Bergamo to Turin performance comparisons in their presentation stands out.

I didn’t even notice that. I went back and checked how Genoa was presented and it was compared against Milan 7763. They also had plenty of benchmarks of Milan against Rome.

The saving grace here is that it was a “preview” and not a launch so it may not have any significance.

 

[Khato]

Lisa Su announced it a quarter or two ago, I believe during an earnings call. I should note, though, that AMD typically talks in terms of revenue share, though not always, so they were probably referencing revenue share but I'm not positive about that.

Thanks, unfortunately I've had no luck finding a reference to such in the last 5 earnings call transcripts. They mostly just seem to mention the % revenue growth and how many new public instances are available. So probable that it was mentioned in one of their other events.

Even during the time Intel had 90% server market share, they never sold processors anywhere near the MSRP.

They clearly aren't getting enough profit off of each server chip to fund future nodes, that's for sure.

Well, they sold them for roughly MSRP at retail. But you're quite correct that all OEM and large scale purchases only pay a fraction of MSRP. Hence the caveat at the end of my statement that the numbers aren't accurate in absolute terms, but the ratios between them should be about right. Primary point was that the performance/$ of the top end SKU can be close to half that of the first step down. Anyway, the point I was trying to make is that while Intel isn't nearly as uncompetitive in the hyperscaler/cloud arena as general benchmarks would imply thanks in large part to their accelerators, they're still far enough behind that they can't charge the margins of a 'top end' SKU.

I didn’t even notice that. I went back and checked how Genoa was presented and it was compared against Milan 7763. They also had plenty of benchmarks of Milan against Rome.

The saving grace here is that it was a “preview” and not a launch so it may not have any significance.

Agreed that it most likely has little to no significance, especially given the other issues with the performance claims in their presentation. But it certainly was an interesting omission.

 

[del42sa]

https://wccftech.com/intel-13th-14th-gen-instability-issues-buggy-microcode-etvb-fix-bios-fix-0x125/

Although Intel has yet to publicly issue a statement regarding the serious matter of instability that affects its high-end 13th and 14th Gen CPUs despite saying that it would do so a few months back, it looks like Igor's Lab has discovered internal documents (NDA) which spill the beans on what has been causing these issues from the start.

update: Intel deny the report:

“Contrary to earlier reports, Intel has not confirmed root cause and is continuing, with its partners, to investigate user reports regarding instability issues on unlocked Intel Core 13th and 14th generation (K/KF/KS) desktop processors.

The microcode patch referenced in press reports fixes an eTVB bug discovered by Intel while investigating the instability reports. While this issue is potentially contributing to instability, it is not the root cause.”

https://videocardz.com/newz/intel-f...ility-issues-bios-with-new-microcode-underway

 

[lightisgood]

Finally, silicon photonics era will arrive on PCIe 7.0 ...?

Optical PCIe 7.0 connection hits a blazing 128 GT/s

Cadence shows off optical solution for next-generation PCIe 7.0 that is not yet finalized.

www.tomshardware.com

 

[lightisgood]

So, I said "This is typical Lisa's AMD".

AMD Rejects Tiny Corp's Request of Benchmarking Instinct MI300X AI Accelerator On MLPerf, Raising New Questions

AMD has declined the request of the AI startup Tiny Corp to put the Instinct MI300X AI accelerator on the MLPerf benchmarks.

wccftech.com

BTW, I'm very looking forward to see Granite Rapids-AP@128C.
It has almost same A100's AI performance.
I'm sure that GNR should break Turin at AI server market.

 

[SiliconFly]

So, I said "This is typical Lisa's AMD".

AMD Rejects Tiny Corp's Request of Benchmarking Instinct MI300X AI Accelerator On MLPerf, Raising New Questions

AMD has declined the request of the AI startup Tiny Corp to put the Instinct MI300X AI accelerator on the MLPerf benchmarks.

wccftech.com

BTW, I'm very looking forward to see Granite Rapids-AP@128C.
It has almost same A100's AI performance.
I'm sure that GNR should break Turin at AI server market.

GNR is RWC and I'm not expecting much. Diamond Rapids would be the real game changer I think.

 

[Hitman928]

So, I said "This is typical Lisa's AMD".

AMD Rejects Tiny Corp's Request of Benchmarking Instinct MI300X AI Accelerator On MLPerf, Raising New Questions

AMD has declined the request of the AI startup Tiny Corp to put the Instinct MI300X AI accelerator on the MLPerf benchmarks.

wccftech.com

Completely off topic for this thread but also a worthless drama article. A startup wanted AMD to pay them $1M and give them free MI300Xs to test them for MLperf. AMD would have been idiots to agree to that.

 

[ondma]

The node advantage AMD has had for years will be gone by the end of the year. All the hopes and dreams for a brighter tomorrow may be gone. Intel said they were going to put AMD in their rear view mirror. Will we see AMD fire sale prices again? One can only hope that is the case.

I think Pat may need to re-evaluate that statement. MAYBE they will be ahead in the low power mobile with LL, but otherwise they are still playing catch up. As far as the process advantage, it just blows my mind that intel is touting their foundry processes and still using TSMC for their flagship products. Using them for ARL and LL is bad enough, but isnt even Panther Lake still using TSMC?

 

[Kepler_L2]

I think Pat may need to re-evaluate that statement. MAYBE they will be ahead in the low power mobile with LL, but otherwise they are still playing catch up. As far as the process advantage, it just blows my mind that intel is touting their foundry processes and still using TSMC for their flagship products. Using them for ARL and LL is bad enough, but isnt even Panther Lake still using TSMC?

Panther Lake is mostly Intel nodes, outside of the iGPU on PTL-H.

 

[ondma]

Panther Lake is mostly Intel nodes, outside of the iGPU on PTL-H.

Oh, yea, Panther Lake is mobile only though. I was thinking of desktop, ARL-R will still be on TSMC.

 

[Khato]

Unfortunate fact for Intel is that ASML can only make so many EUV machines per year. There's no rectifying the prior CEO's mistake of declining to purchase Intel's EUV allocation and letting multiple years worth of production go to TSMC and Samsung. End result is that Intel couldn't maintain their current unit market share if they were to exclusively use Intel 4 and better nodes for a few years.

That's not to say that TSMC may not have the superior process technology, but Intel's usage of TSMC versus internal node is not necessarily indicative of such. With luck Intel will end up releasing same configuration of ARL on both N3B and 20A processes such that we can compare the end result on perf/power/area metrics.

 

[SiliconFly]

Unfortunate fact for Intel is that ASML can only make so many EUV machines per year. There's no rectifying the prior CEO's mistake of declining to purchase Intel's EUV allocation and letting multiple years worth of production go to TSMC and Samsung. End result is that Intel couldn't maintain their current unit market share if they were to exclusively use Intel 4 and better nodes for a few years.

That's not to say that TSMC may not have the superior process technology, but Intel's usage of TSMC versus internal node is not necessarily indicative of such. With luck Intel will end up releasing same configuration of ARL on both N3B and 20A processes such that we can compare the end result on perf/power/area metrics.

True. It takes time to fix the sins of the past. No fab can build capacity quickly. It takes a lot of time. If things go well, Intel will have all the capacity they need in a couple of years.

Even though Intel has been working in a frenzy towards 18A with their 5NI4Y plan, I don't think 18A will have sufficient capacity next year. PTL will need more time to hit volume. Probably mid-2026.

Oh, yea, Panther Lake is mobile only though. I was thinking of desktop, ARL-R will still be on TSMC.

Since, ARL-R is just a refresh, it stays with the existing node.

I think Pat may need to re-evaluate that statement. MAYBE they will be ahead in the low power mobile with LL, but otherwise they are still playing catch up. As far as the process advantage, it just blows my mind that intel is touting their foundry processes and still using TSMC for their flagship products. Using them for ARL and LL is bad enough, but isnt even Panther Lake still using TSMC?

Intel 3 just opened up. 18A next year. They don't have the capacity they need yet. Takes time. Starting next year, Intel will start using it own foundries for more of its own products. It's gonna take a couple of years before they start using their own fabs mostly for their own products.

I'm sure that GNR should break Turin at AI server market.

Not a fan of server parts. But just out of curiosity, how does GNR stack up against competition and/or prev gens? Is it gonna be like SPR/EMR? Or is it bringing in some much needed boost?

 

[Hans Gruber]

20A is a short term node. 18A will be around for at least 2 or 3 years. It will be an interesting few years coming up.

 

[igor_kavinski]

So, I said "This is typical Lisa's AMD".

AMD Rejects Tiny Corp's Request of Benchmarking Instinct MI300X AI Accelerator On MLPerf, Raising New Questions

AMD has declined the request of the AI startup Tiny Corp to put the Instinct MI300X AI accelerator on the MLPerf benchmarks.

wccftech.com

Tiny Corp are asking for a MILLION bucks. They should instead do a kickstarter for that if they want "free" money.

Few other problems:

1) AMD can probably do it themselves for a lot less.

2) Accepting that offer would be tantamount to admitting that their software teams are crap.

3) AMD probably has enough clients rounded up that they don't need benchmarks to prove their claims and they are most likely just showing prospective clients how others using their product are achieving good AI performance, under NDA, of course. This is not unusual in the industry. Plenty of niche companies require NDAs to showcase the true potential of their products to interested parties.

4) The Tiny Corp guy is a pompous idiot. He thinks he can say anything and people will just follow him like sheep. His intentions might be good but he is naive to think that poking a large company like AMD publicly would get him the desired results. AMD is doing just fine without him. He needs to stop trying to psychologically manipulate Lisa Su because she's too smart for him. What he needs is less talk and more hard work. The way normal people achieve success.

And by the way, LISA's AMD got Intel off their lazy, complacent asses so maybe be a bit more grateful for the shift from progressive decline to a more hopeful and fruitful future filled with exciting products from Intel as a direct result of AMD's push?

 

[dullard]

I think Pat may need to re-evaluate that statement. MAYBE they will be ahead in the low power mobile with LL, but otherwise they are still playing catch up. As far as the process advantage, it just blows my mind that intel is touting their foundry processes and still using TSMC for their flagship products. Using them for ARL and LL is bad enough, but isnt even Panther Lake still using TSMC?

Please tell me where on the image below (which node/nodes) Intel is claiming process leadership. Then please tell me how your post regarding process advantage and TSMC relates to that node(s).


Hint: The process advantage in their statements is not until well after ARL and LL.

 

[Ghostsonplanets]

Using them for ARL and LL is bad enough

You know that product decisions are made years in advance with little to no room for changes, right?

Lunar Lake choice of TSMC N3 usage was made back in 2020, with different leadership.

but isnt even Panther Lake still using TSMC?

No. And that's not a rumor but confirmed by Intel themselves.

Panther Lake uses Intel 18A for the Compute + SoC Tile. Panther Lake 32EU iGPU tile and the Foveros tile are also fabbed on Intel Foundry. TSMC will only fab the 96EU iGPU tile and the PCD tile.

So Panther Lake means a return of Intel SoCs being mainly fabbed on Intel Foundry.

 

[lightisgood]

You know that product decisions are made years in advance with little to no room for changes, right?

Lunar Lake choice of TSMC N3 usage was made back in 2020, with different leadership.


No. And that's not a rumor but confirmed by Intel themselves.

Panther Lake uses Intel 18A for the Compute + SoC Tile. Panther Lake 32EU iGPU tile and the Foveros tile are also fabbed on Intel Foundry. TSMC will only fab the 96EU iGPU tile and the PCD tile.

So Panther Lake means a return of Intel SoCs being mainly fabbed on Intel Foundry.

Yes.
The serious dependent of Intel on TSMC, it is the end in this year.

 

[Hans Gruber]

Please tell me where on the image below (which node/nodes) Intel is claiming process leadership. Then please tell me how your post regarding process advantage and TSMC relates to that node(s).
View attachment 101781

Hint: The process advantage in their statements is not until well after ARL and LL.

I will give you a hint. It's 18A.

 

[ondma]

Please tell me where on the image below (which node/nodes) Intel is claiming process leadership. Then please tell me how your post regarding process advantage and TSMC relates to that node(s).
View attachment 101781

Hint: The process advantage in their statements is not until well after ARL and LL.

You tell me why a company that is supposed to compete as a foundry is still buying their silicon from their major competitor.

 

[AcrosTinus]

You tell me why a company that is supposed to compete as a foundry is still buying their silicon from their major competitor.

That is a good thing !!.
Intel is being smarter now instead of tightly linking products to nodes. If the competition can print better chips let them do it, the products have to land on the shelf instead of being hamstrung by bad power efficiency, there is no pride here just business.