Discussion Intel current and future Lakes & Rapids thread

[liahos2]

Really? Where's ICL Xeon then? OnTrack™, right? So much that even NVIDIA had to use Epyc with their A100. And you know they would never do that if it wasn't absolutely necessary.

shipping to hyperscaler clients for revenue in q4 this year. With a broader follow on for enterprise in Q1.

And despite NVDA A100 using EPYC (likely for its better IO)... google is deploying ampere with xeon

Nvidia’s next-generation Ampere A100 GPU available on Google Cloud

Developers and scientists who need compute horsepower of 16 Nvidia’s A100 GPUs can now get it from Google.

www.techradar.com

 

[lobz]

shipping to hyperscaler clients for revenue in q4 this year. With a broader follow on for enterprise in Q1.

my point exactly

 

[liahos2]

my point exactly

That's been the stated rollout since their q4'19 earnings call.

 

[jpiniero]

shipping to hyperscaler clients for revenue in q4 this year. With a broader follow on for enterprise in Q1.

You think the hyperscalers would transition to Rome/Milan and not Icelake.

Further, alder lake looks like a streaming pile on a 1700 pin socket, which will have to go against a 5nm Zen4 chip.

That was kind of the whole point about there not being any 10 nm desktop chips - they wil be competing against AMD on 5 nm so moving to 7 nm is so crucial. But if 7 is now starting to slip too, perhaps they have no choice.

 

[liahos2]

You think the hyperscalers would transition to Rome/Milan and not Icelake.

Actually based on a call I had with Inspur - I think Rome adoption is being pushed back (due to covid) and milan has been delayed in 2h'21.

Also - do you look at intel financials? Their DCG revenue upside vs expectations are greater than how much money AMD makes in their non gaming business over the course of a year. With ASPS going UP not DOWN.

That was kind of the whole point about there not being any 10 nm desktop chips - they wil be competing against AMD on 5 nm so moving to 7 nm is so crucial. But if 7 is now starting to slip too, perhaps they have no choice.

 

[jpiniero]

Facebook loves them some bfloat, which of course Icelake Xeon doesn't have. I could see Rome adoption being pushed back, but only for so long and that would be for Cascade and not Icelake. If they were ready to move to a new platform, you would think they would use Rome/Milan.

 

[ajc9988]

That was kind of the whole point about there not being any 10 nm desktop chips - they wil be competing against AMD on 5 nm so moving to 7 nm is so crucial. But if 7 is now starting to slip too, perhaps they have no choice.

Did anyone truly believe Intel's timeline on 7nm? Intel pushed 14nm due to delays. They pushed 10nm for years. And now, magically, they are supposed to have a working 7nm node on time?

Intel, to their credit, did attempt COAG. But they did this without EUV being ready and without dummy gates, meaning a defect destroys not just the one transistor, but more. They weren't prepared and were too aggressive. It was huffing their own farts, so to speak.

Even after they stripped out elements from the process so it no longer was what they planned, volume was eh and frequency never hit fully where they wanted, with IPC of newer architectures not making up for it.

If Intel misses the boat in 2021/22, they may need to spinoff to keep going, tbh.

Facebook loves them some bfloat, which of course Icelake Xeon doesn't have. I could see Rome adoption being pushed back, but only for so long and that would be for Cascade and not Icelake. If they were ready to move to a new platform, you would think they would use Rome/Milan.

This, in part, is still per core licensing and speed based single thread performance. Then there is still software optimizations. Think how long it took for Matlabs to fix their instructions when AMD was detected, which is funny because it was Intel code that caused it.

Either way, AMD won the largest super computer builds recently, so it literally is just a matter of time in market share. A when, not an if. But they also need to make sure they can produce enough to meet demand....

 

[Hitman928]

Actually based on a call I had with Inspur - I think Rome adoption is being pushed back (due to covid) and milan has been delayed in 2h'21.

Also - do you look at intel financials? Their DCG revenue upside vs expectations are greater than how much money AMD makes in their non gaming business over the course of a year. With ASPS going UP not DOWN.

That was kind of the whole point about there not being any 10 nm desktop chips - they wil be competing against AMD on 5 nm so moving to 7 nm is so crucial. But if 7 is now starting to slip too, perhaps they have no choice.

Wait, you're saying that you think Milan is delayed by ~ 1 year? That would certainly go against AMD's own recent statements and the leaks of engineering samples being tested. Or are you just referring to when Inspur plans on deploying Milan based systems? If so, that would make more sense and I wouldn't really expect any different. How many solutions for AMD has Inspur even offered to date?

 

[liahos2]

Wait, you're saying that you think Milan is delayed by ~ 1 year? That would certainly go against AMD's own recent statements and the leaks of engineering samples being tested. Or are you just referring to when Inspur plans on deploying Milan based systems? If so, that would make more sense and I wouldn't really expect any different. How many solutions for AMD has Inspur even offered to date?

sorry let me clarify. Inspur (along with other checks i've done) have said that rome adoption slower than anticipated and also partially delayed due to quals getting pushed out by customers. Inspur "believe us if you want to or not...but we think Milan pushed to 2h21 from Q1'21". I dont know if they meant for them or for the market. They are 50% of the china Server market though

 

[Hitman928]

sorry let me clarify. Inspur (along with other checks i've done) have said that rome adoption slower than anticipated and also partially delayed due to quals getting pushed out by customers. Inspur "believe us if you want to or not...but we think Milan pushed to 2h21 from Q1'21". I dont know if they meant for them or for the market. They are 50% of the china Server market though

Ok, thanks for the clarification.

 

[DrMrLordX]

You misunderstand the purpose of the XT lineup, which more likely was to iron out process used for the upcoming Zen 3 chips

Zen3 is using N7+. XT chips use N7. There's nothing about the XT lineup which helps them "iron out" N7+. If they needed to buy time, all they had to do was keep selling Matisse instead of rebranding it. The XT chips let them reset price points with an eye towards upwards price movement in October. AMD got away with it because Comet Lake-S offers little credible competition. Expect Vermeer to cost more for the 12c part than the 3900X did last year.

while technically having something competing against comet at each price point.

. . . they already had that.

You ignore Intel's done that multiple times before.

No I haven't. Intel has been rehashing Skylake for ages. Comet Lake-S is the pinnacle of process/design stagnation. For an industry that wants to move forward, it's an abomination.

Further, alder lake looks like a streaming pile on a 1700 pin socket

If Alder Lake-S were out right now, it might be interesting. Launching it more than a year from now does not bode well for the product. Again, hence my commentary on stagnation. Which is why I responded to @Dave2150 in the first place. There was little reason for him to be enthusiastic when the industry is so plagued by delays, intentional or otherwise.

 

[lobz]

That's been the stated rollout since their q4'19 earnings call.

Since the stated rollout has been lied about from quarter to quarter and then casually delayed in between quarterly reports, I'm not too inclined to be impressed when the current, already heavily delayed rollout date is not changed for 2 quarters. But the spins you use are a bit strange on a tech forum. We're not acting as investors here, you don't have to convince anyone.

 

[uzzi38]

sorry let me clarify. Inspur (along with other checks i've done) have said that rome adoption slower than anticipated and also partially delayed due to quals getting pushed out by customers. Inspur "believe us if you want to or not...but we think Milan pushed to 2h21 from Q1'21". I dont know if they meant for them or for the market. They are 50% of the china Server market though

Milan is Q4 20 announcement, Q1 21 availability for everyone outside of a few select customers.

Whoever you contacted is either talking about when they're getting Milan or is redacted you.

We have a zero tolerance policy for profanity in the tech sub-forums.
Don't do it again.

Iron Woode
Super Moderator

 

[uzzi38]

shipping to hyperscaler clients for revenue in q4 this year. With a broader follow on for enterprise in Q1.

And despite NVDA A100 using EPYC (likely for its better IO)... google is deploying ampere with xeon

Nvidia’s next-generation Ampere A100 GPU available on Google Cloud

Developers and scientists who need compute horsepower of 16 Nvidia’s A100 GPUs can now get it from Google.

www.techradar.com

I know for a fact that at least one of the biggest hyperscalers (not actually stating who or their ranking here) has not even received a single ICL-SP sample. Only remote access.

I highly doubt Q4 for Hyperscalers.

 

[naukkis]

How about that recently found nasty bug in Icelake?

Intel Ice Lake CPUs Have a System Crashing Bug

Intel CPUs have been rather notorious for system bugs recently. Starting from 2018's Spectre and Meltdown which used speculative execution to exploit systems, the string of new vulnerabilities just continued to this day. Recently we had CrossTalk exploit which represents a threat to cloud...

www.techpowerup.com

Seems that there isn't fix yet - wonder if server parts suffer from same bug.

 

[uzzi38]

How about that recently found nasty bug in Icelake?

Intel Ice Lake CPUs Have a System Crashing Bug

Intel CPUs have been rather notorious for system bugs recently. Starting from 2018's Spectre and Meltdown which used speculative execution to exploit systems, the string of new vulnerabilities just continued to this day. Recently we had CrossTalk exploit which represents a threat to cloud...

www.techpowerup.com

Seems that there isn't fix yet - wonder if server parts suffer from same bug.

I suspect this is what may have caused a delay for ICL-SP, if there's any at all.

 

[ajc9988]

Zen3 is using N7+. XT chips use N7. There's nothing about the XT lineup which helps them "iron out" N7+. If they needed to buy time, all they had to do was keep selling Matisse instead of rebranding it. The XT chips let them reset price points with an eye towards upwards price movement in October. AMD got away with it because Comet Lake-S offers little credible competition. Expect Vermeer to cost more for the 12c part than the 3900X did last year.



. . . they already had that.



No I haven't. Intel has been rehashing Skylake for ages. Comet Lake-S is the pinnacle of process/design stagnation. For an industry that wants to move forward, it's an abomination.



If Alder Lake-S were out right now, it might be interesting. Launching it more than a year from now does not bode well for the product. Again, hence my commentary on stagnation. Which is why I responded to @Dave2150 in the first place. There was little reason for him to be enthusiastic when the industry is so plagued by delays, intentional or otherwise.

So you really don't pay attention to AMD news, do you? Here is an Anandtech article addressing the process from March:

AMD Clarifies Comments on 7nm / 7nm+ for Future Products: EUV Not Specified

www.anandtech.com

Notice, EUV was removed as a guarantee. More recently, people have speculated new wafers or process refinement for the increase in frequency found on some chips generally.

These XT chips are binned, amount to 2-5% tops, and were just to fill out the stack. But you focus so heavily on those instead of AMD saying we'll have Zen 3 in a couple months. You downplay it as stagnation with significant latency changes and 20% IPC.

You just sound salty. Intel has nothing really in the stack, so you say the entire industry is stagnant.

 

[JoeRambo]

How about that recently found nasty bug in Icelake?

Intel Ice Lake CPUs Have a System Crashing Bug

Intel CPUs have been rather notorious for system bugs recently. Starting from 2018's Spectre and Meltdown which used speculative execution to exploit systems, the string of new vulnerabilities just continued to this day. Recently we had CrossTalk exploit which represents a threat to cloud...

www.techpowerup.com

Seems that there isn't fix yet - wonder if server parts suffer from same bug.

There is a fix already in microcode. At least so was posted by Intel rep in intelliJ thread for this nasty bug.

 

[jpiniero]

So I went back and looked at AT's articles. Even in early-mid 2019, Intel was still implying that Icelake Xeon would be available publicly in 2020. That's obviously not happening now.

Did anyone truly believe Intel's timeline on 7nm?

Sort of, if only because any real slippage means the end of the fabs. They simply won't have the profits to be able to afford anything past 7.

 

[ajc9988]

So I went back and looked at AT's articles. Even in early-mid 2019, Intel was still implying that Icelake Xeon would be available publicly in 2020. That's obviously not happening now.



Sort of, if only because any real slippage means the end of the fabs. They simply won't have the profits to be able to afford anything past 7.

I disagree, to a degree, but generally agree with the sentiment. They won't have profits to continue on the cutting edge.

It is much the same way GloFo dropped as soon as AMD went TSMC. They are doing great things regarding FDSOI transistors, which are not as sexy as the smallest node size race, but it is impressive in its own right.

So to would be Intel's manufacturing. I'd say spin them off, let them open up, and focus on their IP for x86. They've been sitting on core designs for years waiting for the node process. A spin off, in theory, ends the wait. It also allows the items that don't need the cutting edge to use intel's 14nm without a much competition, as Intel will have to use TSMC or Samsung for cutting edge process.

The other alternative is to sell off the fab IP to TSMC. That could be interesting... or to Samsung. Or Nvidia (that would really be fascinating)!

 

[DrMrLordX]

So you really don't pay attention to AMD news, do you?

Zen3 is confirmed N7+, ask in the relevant threads if you don't believe me.

You just sound salty.

You just sound misinformed.

Intel has nothing really in the stack, so you say the entire industry is stagnant.

Now the competition gets to milk the market for profits thanks to no competition.

 

[ajc9988]

Zen3 is confirmed N7+, ask in the relevant threads if you don't believe me.



You just sound misinformed.



Now the competition gets to milk the market for profits thanks to no competition.

So, first, you say ignore listening to the corporation saying they dropped it to avoid confusion and assumptions, while using it where it makes sense, meaning that it may be used with specific products in their lineup. Does that mean the primary core chiplet? Likely not. But removing the guidance says do not discount, even if 7nm+ EUV is the most likely for core chiplets.

Also, saying they get to milk the market isn't stagnation in and of itself, thereby proving you are salty.

Edit:

As a for instance on troubleshooting the current process or wafer changes, we have zen 2 cores on both upcoming consoles. We do not know which graphics cards will get 7nm or 7nm+ from TSMC, etc. There is a large product stack.

But your salt says there can be no other reason.

 

[lobz]

Now the competition gets to milk the market for profits thanks to no competition.

If there's one thing not needing any speculation is this: AMD milking the market for profits? Not happening in at least 5 years.

 

[DrMrLordX]

So, first, you say ignore listening to the corporation saying they dropped it to avoid confusion and assumptions,

Ask in the relevant thread. If you followed the AMD news, you would know better. N7+ is confirmed. I don't feel like digging it up but it's in the Zen3/Ryzen 4000 speculation thread and possibly elsewhere. That data from March is obsolete and was never conclusive. I can't help you if you insist on clinging to fallacies.

@lobz

AMD is already milking the DiY market. Watch what happens in October . . .

 

[naukkis]

There is a fix already in microcode. At least so was posted by Intel rep in intelliJ thread for this nasty bug.

Newer one than that non-working microcode update mentioned in that article?