Discussion Intel current and future Lakes & Rapids thread

[Hitman928]

That statement (verbal) is different from their official press release.

No, it's not. Lisa Su just called out the specific product lines in the earnings call rather than the reporting segment, but they are congruent statements.

AMD Reports Third Quarter 2020 Financial Results

Browse AMD’s company-wide and financial press releases.

ir.amd.com

"View attachment 34943

Enterprise, Embedded, and Semi-Custom is all lumped together in one group, i.e. server CPUs/GPUs, embedded CPUs/GPUs, and consoles all get lumped together. Computing and Graphics is for consumer level CPUs and GPUs.

Lisa Su just made it a little more specific in stating that revenue increase was driven by Epyc, Ryzen, and consoles.

They even pointed out that gross margin was flat due to the high sales of semi-custom chips.

They are pointing out that GM only stayed flat because of increased console sales. If you take those sales out of the equation, GM would have increased.

And then here it quantifies it - 116% increase due to semi-custom and EPYC sales, and I'd bet almost all of it is due to semi-custom Xbox/PS5. 80% of their Q4 7nm allocation is for those. You won't be seeing a lot of Zen 3 for a while :

View attachment 34944

116% increase Y/Y in that one segment. Again, console chips are notoriously low margin. You cannot have flat GM if your revenue increase is being driven so largely by console chips, it's mathematically impossible. I've done a break down before, but in Q1 of this year, AMD reported that console revenue was negligible. That means that in that quarter almost all of the segment revenue was from datacenter products. I broke down what the means for Epyc revenue in Q1 at the time in this thread. Long story short, from Q1 Epyc revenue to Q3, they had "strong double digit growth" for Epyc sales which means even if we are being conservative, Epyc revenue in Q3 was approaching $500M. As far as that wafer allocation rumor, I don't believe it, AMD's financials don't support it at all.

 

[shady28]

116% increase Y/Y in that one segment. Again, console chips are notoriously low margin. You cannot have flat GM if your revenue increase is being driven largely by console chips, it's mathematically impossible. I've done a break down before, but in Q1 of this year, AMD reported that console revenue was negligible. That means that in that quarter almost all of the segment revenue was from datacenter products. I broke down what the means for Epyc revenue in Q1 at the time in this thread. Long story short, from Q1 Epyc revenue to Q3, they had "strong double digit growth" for Epyc sales which means even if we are being conservative, Epyc revenue in Q3 was approaching $500M. As far as that wafer allocation rumor, I don't believe it, AMD's financials don't support it at all.

That's a math fail for you.

Revenue was driven by consoles.

Gross margin was flat because console margin is low. Margin for server chips is high, and margin for client is low (though not as low as consoles).

So, AMD sold a lot more server chips, and a lot more console chips, but overall margin was flat as they offset each other.

I don't really care about margin that much. My main point is that Intel's volume is so much higher than AMD, they cannot displace them right now or anytime in the foreseeable future. And, Zen 3 is probably going to be rare in the client / pc / laptop space for the next quarter or two.

Why? Because 80% of their production allocation from Q4 TSMC is into consoles, and if I were them I'd put the remaining into high margin server. Wouldn't you?

But put this in perspective. TSMCs 7nm allocation to AMD was ~120,000 wafers. That's about what one mega-fab makes in one month. Intel has 16 fabs... <-- that's not going to change anytime soon.

So, expect Intel to lose server share and gain client/PC share, and don't expect to see a bunch of availability on desktop zen 3 for a while.

 

[Hitman928]

That's a math fail for you.

Revenue was driven by consoles.

Gross margin was flat because console margin is low. Margin for server chips is high, and margin for client is low (though not as low as consoles).

So, AMD sold a lot more server chips, and a lot more console chips, but overall margin was flat as they offset each other.

You say I have a math fail but offer no kind of math at all to refute. Nope, margin for Ryzen is not low, not at all. It's not as high as Epyc, but it is not low. AMD has made that clear in their earnings call, it is clear in their financial reports, and it is not that hard to calculate (ballpark) based upon the sale price of Ryzen CPUs. To prevent further derailment of this thread, I'll just link to the thread about AMD's 3Q financial results. All the math is there if you want to refute AMD's own statements and reported numbers:

News - AMD 3Q20 results

AMD had a very strong (i.e. record setting) 3rd quarter blowing past estimates. https://www.globenewswire.com/news-release/2020/10/27/2114937/0/en/AMD-Reports-Third-Quarter-2020-Financial-Results.html GAAP Quarterly Financial Results Q3 2020 Q3 2019 Y/Y Q2 2020 Q/Q Revenue ($M)...

forums.anandtech.com

I don't really care about margin that much. My main point is that Intel's volume is so much higher than AMD, they cannot displace them right now or anytime in the foreseeable future. And, Zen 3 is probably going to be rare in the client / pc / laptop space for the next quarter or two.

I don't think anyone said AMD would take over the majority of the market anytime soon, AMD have not given any kind of guidance for this at all either. What they can do is continue to chip away at Intel's market share while at the same time reducing their gross margin which will eventually effect Intel's ability to maintain the majority of the market share, at least under their current operating model. As I said a few posts back, I think the turning point (not the end point, but the turning point) for this will be if they have major 7 nm issues. If their 7 nm ramp is as bad as 10 nm, I think it will only be 2 - 3 years into 7 nm delays/yield issues that Intel will be forced to significantly change how they do business.

Why? Because 80% of their production allocation from Q4 TSMC is into consoles, and if I were them I'd put the remaining into high margin server. Wouldn't you?

This does not work out logically or mathematically. If so much of their production is for consoles, please show the math how in Q4 AMD is projecting to increase revenue Q/Q by 7% while GM is projected to increase by 1% and Epyc, datacenter GPUs, embedded, and console revenue all combined was only 40.3% of Q3 revenue. I would love to see how the match works out on that.

But put this in perspective. TSMCs 7nm allocation to AMD was ~120,000 wafers. That's about what one mega-fab makes in one month. Intel has 16 fabs... <-- that's not going to change anytime soon.

So, expect Intel to lose server share and gain client/PC share, and don't expect to see a bunch of availability on desktop zen 3 for a while.

We don't know how many wafers AMD has been allocated, this is all based upon sources with unknown reliability. But again, no one is arguing AMD is going to take over Intel's volume next quarter or next year. The threat to Intel is over the next few years with a potential turning point with their 7 nm production.

 

[shady28]

You say I have a math fail but offer no kind of math at all to refute.

I quoted your post where you stated this :

"You cannot have flat GM if your revenue increase is being driven largely by console chips, it's mathematically impossible. "

That's a math fail.

This does not work out logically or mathematically. If so much of their production is for consoles, please show the math how in Q4 AMD is projecting to increase revenue Q/Q by 7% while GM is projected to increase by 1% and Epyc, datacenter GPUs, embedded, and console revenue all combined was only 40.3% of Q3 revenue. I would love to see how the match works out on that.

Q4 production limitations won't be felt until Q1 next year.

If you are questioning the limitations of their fab contracts, just goolge it. The source article came from Taiwan and is in Chinese, but there are a ton of summaries.

AMD reserves 80% of its SoC chips for PS5 and XSX - iGamesNews

We assume as normal that a lot of people want the same product at the same time, that a lot of hype is generated by the current console or the rigorous GPU,

www.igamesnews.com

"AMD has reportedly made deals with SONY and Microsoft to sell their custom chips for PS5 and Xbox Series X, to the point where the Reds have nothing less than 120,000 wafers "

80% of the wafers received by AMD went to manufacture of consoles

AMD reserved round 120,000 wafers @7nm manufacturing process for next-generation consoles from TSMC. Most interesting of all

tech4gamers.com

"Turning the data around, this means that 80% of the wafers provided by TSMC to AMD went to consoles, while with the remaining 20% - around 30,000 wafers, the company has not been able to meet the demand for Ryzen 5000, the Radeon RX 6000 GPUs, and even the AMD Ryzen 3000 prices are skyrocketing due to stock availability issues. In short, AMD has given high priority to consoles, affecting rest of the market."


It doesn't really matter if your napkin math on financials fails to resolve this with their margins and revenue and so on. This is what's happening to production in Q4, it's already affected their launches on the DIY market and will likely continue to affect it throughout Q1 2021.

AMD can't sell what they don't have.

Intel is pretty much going to own the desktop market through the end of Q1 2021 by default.

Nvidia will likewise own the GPU market by default.

By the time they can get capacity, maybe in Q1 2021, Rocket Lake will be out. Edit: this is actually immaterial. AMD won't have chips to sell in any qty, so Intel can do whatever.

Meanwhile, Qualcomm has reportedly become the #1 customer for 7nm thanks to iPhone 12 modems.

AMD has to compete for fab capacity with companies that are much larger and with much deeper pockets than it has.

And this is why I refer to Zen 3 and 68XX as paper launches.

 

[Panino Manino]

But put this in perspective. TSMCs 7nm allocation to AMD was ~120,000 wafers. That's about what one mega-fab makes in one month. Intel has 16 fabs... <-- that's not going to change anytime soon.

Why GF gave up 7nm again?
Was partly due to disbelief that Zen wouldn't be successful enough? That Zen 2 sales alone wouldn't be enough to pay the investment?

 

[coercitiv]

AMD can't sell what they don't have.

Yup, they've been selling empty boxes through the entire year.

 

[John Carmack]

Ladies and gentlemen please stop. There's already a thread for the people who want to talk about free-falling margins.

Question - Intel Q3: Ouch

https://d1io3yog0oux5.cloudfront.net/_c675d259e2193ee11667e182e2fab5ca/intel/news/2020-10-22_Intel_Reports_Third_Quarter_2020_Financial_1424.pdf Desktop volume down 16%, Notebook up 16% but ASP down 7%. Q4 forecast is bad too, they are expecting 14% decline in revenue compared to Q4 last year.

forums.anandtech.com

 

[Gideon]

Why GF gave up 7nm again?
Was partly due to disbelief that Zen wouldn't be successful enough? That Zen 2 sales alone wouldn't be enough to pay the investment?

They probably didn't get to choose.

They were behind their targets for a while and AMD picked TSMC as a target for both Zen 2 chiplet and RDNA1 at least a year before this was announced.
According to this article AMD already had multiple tape-outs at TSMC when this decision was made public, therefore it had to be at least a year earlier. That is about the time it takes to finalize a chip for tapeout and TSMC does not allow the same team of engineers to work with them and a competitor on the same product.

In fact, before the cancellation only one chip in 2019-2020 timeframe was still being planned on GloFo (possibly Renoir). There was an AMD statment saying this at the time, but I'm unable to find it now.
https://www.anandtech.com/show/13277/globalfoundries-stops-all-7nm-development
Now in hindsight, considering Intel's continued issues and TSMCs overbooking (also due-to COVID) and Samsungs inability to produce, the investment probably would have worked out, but nobody could have forseen that in august 2018.

TL;DR:
They missed their targets and amd was going TSMC anyway

 

[Panino Manino]

Yup, they've been selling empty boxes through the entire year.

The facts that AMD is selling well and selling less than it could can both coexist.

They probably didn't get to choose.

They were behind their targets for a while and AMD picked TSMC as a target for both Zen 2 chiplet and RDNA1 at least a year before this was announced.
According to this article AMD already had multiple tape-outs at TSMC when this decision was made public, therefore it had to be at least a year earlier. That is about the time it takes to finalize a chip for tapeout and TSMC does not allow the same team of engineers to work with them and a competitor on the same product.

In fact, before the cancellation only one chip in 2019-2020 timeframe was still being planned on GloFo (possibly Renoir). There was an AMD statment saying this at the time, but I'm unable to find it now.
https://www.anandtech.com/show/13277/globalfoundries-stops-all-7nm-development
Now in hindsight, considering Intel's continued issues and TSMCs overbooking (also due-to COVID) and Samsungs inability to produce, the investment probably would have worked out, but nobody could have forseen that in august 2018.

TL;DR:
They

I'm thinking that maybe AMD could offload some 7nm chips to GF to not be at TSMC's mercy.

 

[Hitman928]

I'm thinking that maybe AMD could offload some 7nm chips to GF to not be at TSMC's mercy.

GF's 7 nm was significantly behind schedule, that led to AMD chosing to go with TSMC for 7 nm. Without AMD, GF didn't really have any customers for 7 nm and so it was pointless to try and keep going with 7 nm and beyond, they were only falling further behind. Had they been able to see the future, maybe they could have agreed to some licensing deal like they did with 14 nm just to recoup some lost R&D on the advanced nodes, but they'd still have to get TSMC to agree to it as well.

 

[shady28]

The facts that AMD is selling well and selling less than it could can both coexist.



I'm thinking that maybe AMD could offload some 7nm chips to GF to not be at TSMC's mercy.

Actually Samsung might have been a better option. And specifically, Samsung 5nm, as they are ramping 5nm on their Exynos for 2021. Given the way they do their naming convention, 5nm Samsung is probably between TSMC 5nm and TSMC 7nm, or between Intel 7nm and Intel 10nm (Intel 7nm is reportedly more dense than TSMC 5nm).

More on topic, it's entirely possible that by the time AMD can ramp Zen 3 up to expose it to more than bot watchers and scalpers, it'll be Q2 2021. If RKL performs like the most recent benchmarks suggest, they may no longer have the desktop performance crown (at least, not in the 8 core and under market). At that point we're also very close to Q3/Q4 and Alder Lake / Zen 4.

 

[majord]

I see with Intel taking a pounding at a technical / performance / execution / short-medium term roadmap level the conversation turns to financials and market share.

Like there was ever any doubt of Intel remaining in a far superior position in total market share , profits etc.. It would take many years and many more market segments tackled by the competion for this ever to be a thing , so why bother discussing it?

About the only metric worth bothering with is trends. And it's all downward at the moment. And upward for the competition.

 

[LightningZ71]

Arguably, it's not going to matter to AMD with respect to volume. I propose that total market demand currently exceeds the capacity of both Intel AND AMD to supply it. Both companies, as long as their products are relatively performance and not flawed, can sell every single chip that they make no matter what. The only thing that they are competing in is average retail prices that their products can support and, by extension, the wholesale prices that their products command.

It doesn't matter how many wafers that AMD can buy, so long as they can afford to buy them at a price that recovers their R&D overhead. They will merily continue along at their current cadence as long as revenue continues to stream in.

Intel is only being hurt at retail on the amount of a premium that they can charge on their products. This hasn't hit their server products very hard yet. It hasn't hurt their atom line either as AMD doesn't really compete very hard there either. It has only really had a material affect on their Gen10 desktop and mobile parts. Even then, they are far above break even.

NO ONE ON THE SUPPLY SIDE WANTS TO DRASTICALLY INCREASE VOLUME!!! It only hurts ASPs for everyone.

Over the next year, everyone will sell everything they make at a profitable price. There's nothing showing that this will change.

 

[shady28]

NO ONE ON THE SUPPLY SIDE WANTS TO DRASTICALLY INCREASE VOLUME!!! It only hurts ASPs for everyone.

Over the next year, everyone will sell everything they make at a profitable price. There's nothing showing that this will change.

Actually TSMC and Samsung do. Both of them can sell everything they make at a premium, and more.

TSMC 5nm and 7nm processes are fully loaded until H2 2021

Then if Biden walks in and lifts restrictions on Huawei and so on, that will add even more demand.

I think what we are seeing is the bifurcation in the market between the haves and have nots. The cost per chip normally went down with a new node, this time that's not happening.

At 5nm, the cost per chip is going up :

Analysts believe that a single TSMC 5nm wafer costs $17,000

On top of this, and I can't find it at the moment, but I saw where TSMC/GloFlo/Samsung were all raising their prices by 20%.


We usually think of Qualcomm, Apple, AMD and so on when we think of chips. But none of those companies control the means of production, and with this high demand and high risk investment costs those companies that do control it (TSMC / GloFlo / Samsung) are looking to make big bank.

 

[LightningZ71]

GloFo? Do they even have a customer for their best process yet? (12LP+). Are they even still making progress on 12FDX?

 

[yuri69]

More on topic, it's entirely possible that by the time AMD can ramp Zen 3 up to expose it to more than bot watchers and scalpers, it'll be Q2 2021. If RKL performs like the most recent benchmarks suggest, they may no longer have the desktop performance crown (at least, not in the 8 core and under market). At that point we're also very close to Q3/Q4 and Alder Lake / Zen 4.

The low availability of Zen 3 was kinda hinted by absolute lack of leaks - the channels were not flooded as were with AMD's previous CPUs.

AMD 2020 roadmap was also super clear about TSMC not being able to be able to supply enough wafers for 7nm production. Look at what AMD planned to produce in 2020:

* continue with 2019 Zen 2
* expand Zen 2 to mobile
* release RDNA2
* release CDNA
* release Zen 3 to desktop and server
* release PS5 SoC
* release XBox SoC

However, the actual market/production expansion list for TSMC 7nm is longer:

* continue with Zen 2 - desktop + servers
* continue with RDNA1 - desktop
* produce new Apple-specific RDNA1 chips
* produce new high-volume Zen 2-based mobile Renoir
* produce low-volume desktop Renoir
* produce new "up to 64c" Threadrippers
* produce new high-volume PS5 SoC
* produce new high-volume XBox SoC
* produce new Zen 3 - desktop
* start sampling Zen 3 - servers
* start sampling very large CDNA Arcturus chips
* produce new large RDNA2 Navi 2x chips

This was a lot for TSMC... AMD knew this and had to prioritize. Simple as that. Going to another foundry might not be feasible due TSMC partnerships, etc.

So yea, Zen 3 is definitely a 2021 product, but no way Zen 4 gets released in "early 2022", let alone Q3/Q4 2021. I guess the same applies to Alder Lake, Intel will try hard to push it in 2021, but the availability is still questionable (largish 10nm die at high frequencies, packaging, advanced firmware).

 

[DisEnchantment]

Intel 8-core/16-thread Rocket Lake-S CPU offers similar performance to Ryzen 7 5800X - VideoCardz.com

Intel Rocket Lake-S makes a new appearance on the Geekbench benchmark site. Intel Rocket Lake-S with 8-cores and 16-threads at 5.0 GHz Alleged Rocket Lake-S CPU has been spotted at the Geekbench website (which at this point should be called “the source of all the leaks”). The CPU Model 167...

videocardz.com

Rocket Lake pops up in GB 1645@~5GHz

 

[mikk]

Geekbench IPC very close to Sunny Cove I would say, the first RKL-S entry there was clearly borked.

 

[shady28]

GloFo? Do they even have a customer for their best process yet? (12LP+). Are they even still making progress on 12FDX?

GloFlo is making bank. The super small nodes get a lot of coverage, but in reality there aren't that many sub 65nm fabs in the world much less 12nm.

For that matter, AMD uses GloFlo for the 12nm IO on Zen 3. Those are really kind of a hybrid 7nm chiplet with 12nm IO. They were also making their RX 580 / 590 GPUs, 5600XT, 5700/5700XT etc. I think GloFlo also made IBMs PowerPC chips, which are arguably some of the most powerful CPUs in the world.

 

[Bouowmx]

I tried finding a specific result Geekbench 5 result of AMD Ryzen 7 5800X that closely matches the aggregate: https://browser.geekbench.com/v5/cpu/5121238
Going to the sub-scores: Integer score is the same

Score	Intel Rocket Lake, 8 cores, 5.0 GHz	AMD Ryzen 7 5800X
Single-core	1645	1666
Crypto	4323	3967
Integer	1416	1420
Floating Point	1696	1814

 

[Dave2150]

GloFlo is making bank. The super small nodes get a lot of coverage, but in reality there aren't that many sub 65nm fabs in the world much less 12nm.

For that matter, AMD uses GloFlo for the 12nm IO on Zen 3. Those are really kind of a hybrid 7nm chiplet with 12nm IO. They were also making their RX 580 / 590 GPUs, 5600XT, 5700/5700XT etc. I think GloFlo also made IBMs PowerPC chips, which are arguably some of the most powerful CPUs in the world.

All of those GPU's you mentioned are no longer in production..... AMD's new GPU's and CPU's are all 7nm

TSMC are worlds ahead of GloFlo.

 

[shady28]

All of those GPU's you mentioned are no longer in production..... AMD's new GPU's and CPU's are all 7nm

TSMC are worlds ahead of GloFlo.

Reference? Everything I see indicates they are still making 570s, 580s, and so on right along with Zen 1+. Please separate your cheer-leading from facts.

 

[Edrick]

The low availability of Zen 3 was kinda hinted by absolute lack of leaks - the channels were not flooded as were with AMD's previous CPUs.

AMD 2020 roadmap was also super clear about TSMC not being able to be able to supply enough wafers for 7nm production. Look at what AMD planned to produce in 2020:

* continue with 2019 Zen 2
* expand Zen 2 to mobile
* release RDNA2
* release CDNA
* release Zen 3 to desktop and server
* release PS5 SoC
* release XBox SoC

However, the actual market/production expansion list for TSMC 7nm is longer:

* continue with Zen 2 - desktop + servers
* continue with RDNA1 - desktop
* produce new Apple-specific RDNA1 chips
* produce new high-volume Zen 2-based mobile Renoir
* produce low-volume desktop Renoir
* produce new "up to 64c" Threadrippers
* produce new high-volume PS5 SoC
* produce new high-volume XBox SoC
* produce new Zen 3 - desktop
* start sampling Zen 3 - servers
* start sampling very large CDNA Arcturus chips
* produce new large RDNA2 Navi 2x chips

This was a lot for TSMC... AMD knew this and had to prioritize. Simple as that. Going to another foundry might not be feasible due TSMC partnerships, etc.

So yea, Zen 3 is definitely a 2021 product, but no way Zen 4 gets released in "early 2022", let alone Q3/Q4 2021. I guess the same applies to Alder Lake, Intel will try hard to push it in 2021, but the availability is still questionable (largish 10nm die at high frequencies, packaging, advanced firmware).

I'm sure there are plenty of AMD threads to post in. This is supposed to be about Intel products.

 

[ondma]

I'm sure there are plenty of AMD threads to post in. This is supposed to be about Intel products.

Rocket Lake would have been a nice product if it had come out 6 months ago. Now that Ryzen 5xxx is out, it just seems kind of meh.

 

[shady28]

I tried finding a specific result Geekbench 5 result of AMD Ryzen 7 5800X that closely matches the aggregate: https://browser.geekbench.com/v5/cpu/5121238
Going to the sub-scores: Integer score is the same

Score	Intel Rocket Lake, 8 cores, 5.0 GHz	AMD Ryzen 7 5800X
Single-core	1645	1666
Crypto	4323	3967
Integer	1416	1420
Floating Point	1696	1814

Yeah I tried doing some of the same, the thing is the scores are all over the place and there isn't enough information to see why. Obviously, a lot of it is OC \ power limit settings, some is RAM speeds, and some may be thermals / environment.

For example, here's a 10700K beating a 5800X on both single and multi-core.

Without system details It is almost impossible to tell what is going on.