News AMD 3Q Earnings Report

[Hitman928]

AMD will announce 3Q earnings after the market closes today. I will update this post as info comes out.

In the mean time, saw this on [WCCF]

AMD Outsells Intel in Korea for First Time in History, Doubles Sales in Europe

Once again AMD finds itself making headlines ahead of its earnings report tomorrow for the headway its products have recently been making around the world. In Korea, the company's processors have reportedly outsold Intel's for the first time in history. According to the research arm of...

wccftech.com

AMD has traditionally done very well in the do-it-yourself channel market, where customers buy parts separately and build their own PCs. Although the pre-built OEM market has been much much harder for the company to penetrate, due to Intel's multi-decade long entrenchment in that space.


Which is precisely why this is such a big deal, AMD has rarely ever outsold Intel in the pre-built market and it has in fact never done so in Korea, a market where both of AMD's competitors NVIDIA & Intel have traditionally been very strong in.

I'm not familiar with the Korean market so I don't know how big it is or how much it is driven by more boutique builders or by large OEMs. A European report linked in the WCCF article states that AMD rose from 7% to 12% of pre-built machines in 3Q in the European region. It will be interesting to see what AMD reports today in their quarterly earnings.

 

[krumme]

Guys. We are here to have fun.
Margins is exactly as predicted. Spot on.
Revenue and profit more or less the same. I think we can reserve the biggest celebrations, enthusiasm, slamming, trolling and hyperbole for another more interesting quarter.

This is purely speculation but if we look into the details I think Epyc1 have had a slighted lower share than they had hoped for perhaps due to it's more inconsistent performance for various loads. It naturally affects epyc2 adaptations because you start from a lower brand perception.
Lisa is a bit of a sand bagging type and the share numbers just wasnt that great. As Mark says there is tons of conservatism. That's the name of the game. And there is also good reasons for that. As it's very high margin product even slightly lower than expected have some impact.
On the counter side I think desktop 3000 series takeoff especially for high end 3900 have been way above what you could expect looking at eg retailing in eu. The 3900 is the new black and I think 3950 will take that position in november. The effect can imo not be underestimated for future brand evolvement. It looks to me Intel simply lost the high end consumer grip in less than a quarter. That's crazy.

 

[TheELF]

When you look at the Fortune 50, most of the biggest customers (Alphabet, Microsoft, Amazon etc.) all use AMD products in the data center and consumer products.

They also all use intel products and the ratio is most probably not 1:1...
Of course it makes sense for them to use AMD in a lot of scenarios where their CPUs make sense to use but that doesn't mean that they don't use intel wherever that makes sense.

Back when I was new in the computer biz, the saying was "Nobody ever got fired for buying IBM." Took a while for that to change of course, but it did change irrevocably.

IBM gave intel all the designs and set them up they also made intel "share" them with AMD,that saying stopped to be about IBM because IBM decided to get out of that market.

Intel has nothing to compete against AMD in technical terms in HEDT and Enterprise. Sales of course is a different story but the momentum is clearly on AMD's side at the moment. We'll see how long they can keep it going.

Intel can provide datacenters and servers with ~6Tb that's terabyte of RAM,RAM and RAM that is persistent nonetheless,and with as many CPU cores combined with AI nervana cores as m.2 slots and power that they can provide.
AMD has no answer to that and won't have any...ever.
As I already said above that does not mean that EPYC sales will go down or anything, I do believe they will continue to grow just don't expect AMD to dominate any market but they will do well enough.

Intel Optane DIMM Pricing: $695 for 128GB, $2595 for 256GB, $7816 for 512GB (Update)

Intel's Optane DC Persistent DIMMs have landed at retail, giving us our first glimpse of pricing.

www.tomshardware.com

10nm Ice Lake CPU Meets M.2: The 'Spring Hill' Nervana NNP-I Deep Dive

Intel shares the fine-grained details about its 10nm Ice Lake chip on an M.2 card.

www.tomshardware.com

 

[coercitiv]

RAM,RAM and RAM

This deserves a special place in the AMD Q3 2019 Earnings Report thread.

Also, thank you @krumme for the nice summary of the current status quo.

 

[Asterox]

Guys. We are here to have fun.
Margins is exactly as predicted. Spot on.
Revenue and profit more or less the same. I think we can reserve the biggest celebrations, enthusiasm, slamming, trolling and hyperbole for another more interesting quarter.

This is purely speculation but if we look into the details I think Epyc1 have had a slighted lower share than they had hoped for perhaps due to it's more inconsistent performance for various loads. It naturally affects epyc2 adaptations because you start from a lower brand perception.
Lisa is a bit of a sand bagging type and the share numbers just wasnt that great. As Mark says there is tons of conservatism. That's the name of the game. And there is also good reasons for that. As it's very high margin product even slightly lower than expected have some impact.
On the counter side I think desktop 3000 series takeoff especially for high end 3900 have been way above what you could expect looking at eg retailing in eu. The 3900 is the new black and I think 3950 will take that position in november. The effect can imo not be underestimated for future brand evolvement. It looks to me Intel simply lost the high end consumer grip in less than a quarter. That's crazy.

The statistics are very accurate, or more precise for the for bigest Geman retailer Mindfactory.As far i now, only Mindfactory shows(über verkauft or sold over)the number of various sold hardwer parts.

- 80% sold CPU are from AMD

- by far Ryzen 5 2600 is best seling CPU on Mindfactory

- as expected best seling Ryzen 3000 is Ryzen 5 3600

AMD Desktop

Bei Mindfactory finden Sie Prozessoren von AMD zum günstigsten Preis: Prämiertes Shopportal | 22 Jahre Kompetenz | Paydirekt | Jetzt bestellen & sparen!

www.mindfactory.de

INTEL Desktop

Bei Mindfactory finden Sie Prozessoren von Intel zum günstigsten Preis: 24/7 Bestellmöglichkeit | Großes Produktsortiment | 100% SSL-Schutz | Online kaufen!

www.mindfactory.de

Do we need to ask, why no one else offers sales statistics in style as german Mindfactory retailer?Are sales statistics atomic physics, well yes and obviously only Mindfactory has atomic physics in the little finger.

 

[beginner99]

I worked in IT for 34 years, the last 15 as a supervisor. Yet you say its a fact that I know little about how IT works ? That NOT fact thats an opinion.

Sadly a lot has changed in the last 5-10 years. especially competence and knowledge going south sharply, at least in non-tech companies.

And of course you ignore the benchmarks where EPYC beat Intels best by 100% or more. And on average are like 80% more. And take half the power (which in the server world equates to 1/4th the total power consumption due to electricity costs). All of this you ignore, and says "they both run". One thing is winning a benchmark, when you totally dominate the benchmarks by a large margin, that when people use other terms for it.

But you don't get that.

Oh, and doing it for 1/2 the price.....

it assumes the guy making the decision actually reads benchmarks, hell that he knows what AMD is (and total server cost isn't half, far from it).

 

[Markfw]

Sadly a lot has changed in the last 5-10 years. especially competence and knowledge going south sharply, at least in non-tech companies.



it assumes the guy making the decision actually reads benchmarks, hell that he knows what AMD is (and total server cost isn't half, far from it).

First, I only retired 3 years ago, and I did, since nothing was getting better. And as for server cost ? 7k vs 13k ? Thats pretty close to half. And power usage ? thats in EPYC's favor also. In a datacenter, thats big, since power usage quite often means more or less AC, and that a big cost also.

 

[Hitman928]

In your own quotes you have Lisa Su saying that GPUs decreased sequentially. Also, from where are you getting 100m revenue for data center GPUs? AMD doesn't disclose revenue by market segment.

GPUs including the workstation and data center GPUs decreased sequentially and were lower driven by. . . lower data center GPU sales. Gaming GPU sales were up sequentially. So was Lisa Su lying about that?

The $100M+ revenue came by paying attention to the conference calls. Starting a few quarters ago they started to get asked how much revenue was due to data center sales and what the breakdown of that was for GPU vs CPU sales. If you do the math, it comes out to averaging about $100M+ per quarter for data center GPU sales up until this last quarter in which they basically just said that Epyc sales were up 50+% while data center GPU sales were down.

 

[jpiniero]

I don't believe AMD when they say they didn't benefit from Intel's shortages or any possible tariff pull ins. Definitely looks like Intel did on the tariff pull in.

Sadly a lot has changed in the last 5-10 years. especially competence and knowledge going south sharply, at least in non-tech companies.

Thing is, Cloud (and Telecom) is a big part of the server business now. You would think both segments would have competent people. Telecom might be interested in those secret x86 instructions Intel puts in the Xeons perhaps.

 

[DrMrLordX]

I get your point, buy you made a mistake to piss on AMD here...you cant do that, you can only piss only on Intel

Not true. In general, posters here go after companies for one or more of three reasons:

1). Personal vendettas
2). Product performance
3). Public perception (easily confused with #1; see contra revenue controversy and similar)

There was a stretch from about 2006 to 2017 where AMD was persona non grata around here, and honestly you didn't get much respect being a K6/K6-2/K6-3 fan either, back in the day. You should have seen how upset the AMD Zoners were about the mean, nasty, pro-Intel forum trolls. If Intel wants better treatment on these forums, they need to start competing again. Watching them continue to earn large quantities of money selling also-ran technology is slightly annoying. But that's how markets work, sometimes. Inertia is a thing.

antimonopoly rage

Old conspiracy, and still unsubstantiated.

this is pure incompetence by AMD to get such a results with products they have in current times

How? Why? Did you even look at the earnings reports? About the only thing that disappointed was their datacenter GPU sales. That may be more due to market saturation/cyclical sales fall-off than anything else.

My quote says "The unit as a whole is down 27%, so telling me they have a 50% increase is pointless. "

Glad you quoted yourself. This statement is truly inane.

AMD has chosen, for whatever reason, to lump their EPYC sales in with console sales. It's odd, and it permits people who want to knowingly sew misinformation to do so. AMD needs to take notes and (hopefully) improve their reporting in the future. 50% increase in EPYC sales is pointless? When AMD didn't even have EPYC2 available for the entire quarter? You must be mad! Growth like that is unprecedented. They're definitely selling more than 5000 or 10000 CPUs. Good grief.

AMD over the last few quarters has had 10 - 12% market share in data center GPUs with $100M+ in revenue. Not earth shattering by any means but they went from practically 0 to double digit market share pretty fast, it was a faster ramp than Epyc. They've mostly found success in cloud servers though a little in machine learning as well. They mentioned being down as a cyclical thing which is probably true.

They may have grown too quickly and encountered temporary market saturation of their own product. nVidia hasn't released anything recently in the datacenter that I'm aware of that would tamp down sales. Unless NV just dropped prices or something. In any case . . . time to sell more Radeon VII?!?!? (no it isn't, ha ha)

Telecom might be interested in those secret x86 instructions Intel puts in the Xeons perhaps.

Kind of makes me wonder if AMD offers the same thing, or plans to do so in the future?

 

[Hitman928]

Intel can provide datacenters and servers with ~6Tb that's terabyte of RAM,RAM and RAM that is persistent nonetheless,and with as many CPU cores combined with AI nervana cores as m.2 slots and power that they can provide.
AMD has no answer to that and won't have any...ever.
As I already said above that does not mean that EPYC sales will go down or anything, I do believe they will continue to grow just don't expect AMD to dominate any market but they will do well enough.

Intel Optane DIMM Pricing: $695 for 128GB, $2595 for 256GB, $7816 for 512GB (Update)

Intel's Optane DC Persistent DIMMs have landed at retail, giving us our first glimpse of pricing.

www.tomshardware.com

10nm Ice Lake CPU Meets M.2: The 'Spring Hill' Nervana NNP-I Deep Dive

Intel shares the fine-grained details about its 10nm Ice Lake chip on an M.2 card.

www.tomshardware.com

While I was speaking strictly about the processors themselves, as far as I'm aware the first gen optane products were a bit of a let down and even the 2nd gen has been losing money overall for intel. Optane is definitely intriguing but has some drawbacks as well, not the least of which being price. With AMD having a PCIe4 server platform and intel coming out with one soon, it will be interesting to see how the market takes to optane.

Same thing with the nervana cores, that's separate from the CPUs themselves. Those are more a competitor to Google and other inference chip makers as well as a little bit to Nvidia and AMD GPUs in the AI market.

 

[tamz_msc]

GPUs including the workstation and data center GPUs decreased sequentially and were lower driven by. . . lower data center GPU sales. Gaming GPU sales were up sequentially. So was Lisa Su lying about that?

The $100M+ revenue came by paying attention to the conference calls. Starting a few quarters ago they started to get asked how much revenue was due to data center sales and what the breakdown of that was for GPU vs CPU sales. If you do the math, it comes out to averaging about $100M+ per quarter for data center GPU sales up until this last quarter in which they basically just said that Epyc sales were up 50+% while data center GPU sales were down.

She didn't say gaming GPUs were up, just that they were doing good. She did not even hint at any kind of numbers. It is infuriating that AMD does not do a breakdown of their numbers by segment.

It is highly doubtful that data center GPUs bring in that much of a revenue for AMD that its decline is the primary driver of the decrease in overall GPU revenue, even though Navi launched in Q3.

 

[Hitman928]

She didn't say gaming GPUs were up, just that they were doing good. She did not even hint at any kind of numbers. It is infuriating that AMD does not do a breakdown of their numbers by segment.

It is highly doubtful that data center GPUs bring in that much of a revenue for AMD that its decline is the primary driver of the decrease in overall GPU revenue, even though Navi launched in Q3.

Again. . .

In graphics, revenue increased year-over-year, driven largely by higher channel GPU sales. Shipments of our Radeon 5000 GPU family featuring our RDNA architecture increased sequentially and we are seeing solid demand for the new products based on their competitive performance and features.

This was during her talking about the Computing and Graphics segment, i.e. consumer or gaming GPUs.

 

[tamz_msc]

Again. . .



This was during her talking about the Computing and Graphics segment, i.e. consumer or gaming GPUs.

I'm not talking about YoY, I'm talking about QoQ. YoY does not give any idea as to how Navi is performing.

If you look at GPUs overall, they actually declined a bit sequentially. And that decline was primarily driven by data center GPUs, which declined just due to some of the buying cycles in the cloud. Overall, gaming did well and we continue to expect that. As we go into the fourth quarter, you’ll see that the data center GPUs will increase, as well as I mentioned in the prepared remarks that client and graphics would also increase.

Also that statement regarding Radeon 5000 series increasing sequentially is a vacuous statement - of course it will increase sequentially - there was no Navi in Q2!

 

[Hitman928]

I'm not talking about YoY, I'm talking about QoQ. YoY does not give any idea as to how Navi is performing.

Y/Y is typically the more telling stat because GPU sales are historically very cyclical through the year. If you want numbers on how essentially a single model is performing, no one is going to tell you that with hard numbers, not Nvidia, AMD, or Intel. The best you will get are qualitative statements like AMD gave and then see if it makes sense in context of the overall numbers.

I went back and looked at AMD's Q2 report and they said that desktop GPU sales were down in Q2 but mobile was up and that they expected growth in GPU sales in the second half of the year due to additional ramping of Navi GPUs. Then in this quarter they said that Navi was doing well and GPU sales contributed to the increase in revenue offsetting the decline in Semi-Custom.

That's the best you're going to get at this point in terms of Navi market performance. You either believe it or you think Lisa Su is lying to company investors.

Also that statement regarding Radeon 5000 series increasing sequentially is a vacuous statement - of course it will increase sequentially - there was no Navi in Q2!

AMD counted the initial Navi stock as sales in Q2 since they had to have the GPUs through the channel and in the hands of retailers in Q2 in order to have them ready to actually be sold at the beginning of Q3. However, I agree it's not very telling that it sold more through an entire quarter versus a small slice of Q2.

 

[TheELF]

This deserves a special place in the AMD Q3 2019 Earnings Report thread.

Yes it does because intel prefers to sell 512Gb dimms for $8k instead of selling 64 cores for $8k and that's what will allow amd to get more sales in the server market.

While I was speaking strictly about the processors themselves, as far as I'm aware the first gen optane products were a bit of a let down and even the 2nd gen has been losing money overall for intel. Optane is definitely intriguing but has some drawbacks as well, not the least of which being price. With AMD having a PCIe4 server platform and intel coming out with one soon, it will be interesting to see how the market takes to optane.

Same thing with the nervana cores, that's separate from the CPUs themselves. Those are more a competitor to Google and other inference chip makers as well as a little bit to Nvidia and AMD GPUs in the AI market.

Optane as well as nervana is only worth it for certain scenarios that's for sure but then again so are server CPUs in general.
I mean intel had xeon phi years ago with 72c/288t for $6294
and even that didn't do all that well financially, or did it? I mean we barely ever hear about them or see any posts about them so I have no idea.

Read more: https://www.tweaktown.com/news/52716/intels-xeon-phi-7290-expensive-chip-72-cpu-cores/index.html

 

[Hitman928]

Yes it does because intel prefers to sell 512Gb dimms for $8k instead of selling 64 cores for $8k and that's what will allow amd to get more sales in the server market.

Optane as well as nervana is only worth it for certain scenarios that's for sure but then again so are server CPUs in general.
I mean intel had xeon phi years ago with 72c/288t for $6294
and even that didn't do all that well financially, or did it? I mean we barely ever hear about them or see any posts about them so I have no idea.

Read more: https://www.tweaktown.com/news/52716/intels-xeon-phi-7290-expensive-chip-72-cpu-cores/index.html

Xeon Phi was an absolute failure both financially and technically. There's a reason why it's been discontinued and Intel is developing its own GPUs instead.

Nervana looks good but I think the main problem is that the people really interested in what it can do and have the means to deploy it are already developing their own hardware to do the same thing (e.g. Google). I don't know how that whole thing is going to turn out but it will be very interesting to follow.

Optane is a wild card. It has some great potential benefits, but in non-DIMM form there will be competition with claims of greatly reduced latency (https://blocksandfiles.com/2019/07/01/sk-hynix-storage-class-memory/) and in DIMM form you are locked into Intel CPUs that have the proper memory controller support to be compatible which creates all kinds of headaches on the server vendor and customer side not to mention it is significantly more expensive than industry standard storage solutions. Then you also have increasingly fast storage options every year that are only going to improve with PCIe4 to work with. So then it becomes a question of is something like Optane really needed/worth it? I don't know, all I know is right now Optane is having a hard time finding customers but we'll see what Intel can do with it.

 

[beginner99]

all I know is right now Optane is having a hard time finding customers

I think optane and nervana and xeon phi all are niche products and hence not sold in big enough volume to make a profit. I'm sure some people need that amount of none-volatile RAM but how many companies actually run terabytes worth of in memory databases? Same for the other two. In contrast to phi, a normal gpu can be used well to be a gpu for consumers or pro users. hence the development cost is split over much higher volume, theoretically at least if the manage to deliver.

AFAIK phi suffered from the issue that to make use of the power you needed the software which meant adjusting or rewriting it. Much simpler to just take a NV gpu with CUDA were that stuff already exists.

 

[TheELF]

Xeon Phi was an absolute failure both financially and technically. There's a reason why it's been discontinued and Intel is developing its own GPUs instead.

Nervana looks good but I think the main problem is that the people really interested in what it can do and have the means to deploy it are already developing their own hardware to do the same thing (e.g. Google). I don't know how that whole thing is going to turn out but it will be very interesting to follow.

Optane is a wild card. It has some great potential benefits, but in non-DIMM form there will be competition with claims of greatly reduced latency (https://blocksandfiles.com/2019/07/01/sk-hynix-storage-class-memory/) and in DIMM form you are locked into Intel CPUs that have the proper memory controller support to be compatible which creates all kinds of headaches on the server vendor and customer side not to mention it is significantly more expensive than industry standard storage solutions. Then you also have increasingly fast storage options every year that are only going to improve with PCIe4 to work with. So then it becomes a question of is something like Optane really needed/worth it? I don't know, all I know is right now Optane is having a hard time finding customers but we'll see what Intel can do with it.

Everybody is interested in AI training and being able to add additional nervana (and normal) cores to whatever AI solution you already have on your system is quite a number.If it's easy to integrate into already deployed systems it's going to do well at least.
Optane is not just storage it's memory and compared to memory it's super cheap especially when you get up to capacities that are impossible to do on ram and you need the special intel CPUs to actually see them as memory,when I talk about memory I mean that the CPU will be able to directly take and put back data from and to the optane ram without having to go through the normal loops of getting it from storage first copying it to ram and only then being able to work on it,it's a hughe speed up for whoever can need it.

I think optane and nervana and xeon phi all are niche products and hence not sold in big enough volume to make a profit. I'm sure some people need that amount of none-volatile RAM but how many companies actually run terabytes worth of in memory databases? Same for the other two. In contrast to phi, a normal gpu can be used well to be a gpu for consumers or pro users. hence the development cost is split over much higher volume, theoretically at least if the manage to deliver.

AFAIK phi suffered from the issue that to make use of the power you needed the software which meant adjusting or rewriting it. Much simpler to just take a NV gpu with CUDA were that stuff already exists.

Well that's why there are also smaller optane dimms available,you get as much as you need.
Phi is/was x86 ,anything multithreaded scales on it,you just have to "compensate" (disable) for the quad hyperthreading which will cause quite some slow down for most distributed computing workloads.

 

[JPB]

Netflix considers switching from Intel to AMD EPYC in its datacenters

 

[DrMrLordX]

@JPB

Oof, that's got to hurt Intel, if it actually happens.

 

[Gideon]

@JPB

Oof, that's got to hurt Intel, if it actually happens.

Yes, though caution is advised asa this is classic rumor-mill journalism. The only "evidence" they have are a few slides from this presentation (which in itself is really interesting). There a Netflix Employee mentions the struggles they had getting 200 Gb/s per server streaming to work in NUMA configurations. He tested it out on both Cascade Lake and Rome. It's true that Rome gets it done with one socket (Intel needs 2), but he never mentioned anything specific about replacing Intel.

I'm pretty sure they will start running some AMD servers in tandem with intel, but it's far from certain that they'll replace their entire stack

 

[JPB]

The benchmarks Intel doesn't want you to see - TweakTown

 

[Markfw]

So the 16 core beats the 18 core Intel...Seems right to me....