News Intel Posts Record Earnings, Desktop PC Volumes Down 10 Percent YoY, First DG1 Discrete GPU Power-On

kinggremlin

Distinguished
Jul 14, 2009
574
41
19,010
These quarterly results seem to indicate Intel is bulletproof. Intel's price cuts haven't really taken hold yet so that impact wasn't seen in the Q3 results, but I certainly expected to see some more significant negative impact on Intel's bottom line from the AMD resurgence. However, it is a bizarre situation that Intel just posted a record revenue quarter, and is continuing to be unable to meet demand yet they plan to roll out massive price cuts starting this quarter. Not sure how that is supposed to improve product availability. It's like they are dropping prices just to avoid getting insulted by internet trolls, even though they are selling everything they can at the current inflated prices.
With reports of TMSC struggling with 7nm demand, it really makes AMD look like a bunch of morons. If Intel can't meet demand at their prices, and AMD's primary fab is struggling to meet demand as well, with sporadic outages of Zen 2 CPU's, why on earth would they price their CPU's so far below Intel? Sure it sort of benefits the consumer in the short term (Intel's announced price cuts in the mainstream are highly underwhelming), but how many millions is AMD leaving on the table that could fund R&D for future generations and benefit us for the long term? AMD could easily charge 25% more for their Zen2 CPU's and not see any drop in volume. AMD can't ever seem to get pricing right. Either they charge too much for garbage (their video cards now), or they charge too little for their good products (CPU's now).
 

InvalidError

Titan
Moderator
These quarterly results seem to indicate Intel is bulletproof.
Intel isn't bulletproof. Corporate sales are the bulk of Intel's business and corporate buyers have very high latency with adopting unknown parts from new vendors. AMD knew from the beginning that it'd be about three years before Zen got significant traction in the corporate world beyond early adopters. By the same token, Intel knew the competitive threat would be limited for the first couple of years. 2020 is where things may heat up for real should supply on both sides allow.
 
  • Like
Reactions: bit_user

twotwotwo

Reputable
Aug 10, 2019
47
18
4,535
Agree with the first poster that Intel can remain very healthy through a lot. Good to see the boss acknowledge some of the tricky things--price pressure and currently lacking process leadership.

Still a big unanswered question for me is how the rest of the 10nm rollout will go. If this were a typical rollout, the 10th gen would be Ice Lake: there would be a plentiful enough supply of the chips, and they'd be a clear step up from the 14nm chip for everybody, instead of this productivity/graphics split they're talking about.

The 14nm shortages are also confusing. Did they have to convert a bunch of 14nm capacity to 10 to get this far on Ice? Some mysterious issue with their operations/execution they're keeping totally silent about? I'd also offer bad market forecasting as a guess, but shortages have gone on too long for it to just be forecasting.

I'm suspicious, with the unusual rollout of 10nm so far, that high volumes of high-end desktop/server 10nm CPUs will actually happen closer to the launch of 7nm+ Zen 3 than Zen 2. By mid-2021, maybe they're against 5nm and another new core design. Does not feel like Intel puts this behind them with their next big product launch anyway.

The dGPU stuff is interesting, also because it goes after one of AMD's other sources of profit. A sort of analogous thing on AMD's side is the (probable) 7nm Renoir laptop chips next year. They probably won't match Intel since AMD never got as good at power management; what I wonder is if they can get close enough to put pressure on Intel's prices. I'm sure each company's happy to take a shot at chipping away at the other's "safe" businesses.
 

Mileta Cekovic

Honorable
Jun 3, 2013
3
0
10,510
These quarterly results seem to indicate Intel is bulletproof.
Actually these quarterly results indicate only that Intel is selling overpriced processors to an non-educated market :(. But as much as these results are financially good, they do not show that things will change unless Intel reacts quickly, as customers will start to understand who is offering better value. It will take some time though, but customers will understand at the end of the day.
It is just a matter of:
  • How long AMD is able to maintain technical leadership (10 years ago AMD could not maintain it longer then 1-2 years with Athlon 64 and that was not enough).
  • Will Intel respond quickly enough and establish at least technical parity with AMD.
 
Intel isn't bulletproof. Corporate sales are the bulk of Intel's business and corporate buyers have very high latency with adopting unknown parts from new vendors. AMD knew from the beginning that it'd be about three years before Zen got significant traction in the corporate world beyond early adopters. By the same token, Intel knew the competitive threat would be limited for the first couple of years. 2020 is where things may heat up for real should supply on both sides allow.

Agreed. Intel's shrinking margin is a sign that pressure is on. Shrinking margin is something you try to avoid as a CEO because it tends to make investors panic. So they will often implement tactics like layoffs, and sales of unprofitable assets to boost the bottom line and make things look better than they are to keep investors from leaving.
 

kinggremlin

Distinguished
Jul 14, 2009
574
41
19,010
Actually these quarterly results indicate only that Intel is selling overpriced processors to an non-educated market :(. But as much as these results are financially good, they do not show that things will change unless Intel reacts quickly, as customers will start to understand who is offering better value. It will take some time though, but customers will understand at the end of the day.
It is just a matter of:
  • How long AMD is able to maintain technical leadership (10 years ago AMD could not maintain it longer then 1-2 years with Athlon 64 and that was not enough).
  • Will Intel respond quickly enough and establish at least technical parity with AMD.
Customers won't understand because they don't care and don't buy CPU's. They buy Intel systems from Dell or HP because they know the brand. The vast majority probably don't even know that Intel makes the CPU. Computers are an appliance to the general public and they don't feel like wasting time researching them.
 

InvalidError

Titan
Moderator
Computers are an appliance to the general public and they don't feel like wasting time researching them.
For a large chunk of the general public, researching PCs really is a waste of time since any potato from recent years will do just fine... except perhaps for those countless poorly optimized FB games that no amount of over-building can help. (My mother complained continuously about laggy FB games on her AMD E-series laptop, upgraded to a Ryzen 2500U which is 3-4X faster, still complains about FB game lag just as much.)
 

bit_user

Polypheme
Ambassador
The 14nm shortages are also confusing. Did they have to convert a bunch of 14nm capacity to 10 to get this far on Ice? Some mysterious issue with their operations/execution they're keeping totally silent about? I'd also offer bad market forecasting as a guess, but shortages have gone on too long for it to just be forecasting.
Fab capacity is neither cheap nor quick to add. So, even if you realize you mis-forecast demand, then you have to be really sure that you were really far off and that it will continue to hold true, to then go and justify adding capacity of an old node fairly late in the cycle. Maybe they lost further time, as that reckoning slowly sank in. Plus, don't forget they had some shakeup in their management, over the past couple years.

The other thing is that because Intel got stuck on 14 nm and AMD is churning out CPUs with way more cores, Intel had to add more cores per CPU, on 14 nm, to stay competitive. This means fewer chips per wafer, so their existing capacity is no longer good for as much volume as before. So, it's kind of like a double-hit to their capacity.
 
Oct 27, 2019
1
0
10
Intel acts like a mafia. bribes here, bribery there, hype there, poaching from competitors, price manipulation, conspiracy with sellers.
what's the rush? what's that noise? everything is going well. AMD will return its investments and market' shares.


  • Dell EMC PowerEdge R6515 and R7515 servers are available worldwide now.
  • Dell EMC PowerEdge R6525 and C6525 servers have planned availability worldwide in October 2019.
  • Dell EMC PowerEdge R7525 servers have planned availability in early 2020.
  • Dell EMC OpenManage advancements are now available worldwide.
  • Dell EMC Ready Solutions for HPC Digital Manufacturing, Research and Life Science, with 2nd Gen AMD EPYC-based PowerEdge servers, will be available worldwide in October 2019.

For more information about how these new servers will be used it is worth a visit to the Direct2DellEMC blog.
 
The 14nm shortages are also confusing. Did they have to convert a bunch of 14nm capacity to 10 to get this far on Ice? Some mysterious issue with their operations/execution they're keeping totally silent about? I'd also offer bad market forecasting as a guess, but shortages have gone on too long for it to just be forecasting.
What's confusing about it? Intel has currently 4 generations of desktop CPUs all running on 14nm and three of them are still actively sold (so obviously also manufactured still) by intel. Plus whatever laptop CPUs are still on 14nm.
I'm suspicious, with the unusual rollout of 10nm so far, that high volumes of high-end desktop/server 10nm CPUs will actually happen closer to the launch of 7nm+ Zen 3 than Zen 2. By mid-2021, maybe they're against 5nm and another new core design. Does not feel like Intel puts this behind them with their next big product launch anyway.
intel is already selling humongous amounts of 10nm to servers.
From the opening link.
The impressive results were fed by a record $9.5 billion in revenue for Intel's data center group, which now comprises 49% of the company's revenue (another record).
This is what intel is selling,AI coprocessors on m.2 and optane dimms,for 8k a pop.
AMD is trying real hard and their products are very good but how do you compete with a single CPU that only has access to normal ram and normal amounts of ram against a cluster of CPUs with hardware acceleration for AI that have access to never before heard of amounts of ram and ram they have direct access to on top of that?
https://www.tomshardware.com/news/i...network-processor-ai-inference-m.2,40204.html
https://www.tomshardware.com/news/intel-optane-dimm-pricing-performance,39007.html
 

bit_user

Polypheme
Ambassador
  • Dell EMC PowerEdge R6515 and R7515 servers are available worldwide now.
  • Dell EMC PowerEdge R6525 and C6525 servers have planned availability worldwide in October 2019.
  • Dell EMC PowerEdge R7525 servers have planned availability in early 2020.
  • Dell EMC OpenManage advancements are now available worldwide.
  • Dell EMC Ready Solutions for HPC Digital Manufacturing, Research and Life Science, with 2nd Gen AMD EPYC-based PowerEdge servers, will be available worldwide in October 2019.
For more information about how these new servers will be used it is worth a visit to the Direct2DellEMC blog.
I had a peek at that page, but it wasn't immediately clear to me which of these servers would support GPU compute.

A couple months ago, I was looking for a GPU compute server from Dell, and was disappointed to find their Epyc-based servers did not offer that option. They were clearly aimed at storage.

So far, Dell seems eager to milk anyone doing GPU compute for all they're worth. Simply prepping their servers for it adds many $k of added costs, before you even add one card. In the end, we opted for a dual-Xeon workstation, though I wish a comparable Epyc-based alternative were available.
 

bit_user

Polypheme
Ambassador
intel is already selling humongous amounts of 10nm to servers.
Really? Not in the form of general-purpose CPUs, at least.

This is what intel is selling,AI coprocessors on m.2 and optane dimms,for 8k a pop.
AMD is trying real hard and their products are very good but how do you compete with a single CPU that only has access to normal ram and normal amounts of ram
IIRC, Epyc CPUs support 2 TB of RAM, each. So, you can configure a dual-CPU server with 4 TB. However, I think that's probably rather niche. You talk about extremes, but AMD is focused on the average use case, which is probably the overwhelming majority of the market.

And, as for advantages, an Epyc CPU packs 128 lanes of PCIe 4.0. Intel (currently) has no answer for that.

against a cluster of CPUs with hardware acceleration for AI that have access to never before heard of amounts of ram and ram they have direct access to on top of that?
https://www.tomshardware.com/news/i...network-processor-ai-inference-m.2,40204.html
According to that article, those M.2's are limited to 32 GB. Also, they don't have much memory bandwidth, for an AI accelerator (stated: 68 GB/sec). AMD and Nvidia's GPUs are in the 1 TB/sec ballpark, also with 32 GB.

All of the mentioned accelerators have access to more, on the other side of the PCIe connection. AMD and Nvidia have fast multi-GPU interconnects, as well.
 
  • Like
Reactions: TJ Hooker
Really? Not in the form of general-purpose CPUs, at least.
Doesn't change the amount of money intel is making off of them.
IIRC, Epyc CPUs support 2 TB of RAM, each. So, you can configure a dual-CPU server with 4 TB. However, I think that's probably rather niche. You talk about extremes, but AMD is focused on the average use case, which is probably the overwhelming majority of the market.
How many dimm slots do you need to reach that with real mem?
Is there even any board out there that could do that?
Also datacenters...how can they not need more mem?
If it is niche right now it's because it's impossible to do...without optane.
And, as for advantages, an Epyc CPU packs 128 lanes of PCIe 4.0. Intel (currently) has no answer for that.
How do PCI lanes leverage all the other things?
What is the scenario where 128 PCI lanes actually are useful?
According to that article, those M.2's are limited to 32 GB. Also, they don't have much memory bandwidth, for an AI accelerator (stated: 68 GB/sec). AMD and Nvidia's GPUs are in the 1 TB/sec ballpark, also with 32 GB.

All of the mentioned accelerators have access to more, on the other side of the PCIe connection. AMD and Nvidia have fast multi-GPU interconnects, as well.
That's the beauty with DC,if your conventional dGPU accelerators aren't enough anymore you can stick additional ones into m.2 without having to scrap your whole system or wait for the next gen of dGPU.
Also if they are working on their own dataset they don't need any huge bandwidth,plus they come with extra CPU cores as well,which no GPU does.