Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/amd-taking-server-market-from-intel-many-questions.16529/page-2
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021071
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

AMD Taking Server Market From Intel, Many Questions

tooLongInEDA

Active member
There is a challenge that Intel is operating under the IDM + Foundry business model. From software to hardware, from design to manufacturing, from desktop CPU to server CPU, from AI/ML to high performance computing, from inhouse clients to external foundry customers, Intel is running too many things and pursuing too many targets, IMO.

It is "Everything Everywhere All at Once".

Can Intel cut something out in order to be more focused?
Agree with the general sentiment on doing too much.

But are they actually going the other way and doubling down on vertical integration (with increased foundry) ? Though arguably with more outsourcing to TSMC there is some movement the other way.

If we think of this as a supply chain, Intel has to complete with the fabless model where Apple and some others are arguably stronger in design/project management - let's rate those guys 95% and say that Intel is perhaps at 90% today. The same applies on the wafer fab side - TSMC is certainly ahead on a "combined foundry metric" (assume we can weight and lump into a single number performance, area, cost/yield, service level, etc). Unless there is some inherent offsetting efficiency gain from keeping everything in house (and is there ?) you end up slightly worse off. [This is probably a poor rehash of Dan's book - which I really should read].

This is all rather like the arguments for and against in-house EDA vs assembling a best in class third party EDA toolchain - which were resolved in favour of third party EDA a generation ago (with Intel late to the party here in some areas - as were IBM). That's one reason why bringing Lip-Bu Tan (from Cadence) onto the Intel board is interesting - it's getting someone in who gets the non-vertical industry model and understands how to build those partnerships and relationships. Lip-Bu is also interesting in having the much less brash and sometimes almost humble leadership style you sometimes find in Asia.
 

blueone

Active member
Can Intel cut something out in order to be more focused?
Excellent question.

Intel already cut Optane and SSDs, and they're trying to divest Mobileye (however incompetently). I agree with all three decisions.

After touting $3.5B in AI/ML revenue in 2019 they've gone silent on AI. Habana has the Gaudi2 SoC on their website, fab'd on TSMC N7 process it looks like. Since Intel appears to have gone silent on AI revenues, I assume there's nothing good to say, and it fascinates me that Habana products are not listed on the Intel.com product page. Interesting, eh? Perhaps the untold strategy is follow Nvidia into GPU-based AI/ML, and dump Habana, after they already dumped Nervana. AI is one of those fields where you absolutely must have a good understanding of the application problem space to judge processor products, and I admit I'm lacking in that area.
 
Last edited:

Tanj

Active member
There is a challenge that Intel is operating under the IDM + Foundry business model. From software to hardware, from design to manufacturing, from desktop CPU to server CPU, from AI/ML to high performance computing, from inhouse clients to external foundry customers, Intel is running too many things and pursuing too many targets, IMO.

It is "Everything Everywhere All at Once".

Can Intel cut something out in order to be more focused?
The foundry needs to split from the chip design side. They can't keep being tied together running a 3-legged race on a field of sprinters. The foundry needs an existential understanding of other customers, and the design house needs access to every foundry. The race is to get them standing on their feet before launch on their separate paths.
 

blueone

Active member
The foundry needs to split from the chip design side. They can't keep being tied together running a 3-legged race on a field of sprinters. The foundry needs an existential understanding of other customers, and the design house needs access to every foundry. The race is to get them standing on their feet before launch on their separate paths.
The design groups already have access to other foundries (exclusively TSMC, as far as I know) for various products. FPGAs, GPUs, networking chips, AI chips. Aren't the CPU teams already signed up with TSMC for selected chiplets in 3nm process?
 

Tanj

Active member
The design groups already have access to other foundries (exclusively TSMC, as far as I know) for various products. FPGAs, GPUs, networking chips, AI chips. Aren't the CPU teams already signed up with TSMC for selected chiplets in 3nm process?
They still have mutual goals regarding their most important chips. Yes, this is a step and a practice for what is to come. But they are not yet fully free to make their best choices. Let's hope they launch each other, not hold back.
 
Top