Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/what-are-the-odds-that-meteor-lake-cpu-tile-will-be-fabricated-by-tsmc.16348/
        )

    [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] => 2021370
            [XFI] => 1050270
        )

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

What Are The Odds That Meteor Lake CPU Tile Will Be Fabricated By TSMC?

Unless Intel has a separate team working on the CPU tile with TSMC's design kit in parallel to Intel's own, it's hard to believe Intel can switch from Intel manufacturing to TSMC fab in such short time frame.

On the other hand, Intel doesn't have enough and right products that the market needs. Intel missed the historical semiconductor market boon across all leading edge and mature node products that almost every semiconductor company has been enjoyed. Intel's flat 2021 revenue growth and shrinking 2022 revenue is a strong indication that Intel desperately needs right products to sell.

Without the right products to offer, Intel can slip into a nice but second tier product company. It's more serious than if Intel can regain leadership in manufacturing. Apple already left Intel, Microsoft, Google, Amazon, and many other traditional Intel customers are doing their own inhouse semiconductor projects or buying products from Intel's competitors. Intel's customers can't wait and won't wait for Intel.

Is the Meteor Lake product line competitive? If the answer is yes, Intel should do everything to deliver it on time, including outsource CPU tile to TSMC.
 
Last edited:
Is the Meteor Lake product line competitive? If the answer is yes, Intel should do everything to deliver it on time, including outsource CPU tile to TSMC.

Pat Gelsinger has financial fiduciary responsibilities, and if Intel’s fabrication problems will result in significant financial distress, and if Pat has a reasonable fabrication alternative, such as TSMC, Pat is dutybound to procure such alternatives.
 
As an IDM it is more difficult to gauge capacity requirements. If I were Pat I would plan for the worst and hope for the best. Hopefully Pat has CPU teams designing at both Intel and TSMC to make sure they have the capacity available in case of emergencies. That is the beauty of tiles and chiplets. Less susceptible to yield ramping problems and you can multi source.
 
My understanding is that Meteor Lake is Intel’s first attempt to put many different tiles (CPUs, GPUs, memory,..) together, and to challenge other HPC competitors in a meaningful way. It is a new ball game even for Intel. Many things including process, packaging, testing, software driver, and motherboard reference design need to be ready in time. If Pat can pull everything together really quickly, he is a magician.

According to a last year’s UDN report (see the thread), Intel has completed 3 CPUs and 1 GPU engineering runs on TSMC N3. It is very reasonable to assume that one of those CPUs is for Meteor Lake.

 
As an IDM it is more difficult to gauge capacity requirements. If I were Pat I would plan for the worst and hope for the best. Hopefully Pat has CPU teams designing at both Intel and TSMC to make sure they have the capacity available in case of emergencies. That is the beauty of tiles and chiplets. Less susceptible to yield ramping problems and you can multi source.

Odds?
 
Unless Intel has a separate team working on the CPU tile with TSMC's design kit in parallel to Intel's own, it's hard to believe Intel can switch from Intel manufacturing to TSMC fab in such short time frame.

On the other hand, Intel doesn't have enough and right products that the market needs. Intel missed the historical semiconductor market boon across all leading edge and mature node products that almost every semiconductor company has been enjoyed. Intel's flat 2021 revenue growth and shrinking 2022 revenue is a strong indication that Intel desperately needs right products to sell.

Without the right products to offer, Intel can slip into a nice but second tier product company. It's more serious than if Intel can regain leadership in manufacturing. Apple already left Intel, Microsoft, Google, Amazon, and many other traditional Intel customers are doing their own inhouse semiconductor projects or buying products from Intel's competitors. Intel's customers can't wait and won't wait for Intel.

Is the Meteor Lake product line competitive? If the answer is yes, Intel should do everything to deliver it on time, including outsource CPU tile to TSMC.

It may well be TSMC, but I don't see in any way how it will help Intel to stay relevant.

Sales of desktop + notebook CPUs is their fundamental revenue source, with Xeon always being a demand driven market, and dependent on "free" RnD from desktop side coming.

Many-core "cloud" CPUs flatlined, and regular hosting grade ones are back to becoming kings again. It's almost as if the cloud era is being undone in just a year.
 
UDN: https://udn.com/news/story/7240/5662232
"The supply chain pointed out that the products that Intel gave to TSMC this time, including a graphics chip and three server processors, are all Intel core products. The finalized products will be moved to Nanke 18b for mass production; these four products will be officially produced and delivered in May next year, and will be mass-produced in July next year. A year earlier."

I'll just speculate a little:

Intel execution, < 14nm, has been poor and slow
Intel 4 HVM1 seems somewhat late to arrive (Ireland)
Indications Intel has a good deal with TSMC, that might incentivize doing a CPU tile on 3nm "just to see" (kind of like Qualcomm recent Snapdragon comparison with TSMC vs. Samsung)

So, percentage-wise, I think there's a 50% chance a 3nm CPU tile on TSMC exists, or will exist, as an experiment.
 
As an IDM it is more difficult to gauge capacity requirements. If I were Pat I would plan for the worst and hope for the best. Hopefully Pat has CPU teams designing at both Intel and TSMC to make sure they have the capacity available in case of emergencies. That is the beauty of tiles and chiplets. Less susceptible to yield ramping problems and you can multi source.

"As an IDM it is more difficult to gauge capacity requirements."

After so many years in business Intel's design/sales/marketing divisions should have the similar or better capability than fabless companies in forecasting market demand. Why Intel's manufacturing division or Intel as a whole is having difficulty to guage capacity requirements?
 
Last edited:
Also for Pat visiting TMSC in Aug .

Scenario A: Intel have too much 12th Gen CPU so would like to delay 14th Gen CPU.
This is unlikely. Intel 4 is a milestone for Intel and Pat. It's more important than inventory control. Also Intel can choose to produce less 13th Gen CPU.

Scenario B: Meteor Lake CPU Tile Will Be Fabricated By TSMC.
However, the rumor is Intel Taiwan is holding previous mass recruitment. Not sure it's related to Intel previous announcement of freezing all hiring in its Client Computing Group (CCG).
If Intel give CPU Tile to TSMC, Intel should be hiring more people in Taiwan instead of freezing.

Scenario C: No delay. Just scale down.
This is also possible.
 
Honestly the odds that the CPU Tile is fabricated by TSMC is almost 0.
1. Intel would on more then one occation lied to investers
2. This would indicate additional volume that was not planned
3. The time from tape-out to production for intel is more then 12 month

Whats likely happend is the following.
Everything on intels plan is basically delayed
SPR --> GA Q2/Q3 2023
Raptorlake Desktop --> Dec 22
Raptorlake Notebook --> Q2/23
Intel ARC (TSMC 6N) --> China only

The problem is that especially Raptorlake is launching to late. Makes no sense to launch Raptor and Meteor inbetween 3-4month
 
Honestly the odds that the CPU Tile is fabricated by TSMC is almost 0.
1. Intel would on more then one occation lied to investers
2. This would indicate additional volume that was not planned
3. The time from tape-out to production for intel is more then 12 month

Whats likely happend is the following.
Everything on intels plan is basically delayed
SPR --> GA Q2/Q3 2023
Raptorlake Desktop --> Dec 22
Raptorlake Notebook --> Q2/23
Intel ARC (TSMC 6N) --> China only

The problem is that especially Raptorlake is launching to late. Makes no sense to launch Raptor and Meteor inbetween 3-4month
Tigerlake mobile sold for a while even alongside alderlake mobile. I wouldn't read too much into that. They're on different nodes and different architectures so they could launch at nearly the same time and focus on different segments.
 
Tigerlake mobile sold for a while even alongside alderlake mobile. I wouldn't read too much into that. They're on different nodes and different architectures so they could launch at nearly the same time and focus on different segments.
To add on to this. Cometlake and icelake sold concurrently. With icelake only really being found in premium thin and lights, while cometlake was found in the cheaper/larger laptops/desktop.
 
Back
Top