Intel CEO Pat Gelsinger's missteps have strained TSMC partnership, hindered turnaround efforts

[VemigoAmigo]

Bulldozer was just a flat out bad design. Very low IPC, very high cache latencies, and just all around not well thought out architecture. Heck, if anything the fabs did a wonderous job, given AMD managed to get close to 5GHz on the thing despite its architectural problems.

FX-9590 boosted to 5 GHz but yeah

Worlds first 5 GHz chip, sadly performance was terrible and AMD was hit with a lawsuit for faking core count on FX-8000/9000 series. FP and scheduler was shared.

 

[gamerk2]

FX-9590 boosted to 5 GHz but yeah

Worlds first 5 GHz chip, sadly performance was terrible and AMD was hit with a lawsuit for faking core count on FX-8000/9000 series. FP and scheduler was shared.

Over at Toms we had ENDLESS discussions on BDs design (I hated it). I remember when the first cache access times were leaked; my exact words: "They must be fake; it impossible access times are *that* bad".

They weren't fake.

 

[maxxcool7421]

AMD also showed, the hard way, that depending on external fabs can be disastrous. Does nobody else remember Bulldozer? Owning their own fabs allowed Intel to dominate, both on the market and at the bank, for 30+ years. Even now their per chip cost is significantly lower than AMDs since there is no middle man taking a sizeable cut.

that's before adding in restrictions. There is only so much TSMC available. AMD knows this, because their mobile lines and GPUs are hamstrung by it. Intel dominates in mobile, in no small part, because they can supply the tens of millions of chips necessary. AMD cant.

What intel really needs is to utterly gut their worthless boated middle management, and the leadership, and install some younger, more driven, engineers. the bloat at Intel is absolutely insane, and their corporate culture is one that drives away innovation. Fab or not, that is a recipe for disaster.

Bulldozer was 100% AMD design faults

 

[HardReset]

Intel did not have supply issues at all. They focus alot on mobile market, as laptops is way more profitable than desktop and Intel has like 90-95% x86 laptop marketshare.

Back during COVID lockdowns, Intel delivered millions of chips and AMD struggled hard with capacity at TSMC, meaning AMD lost out on alot of sales here. Intel could deliver, AMD could not.

Also, AMD mobile is a fragmented mess.

Intel acknowledges supply issues, will prioritize premium chips | TechCrunch

Intel interim CEO Bob Swan issued an uncharacteristically frank letter today, highlighting the company’s supply issues. The executive blames the

techcrunch.com

Intel admits 'tight' supply for cheap PC chips, focuses on high-end

In an open letter, interim Intel CEO Bob Swan admitted the company's supply of CPUs for the "entry level" PC market is "undoubtedly tight," so if you have trouble finding a cheap laptop for the holiday season then you know why. Between consumers upgrading their PCs, gaming and commercial systems...

www.engadget.com

https://www.cnbc.com/2018/09/14/jp-morgan-says-intels-chip-shortage-is-worsening-and-will-hurt-pc-sales.html

PC makers: Intel CPU shortages are here to stay ... for six months

If it's not one DRAM thing, it's another

www.theregister.com

Intel Cutting Retail Processor Supply for Holiday 2018

Prices of retail packages of Intel Core desktop processors could continue to rise over Q4-2018, as the company has reportedly cut their supply, in favor of tray/reel shipments to OEMs. This could mean DIY favorites such as the Core i5-8400, the i5-8600K, i5-9600K, or even Core i7 models such as...

www.techpowerup.com

Intel Chip Shortage, In Its Second Year, Hits Revenue Of No. 5 PC Maker Acer

Acer says Intel's CPU shortage since 2018 sent its revenues 13.1% lower in February over January. The shortage is still unfixed, analysts say.

www.forbes.com

Intel Humiliated by Losing CPU Market Share to Eager AMD – TNN

Interested in Intel Humiliated by Losing CPU Market Share to Eager AMD? Click here to find out! ⭐ TNN offers IT, Phone, and Copier Service and Support for California Businesses.

www.tnnsupport.com

 

[guillejp]

What makes good engineering sense tends to make little financial sense. There's a reason why focusing on short-term gains instead of long-term engineering sends stock prices to the moon. For a decade or two.

Yes, they need both angles, but definitely needed more of the engineering side at the head. Again, maybe Pat wasn't the ideal candidate to bring that in.

 

[Dimitrios]

Does he still have that "rearview mirror" or did INTEL have to pawn it?

 

[VemigoAmigo]

Intel will be leaving TSMC faster than they used them anyway. It was a temp solution.

TSMC 3nm helped lower power with Lunar Lake and Arrow Lake but clockspeeds are much lower than Intels own nodes. Hence the 5% gaming regression.

18A is still the most important node in Intel history. Clearwater Forest and consumer chips next year on 18A.

Using TSMC was nothing but a stop gap solution. They don't want to rely on TSMC going forward and we don't really need another company to rely on TSMC. TSMC is heavily overbooked as it is.

AI GPUs is like 75% of TSMC output right now and next 24 months are 100% booked. TSMC can't really meet demand. Nvidia has like 12 months wait for enterprise AI orders right now.

 

[msroadkill612]

GF was not always bad. They fell behind TSMC. See the danger in relying on a third party fab?

None of this has anything to do with my point. Good job bud : )

The above is the narrative that suited Intel but both miss the main points.

1/ AMD/GF mainly demonstrated the folly of owning fabs, the second ones chips stopped selling, a fate which now affects intel. They are a costly millstone on its back. Flicking GF was a life saver for AMD. Intel needs lottsa luck extract any value from what was once listed as a major asset. They cant be in the fab biz as they cant be trusted.
2/ Its true that early Zens were ho hum. Even at a node behind tsmc, intel were competitive.
The killer was that Zen got amd's chiplet architectures, foot in the door. For enough buyers, especially enthusiasts, a 6 core for ~the price of a 2 core OR an otherwise unaffordable 8 core, swung the deal, and word got around that they worked just fine.
Chiplets really simplified product revisions vs the huge cost & risks of revising Intel's monolithic offerings. AMD's frequent revisions of zen gradually eroded the inherent downsides of chiplets, & they found themselves with an architecture far better able to cope w/ the demise of Moore's law. Intel was compelled to get out of their depth revising products that their hubris had allowed to stagnate.

I feel a lone voice in attributing most of Zen's success to its chiplet (MultiCoreModule) architecture & its complementary Infinity Fabric bus.

 

[gamerk2]

Yes, they need both angles, but definitely needed more of the engineering side at the head. Again, maybe Pat wasn't the ideal candidate to bring that in.

Ideally, what I've personally seen works is making most of the middle management Engineers, and giving them the budget to do their jobs, but keeping a financial guy on top to make sure spending limits and profit goals are maintained.

The problem with making all of management with financial backgrounds is you end up focusing almost exclusively on the next fiscal quarter, to the detriment of long term sustainability.

 

[guillejp]

Ideally, what I've personally seen works is making most of the middle management Engineers, and giving them the budget to do their jobs, but keeping a financial guy on top to make sure spending limits and profit goals are maintained.

The problem with making all of management with financial backgrounds is you end up focusing almost exclusively on the next fiscal quarter, to the detriment of long term sustainability.

Beautiful.