Analysis, Please? It seems Qualcomm Paid loads for 3 developers

https://hardware.slashdot.org/story/...ectly-at-apple

Well, its their money, why shouldn't they pay for something they want? There is no telling what exactly they are paying for, not being a member of their board.

Apparently Nuvia are aiming at the top end of the Arm market, and they are headed up by a guy who was a big cheese in Intel CPU development. Qualcomm have hot Arm processors themselves, but off the shelf stuff, and do some serious Arm cores for the mobile market.

It seems Qualcomm is buying Nuvia ( who have never actually made a product to assemble a cpu hardware 'dream team.' "Hardware" is a questionable name for it, as these days the hardware is all imagined in the engineer's head, and described in VHDL or occasionally Verilog. Then it is sumulated in software instead of being built. From that it goes to silicon. So my generation has seen the end of Valve TVs, (They're now reserved for High KV applications, RF Woodglue dryers, plastic welders, Broadcasting, etc), the virtual end of discrete components, and the beginning & ascendancy of ICs.

In business world you never know unless you are an insider, sometimes you want to buy an idea, for this you need to buy the guy who has it, and your end goal may be to bury this idea so it won't interfere with your business. I'm not implying this is what is going on, merely illustrating how deceptive can be appearances.

Had to laugh at this bit:

These bits are telling:

Even more telling...

My reading is that Qualcomm wants to compete with Apple, and what better way to do that, than to buy their opposition...

So why do you think they paid "loads" for them business_kid?

Look the first time you design a particular type of hardware, it's a lot of work. You come away with a working knowledge of the design, and a set of resolutions for next time. The second one is often a great product.

Everybody has multicore ≅2Ghz Arm designs. Apple (who, incidentally don't sell hardware components to anyone IME) have a 4.5Ghz turbo in their M-1, which is enviable. The nearest competitor is Ampere computing with 80×3.0/3.3Ghz A-76 cores which stay in turbo so they're always at 3.3Ghz. Sure, Qualcomm want multicore cpu designers. Qualcomm aren't in the habit of continuing less competitive designs

I think

• Apple have already ditched x86_64. Everything was already on Arm except the Mac laptop, and the M-1 allowed that to happen.
• Qualcomm aren't competing with Apple. But they want to outperform their M-1. They are in the components business, unlike Apple.
• They are interested in maintaining their top dog status in the mobile Arm race. They want monopoly status on high end Arm, like Intel had for a long time on X86.

But I start these threads to get insights from those with a different viewpoint - it's a lazy education for us all.

I agree with this statement from the article:And believe this clearly is important to Qualcomm:Yes there are certainly plenty of details not public. Still, it's not just 3 engineers, it includes the work their startup accomplished, along with their individual pedigrees, the most obvious is their former employment with Apple. I'm not saying it's a corporate coup, but that it helps whatever strategy Qualcomm is envisioning here.

And where I did I say otherwise ? Where did I say it's easy ?

Well, you asked for people's "thoughts" and I gave you mine based on the same link you posted in your OP, and my "thoughts" haven't changed either...

Carry on.

I think my thoughts actually have changed. Many of the comments are very insightful.

Qualcomm are forecasting new hardware in the 2nd half of 2022. Designing a top-of-the-line CPU is actually about a three year project for a design team, even one with precompiled Arm cores. Going from completed design through silicon takes 6-12 months these days, while one of the few 5nM or 7nM foundries schedules you a production slot. IIRC Samsung, TMSC, & Apple are the only foundries. So now I'm thinking

• With their Apple connections, the Nuvia lot may have already secured a production slot at Apple or one of the other foundries.
• A substantial chunk of whatever they were working on may well just be slightly tweaked to fit into Qualcomm's design.

The 'off the shelf parts' that the Qualcomm engineers were using would have been designed with significantly larger fab size in mind. That approach would be like building with Lego - your options are limited. I remember long R&D delays with the Qualcomm Snapdragon 800 - possibly the 'Lego approach' was lacking. A 'pre-fab' design would be like the windows users who Post on LQ and expect to configure everything in linux with point & click. My sense is that to break 3Ghz with Arm cores takes world class engineers designing at a low level. So that's why Qualcomm paid what they did. When you look at Arm designs above 3Ghz, I Can only think of

1. Apple's M-1: base freq 2.5Ghz; Turbo 4.5Ghz. 8 cores +GPU for a laptop
2. Ampere Computing's EMAG: base freq 3.0Ghz; turbo 3.3Ghz.

The Ampere Computing thing is a class apart, as it has a ridiculous 80 cores and a proprietary water cooled heat sink allowing turbo 100% of the time. It is strictly a server unit with larger Arm Neoverse A-76 cores instead of A-72 ones.

Considering the numbers of Arm designs out there, that's a very small number of ≤3Ghz devices.