Jow Forums's thoughts on 5nm

TSMC 7nm is 10% smaller than Intel 10nm, TSMC is ahead

But 10nm is not equal to 7nm. Don't you think that tsmc labeling their chips as 7nm while intel, with a very similar size, labels it as 10nm is a problem? If you look at each company there is a significant difference between two generations of chips, while the difference between intel's 10nm and tsmc's 7nm is rather small.

>What does "feature size" actually mean?
depending on the chip manufacturer, it's a marketing term nowadays, see glofo.

Attached: sram-density-tsmc-5-est.png (1000x759, 190K)

It actually has superior transistor density

Attached: Slide4.jpg (1256x707, 88K)

Attached: DgbbPmq.png (3840x2160, 317K)

which, by the way, is one of the reasons it doesn't work
>year of production 2017/2018
oh the optimism

It works, just with semi broken cpus that can only clock really low

>10% smaller
54 x 44 = 2376
54 x 40 = 2160
2160 (54 x 40, TSMC 7nm) is 11.36% smaller than 2376 (54 X 44, Intel 10nm).

It's also an abortion.

wow, a truly significant difference. Tsmc's 7nm is 25% smaller than it's 10nm. Intel's 10nm is 35% smaller than its 14nm.
It'd be ok if the 11% were only manufacturer-to-manufacturer variations, but they are labeled as completely different generations and it misleads people.

10% can be significant. It's the difference between 4.5GHz and 5GHz

They still cannot even make 10nm desktop CPUS.

neither can samsung or glofo

Wait until 2022 or something.

By "a truly significant difference" I was poking fun at the nerd correcting 10% to 11.36%.

Back to the point, tsmc's 10nm is 22% bigger than intel's 10nm. I believe we should stop using feature size as the measure of fabrication capabilities because it's clearly subjective.

Wake me when those 1nm transistors can be put on a working chip and is being sold in gpus and cpus.

TSMC 1nm = 2025
Intel 10nm = 2125

Feature size means just that. The size of a feature. It isn't a single thing, SRAM cell size, xtor size, track height, etc. Those are all various features which are commonly compared.
intel's 10nm process was said to achieve a transistor density up to 100 million xtors per mm2, though its incredibly unlikely that they'll utilize this density in anything coming to market in the next couple years.
TSMC's 7nm DUV process is said to achieve transistor density up to 89 or 90million per mm2. Though if you look at specific features on the FEOL the TSMC process has smaller SRAM cells and etc.

If intel spent like a decade to get 10nm working sort of you'd think 5nm starts to being close to the limit.

Intel came out with a 10nm CPU (Core i3-8121U) last may.
Since then, Intel had yet to make another 10nm CPU.

They royally fucked up the design and were (still are) under too much pressure to just shitcan it and start over.

Must explain what happened to Cannon Lake.

Then they're fucked indefinitely

While Intel's procrastinates on "10nm", what about AMD & 5nm?

AMD gets consistent improvements in perf/watt and likely raises base clocks across their whole product line as costs actually slightly decrease.

Attached: 1541702343913.jpg (679x263, 88K)

>Zen 4
>5nm
>2021

Attached: amd_roadmap_cpu_q2_2018.png (1600x900, 274K)

Need to note that tsmc's N7 high performance is also not at the 90million mark either, that's for phones and other low power stuff. I think the high power stuff is reported at around 74million/mm^2

They might as well say that desktop 10nm is dead and they're just waiting for their own 7nm.

>they're just waiting for their own 7nm.
And by time Intel comes out with their 7nm CPUs, AMD is well into 5nm, if not 3nm.

Attached: 01ceb804-28be-45ae-8ba5-b204df61c1fa.jpg (690x388, 45K)

>log scale comparison for actual feature sizes
That's disappointing. Turn it into a normalized per-core performance chart by using an inverse function and you'll have a better show of who is good at what size.
The irony of Intel's 10nm being failed is not lost with that example.

It would have been such a beautiful slap in the face to the whole industry if they succeeded though. Think of the keynote speech and investor relations slides.
>We at Intel are proud to unveil:
>10 times more throughput
>10 times less power per instruction,
>10 times more transistors per dollar
>Ladies and gentlemen, our 10nm lithography and our all-new family of 1000 series consumer desktop products.
>Intel. Enabling your future today.
I bet 80% of the high-ups have their head in the guillotine and alcoholic depression is rampant.

PUT ME IN THE SCREENSHOT

Attached: 1555800838525.webm (750x692, 2.55M)

Attached: file.png (1349x1621, 362K)

costs aren't actually decreasing, since wafers are climbing steadily and new tooling for this kind of thing never was cheap and ASML pretty much has a monopoly right now

>Ladies and gentlemen, our 10nm lithography and our all-new family of 1000 series consumer desktop products

actually the fun thing to think about is they planned to have 10nm no later than 2018. Imagine the 8086k running at 5ghz at the same power as a 2600x. Or even intel beating AMD to the moar cores meme and the 8086k could have been a 12 core chip at 5ghz for the same power as the real 8086. ryzen would have been BTFO