Read more.Quote:
Insiders say Intel will update numbering convention to "match the industry standard."
Printable View
Read more.Quote:
Insiders say Intel will update numbering convention to "match the industry standard."
Great, so when are they going to change their TDP measurements?
So we can look forward to 70 and 60nm products can we? Hmm I don't think so.
It will mean the Intel process nodes might get moved down a few numbers. For example Intel 14NM is actually closer to 10NM/12NM nodes from other companies in certain aspects,and their 10NM process node is closer to TSMC 7NM in certain aspects.
If they made decent products that didn't single handedly cause global warming, then people wouldn't give two poops about their transistor density.
Some aspects yes, but the "good yields" vs "barely usable" is probably the key metric here. That and Intel's 10nm sucking for desktop performance (to be fair, TSMC 10nm was mobile only as well, you can't win them all).Quote:
Originally Posted by CAT-THE-FIFTH
But hey, the computer industry has always had a fair dose of carefully chosen benchmarks :D
This genuinely made me laugh!Quote:
Originally Posted by ultrasbm
I'm pretty sure AMD tried this back in the day, where their products weren't as fast as the Intel ones, and Intel ripped into them for it, naming things based on their perceived performance rather than what they actually were....
Ho hum...
Hmm, this gives me an idea for my response to this press release:Quote:
Originally Posted by bae85This genuinely made me laugh!Quote:
Originally Posted by ultrasbm
You have stolen my dreams and my childhood with your empty words.... how DARE you.
Timing is everything
The problem here is Intel is having to make relatively large chips on 10NM - if they had a "glued" design like AMD has,it would most likely be useable. I am kind of surprised with Intel having access to EMIB,etc they haven't moved to chiplets yet.Quote:
Originally Posted by DanceswithUnix
Fishlets maybe , everything Intel does of late is fishy.
Haven't they announced they're planning to do that, is it called foverous or something, not sure if that relates to desktop, mobile, or server though. I sort of lost interest in what Intel was doing a decade or so ago when they just kept churning out the same basic 'Core' design from 2006.Quote:
Originally Posted by CAT-THE-FIFTH
Why would they do that when they can keep milking their fanboys?Quote:
Originally Posted by Tabbykatze
Foveros tile is a 3d packaging form. Problem is with them still being kinda stuck on 14nm it doesn't help that much...Quote:
Originally Posted by Corky34
Throwing transistors at an issue is a common way of doing it and Intel has continued that for at least a decade. But when you throw transistors at it on a poor node then packaging really doesn't mitigate the power draw and heat that ensues
Wasn't the original core architecture a spin off from the pentium M line, or have I garbled that Baldric-fashion?Quote:
Originally Posted by Corky34
Only on certain aspects, not on others. It's just to try and make it look better to ignorant investors who know the difference between 10,12 and 14 on a page but don't understand what that means beyond it. So by being able to print 10nm when AMD is on 7nm looks better than 14nm when AMD is on 7nm.Quote:
Originally Posted by CAT-THE-FIFTH
Hexus actually did a good article on this based on Der8auer's videos etc:
Quote:
Originally Posted by Hexus
Quote:
Originally Posted by Hexus
So to compare apples with apples, what kind of transistor, what doping materials, what are the operating voltages and TDP, leakage currents, overall density, net per-transistor spacing, and overall transistor density per unit area, and ditto but normalised for power input/heat output and efficiency etc.Quote:
Originally Posted by hexus
I have a cunning plan....Quote:
Originally Posted by ik9000
Back in the Athlon XP days, AMD had an IPC advantage, but didn't clock as highly... so my old Athlon XP 2200+ clocked at 1.8GHz, actually competed with a 2.2GHz Pentium 4, hence the naming convention to make them sound the "same speed". It was a smart move when Joey Average typically used to just look at clock speed as the one defining number for CPU choice.Quote:
Originally Posted by [GSV]Trig
Su
Quote:
Originally Posted by ik9000
Quiet now Darling...Quote:
Originally Posted by 3dcandy
I'm not going mad: https://en.wikipedia.org/wiki/Pentium_M
Quote:
Originally Posted by wikipedia
Wouldn't stop the problem that Intel's 10nm doesn't clock high. Good for laptops where that isn't expected, but not for desktops.Quote:
Originally Posted by CAT-THE-FIFTH
That and because the yield has sucked so badly at one point Intel not only halted their conversion of 14nm fabs to 10nm but supposedly converted one back to 14. The idea was to go straight to 7nm, so there never will be any volume in 10nm. But they promised the stock holders 10nm, and something was delivered.
It was. The Pentium M was also in turn a re-work of the Pentium 3. It was a Trigger's broom job where all the bits were re-worked to be better for mobile, but the inspiration was all P3. OFC the P3 was derivative of the P2 which was a cost reduction of the Pentium Pro. This stuff has lineage, and damn that Pentium Pro design was a good one.Quote:
Originally Posted by ik9000
I still find it amazing that Intel's bold roadmap of the time was to push the P4 on to great heights that it was clearly never capable of, and when it all fell apart they got saved by a small team in Israel that had been tasked with knocking out a laptop chip.
Pure marketing. Nothing more.
"If you can't beat them, join them."
Intel 14nm -> 11nm
Intel 10nm -> 8nm
Intel 7nm -> 5 or 6nm
They're all made up numbers for several decades now compared to actual transistor dimensions & line sizes.
No doubt all intel benchmark readings should also be doubled at the same time... :rolleyes:Quote:
Originally Posted by tygrus
Not just intel, amd is just as bad.
The problem is that benchmarks give you values with defined definitions. both TDP and NM node values have ether flexible or divergent definitions.
NM Node size really is a meaningless number, a smaller node doesn't automatically make a chip good or even better, it makes it potentially better that's all
In many ways intel was pushing "die shrink is an important feature that makes our cpu's better" for so many years esp during the core2 and early core i years that now they've hit a bottle neck where each die shrink is getting harder and more complex they've been stuck on their 14nm process and are trying to wring it for all they can their narrative of "why are the opposition still using that smelly big nm process not like our shiny new smaller one" is coming back to bite them in the bum, because we all took the message on board that cpu's have to have new smaller nm processes to be good.
And please don't get me started on the utter steaming bullrubbishrubbishrubbishrubbish that is TDP, at this time TDP from both intel and AMD is a 100% made up number for the purposes of marketing and nothing else.
If you want to know the actual thermal output of a cpu look at the wattage draw, that's it, more electron movement = more heat, at a near 1:1 ratio.
But the problem is Intel and AMD both use TDP to differentiate the product stack. I'd be closer to AMD's call right now but yup both aren't great.
NM again - if they hadn't promised us 10nm for years probably wouldn't be such an issue
AMDs TDP is pretty close to the mark and honours it reasonably well when being measured.Quote:
Originally Posted by Pob255
This feels like the start of a massive campaign of bullrubbishrubbishrubbishrubbish by Intel. "Our 10 has fancy buttons so its really like a 7", "Our 8++++ should be compared to the 7 that TSMC made 3 years ago not the 5 that they're making now", "Yes AMD's processor is faster in every other way but if you stand on one leg and squint you'll see our CPU is almost 3% faster.. win for Intel.
fixed that for you.Quote:
Originally Posted by maxopus
"14nm? We've on 10nm! We've always been on 10nm! Never heard of this 14nm business, you must be mistaken."
Intel has shipped over 30 million Tiger Lake 10nm chips, according to their CEO, and recorded over $20B in free cash flow last year. They don't announce their yields, but something must have improved. AMD's FCF was less than $1B for the year.Quote:
Originally Posted by DanceswithUnix
Intel's first 10nm desktop chips aren't coming until Alder Lake, in second half 2021, on the 10ESF process, so on what are you judging Intel's 10nm desktop performance?
Lakefield is their first Foveros chip in production. It is 10nm ... shipped in q1 2020.Quote:
Originally Posted by 3dcandy
The Xe-HPC/Ponte Vecchio chip is the coming big jump in that technology ... a 47 tile GPU. No 14nm on it either.
If chiplets are superior, then why is AMD making monolithic laptop and game console chips?Quote:
Originally Posted by CAT-THE-FIFTH
Intel is moving to 3D chiplets. AMD will too according to Norrod's 2019 statements in the tomshardware article "Ryzen Up: AMD to 3D Stack DRAM and SRAM on Processors".
3 years late? cough coughQuote:
Originally Posted by JayN
JayN you still haven't answered that question about whether you work for, are paid to promote, have shares in, or otherwise receive monetary or similar benefit from Intel. It's really simple to answer. Until then we'll just give you the label of Intel fanboy and treat everything you say with a healthy dose of salt due to its seemingly blinkered bias.
If Intel could make 10nm desktop chips then they would. Like they planned to originally. Yet years on, they still don't.Quote:
Originally Posted by JayN
If they could make just a handful of chips at 10nm to paper launch to the press even though the public had no chance of actually buying them, then they would and we would have impressive 10nm benchmarks.
Or am I supposed to believe that Intel have the ability, but are giving AMD a head start out of the kindness of their hearts? For a convicted monopolist that would be an interesting turnaround to say the least :D
Oh, and Tiger Lake are laptop chips that top out at about 30W. Now, Intel are finally launching their 10nm Ice Lake Xeon server chips, that's taken way longer than it should have done with low power ARM chips breathing down Intel's neck. So there is clearly progress, but by now there should be zero new releases on 14nm.
Why not? The rest of the industry certainly has no problem improving a process somewhat and renaming it (10nm to 8nm, etc.). Intel had 10nm and now 10nm SuperFin. It could have been called 8nm or 7nm, like other fabs do it. Perhaps Intel won't go for a rename with this specific process, but it could improve 10nm a bit more and then rename it.Quote:
Intel clearly isn't going to be renaming its 10nm SuperFin process
Intel has said before that 7nm is on track. If it can get a pseudo-7nm process based on 10nm, would anyone call it a liar to its face? :)
Anyway, at least Intel seems to finally have 10nm ready for desktop, which is nice.
It does? Where? Last I heard they were back-porting it to 14nmQuote:
Originally Posted by ET3D
Because they are superior in some ways (scalability, mixing process nodes), but not in others (they're more complex and add overhead), and because AMD doesn't yet have chiplets working on the GPU side, and both consoles and mobile have integrated GPUs.Quote:
Originally Posted by JayN