Intel Thread

[Malcador]

https://newsroom.intel.com/news-releases/intel-changes-technology-organization/#gs.bbpryz

Bit of a reorganization, big change is the Chief Engineering Officer leaving.

[Malcador]

Large data breach at Intel - https://www.tomshardware.com/news/massive-20gb-intel-data-breach-floods-the-internet-mentions-backdoors

 

[Keyrock]

Sometime earlier in Santa Clara, California...

[Zoraptor]

Some more information (though not much) on Intel's consumer GPU offerings via Architecture Day. External fab, hardware RT, launch 2021 expected. A few other interesting bits like them licensing a GDDR controller rather than designing their own.

Otherwise, not much from the event unless you're interested in laptops (better iGPUs/ Tigerlake SoCs) or are invested in making jokes about Intel's node size conventions: Tigerlake now isn't 10nm++ but 10nm SuperFin. Presumably, AlderLake will now be 10nm SuperDuperFin instead of 10nm+++. They've also retconned Cannonlake's 10nm iteration to be 10nm- presumably to stop people calling icelake's process 10nm+.

Of all the things wrong Intel responds to it's memes about 14nm++++++.

[Malcador]

Comments by a supposed Intel employee on what's going on inside - https://mobile.twitter.com/chiakokhua/status/1288402693770231809

 

[Keyrock]

SuperFin, eh? Intel can call it whatever they want, let's just hope they can mass produce chips with it. More interesting than that, to me, anyway, is the news that Alder Lake will apparently utilize a littleBIG approach. ARM chips in mobile devices have been doing this for years and it's a smart approach to power saving. More useful in laptops snd servers than desktops and workstations, methinks.

[Zoraptor]

6 hours ago, Keyrock said:

SuperFin, eh? Intel can call it whatever they want, let's just hope they can mass produce chips with it.

Yeah, they've only ever done low core count chips on 10nm, and the cannonlake SKU didn't even have a functional iGPU.

They still only have 4c/8t offerings with 10nm and all(?) the Tigerlake offerings top out at 4/8 too, but they are meant to launch server chips Soon(TM) and that would imply better and more reliable yields- or that they're doing chiplets. Alderlake desktop's only meant to go up to 8c (8 single threaded big cores, 8 atom small cores) and for desktop the big question is probably what clock speeds 10nm can get to, the problem with the current 5.3Ghz flagship chips is that if their replacements cannot match that a lot of any IPC improvement is canceled out- and a 9900k may still be faster if its HT virtual cores are faster than an atom/ Microsoft doesn't do scheduling properly or whatever.

[Keyrock]

I imagine there will be growing pains with littleBIG at first, but they'll get the scheduling ironed out across Windows, Linux, BSD, etc eventually, though it may take a year or two.

I think it's a great idea, and overdue, quite frankly, anywhere power consumption is a primary focus. On a desktop chip, on the other hand, I'd rather they didn't waste die space on low power cores, at least not on any enthusiast-grade chip.

[ComradeYellow]

https://www.tomshardware.com/news/intel-announces-rocket-lake-q1-2021-release-ahead-of-amd-zen-3-announcement-pcie-4

LOL that was quick.  They literally just released 10th gen.  Oh well, perhaps now I'll just buy 'Rocket Lake' and the Geforce 3080 all in one shabang, provided AMD doesn't release KILLER hardware before then (doubtful, Nvidia is kicking ass with DLSS and RTX and the 3080 is a beast but never say never!)

[Zoraptor]

Pretty sure Alder Lake is slated for next year too so they may have a new gen every six months. Then again Alder Lake is meant to be desktop on 10nm so it will almost inevitably slip as every other 10nm product has.

The most amazing thing about Rocket Lake (even more than it being 16 core/ 24 thread) is that for once it won't need a new motherboard.

[AwesomeOcelot]

1 hour ago, Zoraptor said:

The most amazing thing about Rocket Lake (even more than it being 16 core/ 24 thread) is that for once it won't need a new motherboard.

Wait, what? 16 cores?

[Zoraptor]

Gah, Rocket Lake is only up to 8 cores. The weird 16/24 config is Alder Lake's big/ little one.

[Keyrock]

I am curious how much performance Intel can squeeze out of 14nm+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

In 2030 AMD will be putting out 1 nm chips while Intel will be putting out 14nm+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

I keed, I keed, I hope I keed 

Edited October 8, 2020 by Keyrock

[Malcador]

https://www.tomshardware.com/news/intels-long-awaited-fab-42-is-fully-operational

Star Wars jokes apply here. 

[Keyrock]

What if I make a Star Citizen joke instead? 

 

[Malcador]

1 hour ago, Keyrock said:

What if I make a Star Citizen joke instead? 

 

Try sticking to things that exist

[ComradeYellow]

Yup Intel remains a $240 Billion company so it doesn't look like they're in a grand rush to innovate their architecture, so as a consumer and not a company shill I'm very glad AMD is forcing them out of complacency.  Should hopefully see some new chip designs within a few years.

[Zoraptor]

To be fair to Intel they certainly do have architectural improvements, and they are pretty decent- on laptop. Both Ice Lake and Tiger Lake had decent IPC gains compared to skylake.

Their problems are that firstly the architectural improvements were tied to the failed 10nm node (indeed, upcoming Rocket Lake is pretty much literally Sunny Cove (iirc) backported from 10nm to 14nm) and secondly that unless you can get the new arch also running at 5Ghz+ clocks a lot of that IPC gain will be 'masked' by loss of clock speed.

Rocket Lake should have a decent (20% seems likely) IPC gain, but since it almost certainly won't be be hitting 5Ghz due to heat the practical improvement will be less than that.

[Keyrock]

Alder Lake is what interests me from Intel, not for a desktop, but I may be in the market for a new lappy in a year or two and littleBIG seems extremely well suited for a lappy.

I'm also looking forward to seeing Intel's discrete GPUs.

[ComradeYellow]

5 hours ago, Zoraptor said:

 

Rocket Lake should have a decent (20% seems likely) IPC gain, but since it almost certainly won't be be hitting 5Ghz due to heat the practical improvement will be less than that.

Sounds too much like Zen 3 imo, it simply wouldn't be Intel if they weren't sporting +5Ghz boost clocks.

As I think @Keyrock mentioned, it's the fact that they're still using 'Core' that bothers me.  Time for something completely new, like 'Ocean' or something (I mean come on it's Intel, it's gotta sound watery )

[Zoraptor]

There's a limit to what they can achieve in heat dissipation, even having moved back to solder from thermal toothpaste. Rocket Lake's architecture was initially designed for a far more power efficient node, and a far more dense node, so the chips will be larger and at equivalent clocks hotter than a skylake based equivalent. That's also likely the reason for consumer topping out at 8 cores. The practical improvement should still be better than anything Intel has had since Sandy Bridge.

Core is still competitive as an architecture, and any new naming scheme would be a rebranding. At its heart Core descends from a Pentium Pro via Pentium III, so it wasn't itself a revolutionary design- much as with the Bulldozer -> Zen transition a lot of it appearing revolutionary is because its predecessor (Pentium IV) was rubbish.

[Keyrock]

Some CPU porn for y'all:

[AwesomeOcelot]

Time for Passmark to change their CPU benchmark.

[ComradeYellow]

The only people I feel bad for here is AMD employees, I surely hope they get bonuses (in some form or another) from Lisa Siu's slave driving in her successful quest to be a hardware powerhouse.

Back to the drawing board for Team Blue, watch them blame AMD's success on "Russian Meddling" 

[Azdeus]

1 hour ago, AwesomeOcelot said:

Time for Passmark to change their CPU benchmark.

I'm going to have a blast looking at the userbenchmarks site to see what hoops they are going to jump through to invalidate AMD being single thread king.