• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Intel kills 10nm ?? oO

How would the chiplets combine to make 1x16 PCI-E link anyway? Someone did not engage brain before deciding the chiplets had PCI-E lanes off package.
 
4c/8t low power CPUs only to start with but finally starting to see IPC improvements from Intel again, which is great news. How relevant it'll be to desktops depends on how quickly they can mature the 10nm process I suppose.
 
Sargatanas2511 said:
Wait until independent reviews are out for Zen 2 and Ice Lake, I have a feeling Intel will struggle to demonstrate 18% average when it actually arrives.
Click to expand...

I’m sure Intel will have it’s 18%

The problem Intel will have is trying to get sunnycove to 5ghz which is unlikely

They’re say clock for clock it’s 18% but when the 9900k is 5ghz and sunny cove is I dunno 4.4ghz for example then it’s less than 18%
 
Grim5 said:
I’m sure Intel will have it’s 18%

The problem Intel will have is trying to get sunnycove to 5ghz which is unlikely

They’re say clock for clock it’s 18% but when the 9900k is 5ghz and sunny cove is I dunno 4.4ghz for example then it’s less than 18%
Click to expand...

Used to be the case for Broadwell and Skylake as well; when they struggled to get good high clock when they first moved to the 14nm node.
 
Grim5 said:
I’m sure Intel will have it’s 18%

The problem Intel will have is trying to get sunnycove to 5ghz which is unlikely

They’re say clock for clock it’s 18% but when the 9900k is 5ghz and sunny cove is I dunno 4.4ghz for example then it’s less than 18%
Click to expand...

There are potential architectural mitigations to add onto it too, such as Zombieload and any other exploits that may appear. Plus Intel will be using absolutely bestcase examples even now let alone in a year or so when it appears.

Also it was Skylake from 2015 they mentioned as the 18% improvement over so is there any IPC improvement from Skylake to Coffee Lake already?
 
They added AVX512 to it, almost no use in most workloads, but in a few benchmarks, great, benchmarks they have a huge influence in. AVX512 could well mean that it's 50-150% faster in some AVX examples and 1-2% faster in everything else, or even slower with hardware mitigations built in.

I've also mentioned the possibility that Intel won't have 10nm available anywhere near as soon as they say. They were lying about the launch of 10nm last time, even right before it, even launching a few weeks of risk production with dire yields but the products were only available like 4 months later, and by products I mean, literally two very low volume products.

If Intel wants to spoil AMDs 7nm announcements, then saying they have 10nm coming soon, claiming it's got a huge IPC gain and claiming lots of other things is pretty much exactly what I'd expect them to do even if 10nm is still dire and with no chance of real volume and real products any time soon.
 
Last edited:
N19h7m4r3 said:
Interesting fine print on their performance slides for that. Is Ice lake still affected by the security issues, and patches?

Tuv49WP.png

b9pzc3N.jpg This one is even better!
Click to expand...

Nice find :) Maybe Intel knows about new security flaws reported to them and not yet published as of today lol

But regarding optimization, it's true that there are still programs heavily optimized for Intel. Not everyone buys a CPU only to run Cinebench. For example, Linpack is widely used in scientific computing.
 
N19h7m4r3 said:
Interesting fine print on their performance slides for that. Is Ice lake still affected by the security issues, and patches?

Tuv49WP.png

b9pzc3N.jpg This one is even better!
Click to expand...

So they 'possibly 'install not publicaly available updates that hobble performance, then test with software that has only been optimised for Intel processors - magic 18% IPC difference, Profit! This all could be the new instructions of the Ice Lake architecture and a recompile to take full advantage of it, why can't Intel just be completely open instead of trying to make headlines..

Looks like they are trying to spin their way out of a hole.
 
DragonQ said:
4c/8t low power CPUs only to start with but finally starting to see IPC improvements from Intel again, which is great news. How relevant it'll be to desktops depends on how quickly they can mature the 10nm process I suppose.
Click to expand...
It may never come to desktops, it's taken them a year just to double the core count (from the i3 8121U).
Gabbatek said:
10nm has had repeated problems for intel, but amd can do it? I wonder what the issue for intel is. Maybe the performance isn't worth the cost.
Click to expand...
From what i gather it's mainly down to replacing copper with cobalt for the interconnects in the last few layers, if i remember correctly a paper was publish years ago looking at how cobalt was better than copper when wire traces were reduced beyond a certain size, with copper you end up with a larger diffusion barrier than the trace itself whereas cobalt doesn't diffuse into the surrounding silicon, one of the problems Intel's been having is controlling the thermal differentials between cobalt and the surrounding silicon, because they expand and contract at such different rates it causes the cobalt interconnects to crack.
 
It's been said there are multiple fairly major issues with 10nm, struggled with quad patterning, struggled with cobalt and the gate design. I believe someone has implied that the gate design may have been dumped to help yields.

In general I'm just not going to believe Intel has fixed 10nm nor that yields are great till we see evidence. When they are launching a new node but also launching 10 core (presumably no igp, and still only what 200mm^2 or so) chips on 14nm you suspect that yields are still poor. If yields were normal for quad quad apus and were launching in volume you'd think they'd be 3-6 months max from 8-12 core desktop chips.

Intel have become so untrustworthy with their information and blatant lies on 10 and 14n for years now that when we see a midsized 10nm shipping in volume is the day I'll believe they can do that, when they start doing larger server parts on 10nm again, that's when I'll believe it.
 
Yea and apologies if what i said made it seem like the cobalt thing was the only issue, it's not, however personally i think it was the straw that broke the camels back, SAQP was also pretty dumb IMO, from an alignment and cost perspective, and then iirc that forced them to change the gate design because they couldn't always get things to lineup how they wanted to.

IMO they should've just gone full on EUV but even that would've had it's own problems as that was far from perfected when they first started on 10nm.
 
Gabbatek said:
10nm has had repeated problems for intel, but amd can do it? I wonder what the issue for intel is. Maybe the performance isn't worth the cost.
Click to expand...

Intel is trying to implement 10nm at a foundry level and many aspects of their "10nm" where important to CPU design are far more advanced, more like 7nm, than 10nm at other semi-conductor foundries ostensibly with the intention of getting a 3 year lead on the competition but seems to have ended up more like 3 years behind - AMD has used a general purpose 7nm node made by someone else.

drunkenmaster said:
It's been said there are multiple fairly major issues with 10nm, struggled with quad patterning, struggled with cobalt and the gate design. I believe someone has implied that the gate design may have been dumped to help yields.

In general I'm just not going to believe Intel has fixed 10nm nor that yields are great till we see evidence. When they are launching a new node but also launching 10 core (presumably no igp, and still only what 200mm^2 or so) chips on 14nm you suspect that yields are still poor. If yields were normal for quad quad apus and were launching in volume you'd think they'd be 3-6 months max from 8-12 core desktop chips.

Intel have become so untrustworthy with their information and blatant lies on 10 and 14n for years now that when we see a midsized 10nm shipping in volume is the day I'll believe they can do that, when they start doing larger server parts on 10nm again, that's when I'll believe it.
Click to expand...

I've heard and not just from random people that Intel has supposedly been shipping Nikon EUV equipment into 10nm fabs which doesn't make much sense at all, Nikon are supposedly defunct in that area. And behind some of the rumours of them converting 10nm fabs to 7nm instead. So I have no idea what is going on.
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom