Intel - Going Nowhere, Fast.

Burr · 29 May 2018 at 21:07

I think we can safely say the next couple of years are going to be very interesting in the CPU space.
https://semiaccurate.com/2018/05/29/is-intels-upcoming-10nm-launch-real-or-a-pr-stunt/

eddiew · 29 May 2018 at 21:38

Burr said:
I think we can safely say the next couple of years are going to be very interesting in the CPU space.
https://semiaccurate.com/2018/05/29/is-intels-upcoming-10nm-launch-real-or-a-pr-stunt/

If by interesting you mean Intel will stand still until 2020, producing 10nm parts that are no better (or worse) than their current 14nm line, meanwhile AMD sneak up from the back and claim peak performance in mainstream territory... yes? But my hunch is that AMD aren't going to break 5.0 on all cores even with the next gen, so it's as much up to the games and application developers to properly properly use all available threads in order to make Ryzen shine like it should. If you've got one game that's limited by 1-2 heavy threads, my guess is you'll never see a better answer than an 8700k

That aside, I'm increasingly convinced that the right thing to do about my crappy motherboard is replace it - with the best X470 I can lay my hands on. Then keep my 1600X, and switch it for a 3000 series next year. Bit disappointed, had really hoped for an i7 refresh with better clocks and soldered IHS but... not happening is it?

4K8KW10 · 29 May 2018 at 21:45

eddiew said:
If by interesting you mean Intel will stand still until 2020, producing 10nm parts that are no better (or worse) than their current 14nm line, meanwhile AMD sneak up from the back and claim peak performance in mainstream territory... yes? But my hunch is that AMD aren't going to break 5.0 on all cores even with the next gen, so it's as much up to the games and application developers to properly properly use all available threads in order to make Ryzen shine like it should. If you've got one game that's limited by 1-2 heavy threads, my guess is you'll never see a better answer than an 8700k

That aside, I'm increasingly convinced that the right thing to do about my crappy motherboard is replace it - with the best X470 I can lay my hands on. Then keep my 1600X, and switch it for a 3000 series next year. Bit disappointed, had really hoped for an i7 refresh with better clocks and soldered IHS but... not happening is it?

It is exactly about time for AMD to ramp the production of Ryzens significantly up, and prove that Intel is a very unreliable supplier

What goes around, comes around. After so many years of playing dirty, Intel was finally caught with the karma.

Burr · 29 May 2018 at 21:46

There are many factors - but watching how Intel deals with the 10nm issue going forward is one of them. Zen 2 IPC/clock is another.

eddiew · 29 May 2018 at 21:56

4K8KW10 said:
What goes around, comes around. After so many years of playing dirty, Intel was finally caught with the karma.

While I do agree there's a certain satisfaction involved, I'm not sure the current situation benefits the consumer. Would we not have been better off if Intel had both had a fire lit under them AND proven capable of actually making a better product than Covfefe?

Although I guess if what you want is at most 6 very fast cores, you can safely pick up an 8700K and know it's still going to be current in 2 years

Burr · 29 May 2018 at 22:02

eddiew said:
While I do agree there's a certain satisfaction involved, I'm not sure the current situation benefits the consumer. Would we not have been better off if Intel had both had a fire lit under them AND proven capable of actually making a better product than Covfefe?

That product will come (i'd bet) but it'll take 2 years+ by the looks of things.

eddiew · 29 May 2018 at 22:22

Burr said:
That product will come (i'd bet) but it'll take 2 years+ by the looks of things.

Which is why I'd be happier if Intel's 10nm actually worked now... Whether this is good long-term is yet to be seen. Maybe we'll see proper engagement and development on both sides and they'll pass the crown every couple of years

Or maybe AMD will run away with it for 5 years, get lazy, and we'll all be wishing Intel would just produce something competitive to bring the improvements back.

Burr · 29 May 2018 at 22:29

eddiew said:
Which is why I'd be happier if Intel's 10nm actually worked now... Whether this is good long-term is yet to be seen. Maybe we'll see proper engagement and development on both sides and they'll pass the crown every couple of years Or maybe AMD will run away with it for 5 years, get lazy, and we'll all be wishing Intel would just produce something competitive to bring the improvements back.

Yeah the situation sucks. But the CPU scene has been sucking for a long time (pre-Ryzen), at least things are fired up now.

4K8KW10 · 29 May 2018 at 22:46

eddiew said:
Which is why I'd be happier if Intel's 10nm actually worked now... Whether this is good long-term is yet to be seen. Maybe we'll see proper engagement and development on both sides and they'll pass the crown every couple of years Or maybe AMD will run away with it for 5 years, get lazy, and we'll all be wishing Intel would just produce something competitive to bring the improvements back.

It depends on what will happen with the transition to new technologies for manufacturing of processors. It may be possible that other companies join in the race.
It may also be possible that AMD and Intel abandon their core business.

See the other thread for 7nm process for reference and the links there.

Silent_Scone · 30 May 2018 at 07:23

Just as well they’ve got nothing to worry about, then.

jigger · 30 May 2018 at 15:27

Silent_Scone said:
Just as well they’ve got nothing to worry about, then.

Flawed chip design, fragmented platforms, broken manufacturing and the competition running rings round them with no possible fix until 2021. Seems Intel have a lot to worry about.

Apart from the push into PCIE graphics it's all bad news comming out of Intel and worse to come on the horizon.

mmj_uk · 30 May 2018 at 16:46

A few years ago AMD fans were arguing that IPC was completely irrelevant and only core would matter, Intel already have native 18-core chips for sale (albeit expensive) whereas AMD don't even have a native 6+ core chip in the Ryzen/Ryzen2 line. If Intel were go non native and start strapping quads together I'm sure they'd have a much easier time in terms of yields.

jigger · 30 May 2018 at 16:52

mmj_uk said:
A few years ago AMD fans were arguing that IPC was completely irrelevant and only core would matter, Intel already have native 18-core chips for sale (albeit expensive) whereas AMD don't even have a native 6+ core chip in the Ryzen/Ryzen2 line. If Intel were go non native and start strapping quads together I'm sure they'd have a much easier time in terms of yields.

Intel need a secure architecture to work with first. The security flaws are laughable.

humbug · 31 May 2018 at 12:12

mmj_uk said:
A few years ago AMD fans were arguing that IPC was completely irrelevant and only core would matter, Intel already have native 18-core chips for sale (albeit expensive) whereas AMD don't even have a native 6+ core chip in the Ryzen/Ryzen2 line. If Intel were go non native and start strapping quads together I'm sure they'd have a much easier time in terms of yields.

What does that matter? have you not been following events?

humbug said:
Right my bad.

ST @ 4Ghz
Ryzen 1: 164 (100%)
Ryzen 2: 168 (102.5%)
Coffeelake: 174 (106%)

MT @ 4Ghz
1600X: 1326 (100%)
2600X: 1384 (104.5%)
8700K: 1325 (100%)

The IPC on Ryzen is virtually identical to Coffeelake

As for your other point, the fact that Intel are still using monolithic dies for their HEDT chips is not a good thing, its a bad thing, monolithic dies are extremely expensive because of their size and the very low yields you get from wafers.
And even with that Intel still have to compromise with a different architecture to make those large CPU's, the Mesh Interconnect which is less efficient than the Ring Bus you find on Coffeelake, and because its less efficient than coffeelake its also less efficient that Ryzen.

https://forums.overclockers.co.uk/threads/ryzen-2600-vs-core-i9-7800x-in-37-game-benchmark.18820868/

Next year AMD are going 24 core HEDT, 12 core Mainstream and 48 or 64 core server. watch Intel squirm when that happens.

eddiew · 31 May 2018 at 12:26

mmj_uk said:
A few years ago AMD fans were arguing that IPC was completely irrelevant and only core would matter, Intel already have native 18-core chips for sale (albeit expensive) whereas AMD don't even have a native 6+ core chip in the Ryzen/Ryzen2 line. If Intel were go non native and start strapping quads together I'm sure they'd have a much easier time in terms of yields.

Correct comparison is with Threadripper, not Ryzen, since Intel's 18-core is an HEDT part. Yes, TR's 16 cores are made by sticking two 8-core dies together, but AMD have pulled some magic out of the hat and the infinity fabric scales very well indeed.

Rroff · 3 Jun 2018 at 05:43

eddiew said:
Correct comparison is with Threadripper, not Ryzen, since Intel's 18-core is an HEDT part. Yes, TR's 16 cores are made by sticking two 8-core dies together, but AMD have pulled some magic out of the hat and the infinity fabric scales very well indeed.

What makes IF something of note is that it can be used as a general purpose interconnect for different types of processing units, even tie in very disparate parts, with minimal performance/feature impact due to its general purpose capability over specialised application specific links.

Contrary to what some claim here Intel has quite a bit of experience in slapping multiple dies together on the same substrate and plenty of R&D on it goes back years not to mention the multi CPU boards - getting CPU scaling working well like that isn't particularly magic - the Q6600 did it fine.

humbug · 3 Jun 2018 at 13:12

The Q6600 was a dual CPU, literally two CPU's not dissimilar to how dual socket boards are today and have been for many years, only on one socket.

I don't think that's the same thing or Intel would have been doing what AMD are now for years.

For example with Infinity Fabric all the cache pool's of the individual dies become one, all the memory channels become a singular bandwidth, all the individual PCIe lanes usable....
For all intents and purposes the separate dies act and function as one.
Its not 2 or 4 CPUs. its just 1 CPU.

The Q6600 was Intel's answer to AMD's Athlon X2, Intel's first dual core CPU and not a very good one, its two CPU's rather than one CPU with two Integer Units with a singular controller, you know, like multi-core CPU's today.

Hotwired · 3 Jun 2018 at 13:21

mmj_uk said:
A few years ago AMD fans were arguing that IPC was completely irrelevant and only core would matter, Intel already have native 18-core chips for sale (albeit expensive) whereas AMD don't even have a native 6+ core chip in the Ryzen/Ryzen2 line. If Intel were go non native and start strapping quads together I'm sure they'd have a much easier time in terms of yields.

Is this a rehash of Intels sour grapes line that AMD are just gluing chips together?

If the process was a failure there would be some weight in the criticism but since it works and results in cheaper chips it is in fact a massive success and may well get copied by Intel as soon as they work out how to spin the PR of saying they're also going to glue chips together (fancy name, micro details of how their glue is different and revolutionary not an imitation).

humbug · 3 Jun 2018 at 13:33

Hotwired said:
Is this a rehash of Intels sour grapes line that AMD are just gluing chips together?

If the process was a failure there would be some weight in the criticism but since it works and results in cheaper chips it is in fact a massive success and may well get copied by Intel as soon as they work out how to spin the PR of saying they're also going to glue chips together (fancy name, micro details of how their glue is different and revolutionary not an imitation).

More than that, he assumes Infinity Fabric performs worse than Intel's currently lesser equivalent, i say lesser because unlike Infinity Fabric Intel's Interconnect Mesh cannot separate dies, so they still have to make massive monoliths.

Despite all of that Here is the performance difference between Intel Mesh and AMD's Infinity Fabric, the IPC mmj refers to, both 12 thread CPU's, notice the 7800-X is clocked at 4.6Ghz vs the 2600 4.2Ghz? that concludes the Ryzen 2600 has higher IPC than Skylake-X.

mmj is attempting to turn something of a weakness in Intel into a failure of AMD by saying AMD cannot make 6 core singular CPU's, when in fact its a failure on Intel part that they still have to make 18 core CPU's on a singular die.

AMD's "lots of small dies glue together" is very intentional, with it they can make higher core count CPUs, currently 32 going to 48 and rumoured 64 next year, and its far more efficient in manufacturing terms.
Its not that they cannot, its that their way is better.

bru · 3 Jun 2018 at 13:54

Infinity fabric isn't without it's faults but it works well enough, well too work. It has allowed AMD to bring us a whole range of CPU's from quad core to 32 core EPYC.
The second generation has improved certain areas and push clocks up quite a bit.

Intel on the other hand are still trying to get it right, I don't think the mesh system works as well, but of course it would be silly of any of us to count Intel out, they have the money and resources to get their act together, eventually and just like with their 10nm node they are having trouble to start out.
They will get it right and when they do I'm sure they will be back with a vengeance.

On a side note @humbug the Q6600 was a two dual core, quad core glue together not a dual.