Intel shows off its chip packaging powers

It means less resistance (for the electricity), which means less heat is generated. It also means less current is needed, lowering power consumption.
 
So they are going with how AMD has their chips set up with chiplets.
 
Whole market has been going this way; AMD just had to do it sooner to make their architecture viable to produce. It also incurs penalties that they have to design around...
So now Intel will be a few steps behind in the process. Hopefully they fix all the problems. Just fixing those would be like have a few generations of IPC. haha
 
So now Intel will be a few steps behind in the process. Hopefully they fix all the problems. Just fixing those would be like have a few generations of IPC. haha

Well, Anandtech has tested Ice Lake to fix quite a bit. And Intel could easily stuff more than eight cores in a socket if they so desire- they already plan to at 14nm and will be able to stuff even more into the socket with 10nm / 7nm, whichever they make desktop SKUs on.

Quite likely, an Ice Lake 12-core CPU would outpace every Ryzen available across the board except for the Threadripper / Epyc CPUs with 24+ cores. Even running at the same clockspeeds, Intel would edge ahead of the 3950X- they have that much of an IPC increase coming down the pipe, security fixes included.
 
I think Intel did it (multiple silicon in a single processor) first (among desktop CPU manufacturers) with the first quad-core Pentium, which was 2 dual core cpu's in the same package. Obviously the new stuff by AMD and Intel is more complex.

The chiplet concept isn’t new. The technology can be traced to the 1980s, when the industry developed multi-chip modules (MCMs). Saying that one of the cpu vendors copied the other isn't really accurate.

The first modern chiplet was made by Marvell, so you could say they both copied that one.
 
I think Intel did it (multiple silicon in a single processor) first (among desktop CPU manufacturers) with the first quad-core Pentium, which was 2 dual core cpu's in the same package. Obviously the new stuff by AMD and Intel is more complex.

The chiplet concept isn’t new. The technology can be traced to the 1980s, when the industry developed multi-chip modules (MCMs). Saying that one of the cpu vendors copied the other isn't really accurate.
Copied as in following them, not making the technology. They both just need to keep up the good work!
 
Copied as in following them

That's the thing- they're really not. Both are using long-established engineering approaches. In AMD's case, chiplets are necessary for the direction they're going in and for how (and by who) they're producing their CPUs. When you look at the memory latency involved, it's not all roses, and as AMD works to improve performance of Zen in the face of Intel's much higher IPC parts in the pipe, AMD is going to have to get even more creative to keep up, with chiplets working against them.
 
That's the thing- they're really not. Both are using long-established engineering approaches. In AMD's case, chiplets are necessary for the direction they're going in and for how (and by who) they're producing their CPUs. When you look at the memory latency involved, it's not all roses, and as AMD works to improve performance of Zen in the face of Intel's much higher IPC parts in the pipe, AMD is going to have to get even more creative to keep up, with chiplets working against them.
Seems to be doing just fine. ;)
 
Well, Anandtech has tested Ice Lake to fix quite a bit. And Intel could easily stuff more than eight cores in a socket if they so desire- they already plan to at 14nm and will be able to stuff even more into the socket with 10nm / 7nm, whichever they make desktop SKUs on.

Quite likely, an Ice Lake 12-core CPU would outpace every Ryzen available across the board except for the Threadripper / Epyc CPUs with 24+ cores. Even running at the same clockspeeds, Intel would edge ahead of the 3950X- they have that much of an IPC increase coming down the pipe, security fixes included.

Didnt know "That much" was a quantitative form of measurement?
 
Just fine against a five-year-old Intel design (ostensibly) two production nodes larger. "Just fine" has happened because Intel stumbled; had Ice Lake been out two years ago, AMD would still be playing catch up.

But they they didn't get Ice Lake out two years ago, now did they?
Are you seriously suggesting that we should be putting "might have beens" into a comparison?
Fine. IBM *should* have skunked everyone with 5GHz 24 core Power PC's 2 years ago, since RISC is so much more efficient than EITHER CISC process.
Let's go ahead and compare this fictional chip from 2 years ago, shall we?
 
So basically, yes, they are doing good.

For AMD :D

Didnt know "That much" was a quantitative form of measurement?

Well, there's an article being referenced for those that need to catch up.

But they they didn't get Ice Lake out two years ago, now did they?
Are you seriously suggesting that we should be putting "might have beens" into a comparison?

Nope, I'm suggesting that we apply previous performance as a predictor of future performance. Specifically, we can consider both Ryzen and Intel's 10nm stumble to be anomalies. Intel usually (statistically) executes process transitions regularly and without drama, and AMD rarely produces parts that edge ahead. The reasoning for this perspective is that AMD looks like they've wrung as much as they're going to get out of Zen without a significant rework, while Intel has had a new architecture ready for their fabs.

We can certainly predict that Intel will likely get 10nm underway, and if not, 7nm on schedule- that's what they usually do- and we can reasonably question whether AMD will get any more performance (that is, some combination of IPC and clockspeed) out of Zen, because they usually stumble with significant improvements.

Fine. IBM *should* have skunked everyone with 5GHz 24 core Power PC's 2 years ago, since RISC is so much more efficient than EITHER CISC process.

But they're not?

[or, strong, consumer-applicable references please]

Let's go ahead and compare this fictional chip from 2 years ago, shall we?

See above.
 
Intel usually (statistically) executes process transitions regularly and without drama, and AMD rarely produces parts that edge ahead.
That's a strange way of saying one company has always had more R&D and manpower than the other, and that the weaker party had only managed to get ahead when the former had grown complacent with greed. Do you think there is something inherent to Intel's staff that makes their employees more reliable than AMD's? Why don't you tell us all about your personal encounters with them so we can judge for ourselves how that conclusion came about.
 
That's a strange way of saying one company has always had more R&D and manpower than the other, and that the weaker party had only managed to get ahead when the former had grown complacent with greed.

Well, it's not a strange way to say that, because I didn't say that.

Do you think there is something inherent to Intel's staff that makes their employees more reliable than AMD's?

I also didn't say anything about the reliability of Intel or AMD staff.

Why don't you tell us all about your personal encounters with them so we can judge for ourselves how that conclusion came about.

...and this is trolling.
 
For AMD :D



Well, there's an article being referenced for those that need to catch up.



Nope, I'm suggesting that we apply previous performance as a predictor of future performance. Specifically, we can consider both Ryzen and Intel's 10nm stumble to be anomalies. Intel usually (statistically) executes process transitions regularly and without drama, and AMD rarely produces parts that edge ahead. The reasoning for this perspective is that AMD looks like they've wrung as much as they're going to get out of Zen without a significant rework, while Intel has had a new architecture ready for their fabs.

We can certainly predict that Intel will likely get 10nm underway, and if not, 7nm on schedule- that's what they usually do- and we can reasonably question whether AMD will get any more performance (that is, some combination of IPC and clockspeed) out of Zen, because they usually stumble with significant improvements.



But they're not?

[or, strong, consumer-applicable references please]



See above.


So we are comparing a mobile only uarch that is not expected to launch on desktop until the end of next year or early 2021?


AMD already has Zen2+, and a team that is finalizing Zen3 (if they haven't finished) , along with ANOTHER team working on Zen4.


Your jab at trolling is cute, but the humor isn't going to change what he said being true. Intel got to their position (in the last ~20 years) by commiting illegal anti competitive practices.

It's easy to say well AMD didn't do this, but aside damage Hector Ruiez did by selling off their profitable fabs and NAND business, Intel is the reason AMD has stumbled the way they have.


They are a company doing more then 2 others that have 10x+ the R&D budget. By that measurement if Intel or Nvidia were as effective, we would be on next gen interconnects and have working MCM in GPUs without the issues they currently face.

I love all three companies products, but your relentless trolling and then slight backpedaling when called on it is a bit exhaustive.
 
So we are comparing a mobile only uarch that is not expected to launch on desktop until the end of next year or early 2021?

We're comparing what we've got.

AMD already has Zen2+, and a team that is finalizing Zen3 (if they haven't finished) , along with ANOTHER team working on Zen4.

...and Intel doesn't have a succession of designs in the pipe?

[also, references]

They are a company doing more then 2 others

They really aren't.

but aside damage Hector Ruiez did

It's always someone, right?
 
Saw the anand article yesterday, and I'm a huge fan of zen/zen 2 (though my workloads tend to favor that ) and I was impressed with the increase seen in ice lake. Still have questions on when they're gonna roll out in mass, and I'm kinda thinking they're just gonna be in the mobile sector. Be interesting to see a desktop part if it's ever released though.
 
yeah by the time this hits desktops, AMD will likely be very near or actually ready to release Zen 2+

*Maybe even Zen 3, if Intel waits until late 2020 for release
 
Intel is doing just fine, it’s AMD that needs to worry. Desktop sales are shrinking mobile and server sales are where it is at. AMD still does not have a good notebook line delivered in any meaningful quantities and their support for big servers while getting better isn’t at a point where Dell, HP, or Lenovo are ready to give them prime time. The Epyc chips are great but hardware support for them has been rocky.
I am looking forward to my 3900x when it gets here, hopefully by then they have a working Bios.
 
We're comparing what we've got.



...and Intel doesn't have a succession of designs in the pipe?

[also, references]



They really aren't.



It's always someone, right?
1. We don't have IceLake. You saw a tech press test day and that's it. According to Ian @ Anandtech it's barely faster overall then WiskyLake.

You cannot buy an IL SKU period. They will be available soonish according to Intel, but then again 10nm was coming Soon^tm.

I never said that Intel doesn't have any designs in the pipeline. That would be the stupidest thing they could do. Especially after the Billions they wasted on 10nm.

2. As far as citations, Lisu Su has said it in nearly every interview and AMA and Analyst calls. Mark PaperMaster had as well.

Lisu Su implemented leap frogging design teams in both sides of the house. David Wanh has several GPU teams that are leapfrogging like the Zen team is.

Lisa Su has said that Zen3 had working silicon back from the fabs and they are working on tuning and qualification testing in their labs.


As far as it always being someone, yes he was directly responsible for being cash desperate after the illegal crap Intel did by bribing OEMs. He came up with the short sighted plan of selling their very profitable NAND business off (right before SSDs and smart phones exploded) and spinning the fabs off into their own company.

Global Foundries was late getting their 32nm process ready which pushed Bulldozer back a year. This was a result of them leaving AMD and forming an entirely new company.

So while AMD did indeed stumble with BullDozer, had they not been the victim of a short sighted CEO AND been cash starved due to Intel's illegal activities they would have has the reserves to ride BD out and get their next design out.

You cab look at the history of the company and see what happened from the time he took over after Intel was running it's illegal campaign.
 
1. We don't have IceLake.

I didn't say we did. I'm referring to Anandtech's reporting.

2. As far as citations, Lisu Su has said it in nearly every interview and AMA and Analyst calls. Mark PaperMaster had as well.

Then add a link.

As far as it always being someone

Meaning that there are always excuses. AMD is never going to get a 'fair shake', nor is anyone else. Welcome to Business 101.
 
Well, Anandtech has tested Ice Lake to fix quite a bit. And Intel could easily stuff more than eight cores in a socket if they so desire- they already plan to at 14nm and will be able to stuff even more into the socket with 10nm / 7nm, whichever they make desktop SKUs on.

Quite likely, an Ice Lake 12-core CPU would outpace every Ryzen available across the board except for the Threadripper / Epyc CPUs with 24+ cores. Even running at the same clockspeeds, Intel would edge ahead of the 3950X- they have that much of an IPC increase coming down the pipe, security fixes included.
So... You already know the IPC increases for 2020 and beyond? Where's that magic crystal ball hidden. Seriously though, you're trying to compare a future product that doesn't exist with no details against the *current* crop of CPUs. It's like when everyone kept saying the next GPU from AMD would compete against the 1080ti... Cool, but the rtx's are now out. Don't get me wrong, it's fun to speculate, but it's far from a guarantee at this point. Also unknown is how much frequency they will have @ 10 or 7nm. It's possible they will gain some, possible they will loose some a d possible it will stay the same. Point is, nobody knows yet.
The part that you don't even know if the part will be 10 or 7nm is very telling that this is more wishful thinking than anything.
 
Also unknown is how much frequency they will have @ 10 or 7nm. It's possible they will gain some, possible they will loose some a d possible it will stay the same. Point is, nobody knows yet.
The part that you don't even know if the part will be 10 or 7nm is very telling that this is more wishful thinking than anything.

No he isn't, this is a physical part people had limited reviews on. 7nm, yes, we don't know how intel's 7nm may perform just that it's on track. Everyone is so defensive these days, yeesh. I also think real reviews of IL will be slower than shown, and some faster, imagine that.
 
He isn't what? He's basing future designs, frequencies and performance on desktop parts that haven't been released not knowing if it's even going to be on 10 or 7nm. Sure, you can guess that they have an IPC increase based on the limited hands on, which there appears to be. You can assume this will scale perfectly and you can assume they can fill it full of cores without compromise. Still extrapolating something different to a future unknown and saying it will be better than what we have today. Congrats, AMDs next product will likely be better than their current products, w00t.
All im trying to say is we can be pretty sure they will have an IPC increase, but they are going to chiplets like AMD, so maybe some loses like amd has with going between infinity fabric. How much increase, nobody knows yet. We also don't know when they will become available or what process they will be on. I'll wait for reviews to hit, thanks.
 
That's the thing- they're really not. Both are using long-established engineering approaches. In AMD's case, chiplets are necessary for the direction they're going in and for how (and by who) they're producing their CPUs. When you look at the memory latency involved, it's not all roses, and as AMD works to improve performance of Zen in the face of Intel's much higher IPC parts in the pipe, AMD is going to have to get even more creative to keep up, with chiplets working against them.
Umm Intel's future designs include chiplets... Not sure how this is just an AMD hurdle?

https://www.pcworld.com/article/340...t-level-with-co-emib-odi-connections.amp.html

They will use what they learned integrating AMD GPU into their CPU and I'm sure ideas they have and anything else at their disposal to pull it off. But to say chiplets are just an AMD problem is silly, it will be a problem for everyone using it, including Intel.
 
For AMD :D



Well, there's an article being referenced for those that need to catch up.



Nope, I'm suggesting that we apply previous performance as a predictor of future performance. Specifically, we can consider both Ryzen and Intel's 10nm stumble to be anomalies. Intel usually (statistically) executes process transitions regularly and without drama, and AMD rarely produces parts that edge ahead. The reasoning for this perspective is that AMD looks like they've wrung as much as they're going to get out of Zen without a significant rework, while Intel has had a new architecture ready for their fabs.

We can certainly predict that Intel will likely get 10nm underway, and if not, 7nm on schedule- that's what they usually do- and we can reasonably question whether AMD will get any more performance (that is, some combination of IPC and clockspeed) out of Zen, because they usually stumble with significant improvements.



But they're not?

[or, strong, consumer-applicable references please]



See above.
According to said article... "A quick calculation of 1.47/1.42 means that even Intel is only predicting an absolute gain of ~3.5% for Ice Lake over current generation systems". Doesn't seem like a huge gain. Yes, IPC went up, but frequencies came down. Great, so we still aren't sure what we'll really see. *IF* they can get frequencies back up, the IPC increase is great. If the changes they made to gain IPC decreased hitting higher frequencies, well, right now that's the only thing we can say because.. refer back to your own link of what we actually know right now from limited testing and Intel made ~3.5% increase to performance from previous gen.

And just one last qoute from your article:
"If it were not for the vast increase in memory speed, moving from LPDDR3-2133 to LPDDR4X-3733, one might have predicted that the Core i7-1065G7 Ice Lake processor and the Core i7-8565U Whisky Lake processor would have performed equally."

Going from ddr3 to lpddr4 is a huge bump in memory speed and *likely* (unproven at this point) could explain a 3.5% difference.
 
We're comparing what we've got.

If we took what the Anad article on Ice Lake showed and applied that to a desktop variant then it would basically be a wash. As it stands right now, the IPC increase cannot make up for the clockspeed deficit. That would mean that 10nm desktop parts would come out barely faster than Coffee Lake refresh on average and would be slower than it in some areas. Productivity might be in line with Zen 2 at that point, but hard to say. It would be a pretty "meh" desktop CPU release.
 
Yeah, I would expect clockspeeds to increase on the desktop over the mobile part.

I'd hope so, but if they're not reaching close to the clockspeeds of CL Refresh the chips are still going to only show minimal improvements in the one area where Intel currently has a big advantage.
 
No, but you did make it sound like an AMD only problem, which it isn't. It's just AMD has been fighting it for longer.

AMD has chosen to fight that fight, largely as a result of not being able to count on their fab partners to produce the monolithic dies that Intel produces with ease.

It's not a bad choice for AMD, all things considered, but it's also not without consequence, which Intel appears to be avoiding at the moment.
 
AMD has chosen to fight that fight, largely as a result of not being able to count on their fab partners to produce the monolithic dies that Intel produces with ease.

It's not a bad choice for AMD, all things considered, but it's also not without consequence, which Intel appears to be avoiding at the moment.

Yes, yes it's so easy for Intel to create those monolithic designs that everyone is using those 10nm desktop chips... Oh wait they don't exist.
 
Yes, yes it's so easy for Intel to create those monolithic designs that everyone is using those 10nm desktop chips... Oh wait they don't exist.

No, they're making them quite successfully on 14nm. 10nm would be cake.
 
Back
Top