Discussion Intel current and future Lakes & Rapids thread

[IntelUser2000]

I doubt the modems are that low margin. Margins aren't really the problem. Apple has lower margins than Intel but many times more revenue. The focus on margins was what lost the original iPhone deal because they believed the volume wouldn't be there. At potential 200 million sales, they won't be complaining.

The biggest issue is modems bring in small revenue.

Working with Apple has other benefits like learning how mobile markets work, and how to make power efficient silicon. We might see them using the same modem across broader segments like in their own products. Expect to see on-package LTE effort to expand.

 

[Dayman1225]

Icelake U 2/4 @2.6GHz appears on Geekbench 4

Scores:

4151 ST
7945 MT

Max Frequency according to this link was 2588 MHz

Also the L1D$ has increased to 48KB/c and L2 increased to 512KB/c like previously rumored

 

[Gideon]

Icelake U 2/4 @2.6GHz appears on Geekbench 4

Hmm, so it has 48 KB L1 Data Cache (up from 32).

 

[Nothingness]

Icelake U 2/4 @2.6GHz appears on Geekbench 4

Scores:

4151 ST
7945 MT

Max Frequency according to this link was 2588 MHz

Also the L1D$ has increased to 48KB/c and L2 increased to 512KB/c like previously rumored

If all of this is true that's a 30% increase in IPC. Getting closer to Apple IPC Let's how high it clocks!

 

[TheGiant]

If all of this is true that's a 30% increase in IPC. Getting closer to Apple IPC Let's how high it clocks!

Base freq 2.6GHz, are you sure it is properly detecting turbo clock?
If it is true we can celebrate that a "Conroe medium" jump

 

[Gideon]

If all of this is true that's a 30% increase in IPC. Getting closer to Apple IPC Let's how high it clocks!

In this test, it has ~16% better IPC than a similar Cannon Lake part (@ 3.1).
It's Ubuntu vs Windows though, which tends to skew the results considerably

 

[.vodka]

Base freq 2.6GHz, are you sure it is properly detecting turbo clock?
If it is true we can celebrate that a "Conroe medium" jump

Good point.

If that score is actually at 2.6GHz then it's great news! Then we can hope 10nm+ can clock somewhat similarly to 14nm++

In this test, it has ~16% better IPC than a similar Cannon Lake part (@ 3.1).
It's Ubuntu vs Windows though, which tends to skew the results considerably

How much so?

 

[Dayman1225]

Base freq 2.6GHz, are you sure it is properly detecting turbo clock?
If it is true we can celebrate that a "Conroe medium" jump

According to GB4 this is the frequencies it was running throughout the test:

 

[csbin]

Geekbench 4 Linux x86 (64-bit)

Core i5-6200U Processor (Max 2.8Ghz)

Scores:

4001 ST
7588 MT

https://browser.geekbench.com/v4/cpu/9862829

 

[jpiniero]

https://browser.geekbench.com/v4/cpu/compare/7230379?baseline=10445533

Test compared to 6100H (2.7 Ghz). There are a couple tests where the Icelake sample loses, but there's also several where it wins by 30% or more.

 

[Andrei.]

If all of this is true that's a 30% increase in IPC. Getting closer to Apple IPC Let's how high it clocks!

https://browser.geekbench.com/v4/cpu/compare/10436977?baseline=10445533

It doesn't look anywhere near 30%. More like 10-20% depending on workload, which matches what it was rumoured to be. It's also going to need to maintain that IPC in high clocks (memory).

https://browser.geekbench.com/v4/cpu/compare/7230379?baseline=10445533

Test compared to 6100H (2.7 Ghz). There are a couple tests where the Icelake sample loses, but there's also several where it wins by 30% or more.

Don't compare Windows vs Linux, they're using different compilers. Linux and macOS is going to be on top in every Geekbench result.

 

[Dayman1225]

So I got a friend of mine to clock down his 6700K to 2.6GHz on the core, 2.8GHz uncore at 2 core and 4 threads Smeltdown Patches disabled, 2133MHz ram, but still lower latency than the ICL U scores.

He ran GB4 two times like this I chose the higher score to compare, links below:

Link 1

Link 2

I then compared this to the ICL U score:

Link to ICL U score

Compared score link

Yes, I know one is running Windows (6700K system) and the other is running Linux (ICL U) but this is the best I could do for you for direct comparison. Not scientific at all of course. So I typed the sub scores into excel made a graph and this is what I got.

Overall I got an average percentage change of:

ST: 24% (rounded to 1 decimal place)

MT: 14.3% (rounded to 1 decimal place)

 

[Arachnotronic]

So I got a friend of mine to clock down his 6700K to 2.6GHz on the core, 2.8GHz uncore at 2 core and 4 threads Smeltdown Patches disabled, 2133MHz ram, but still lower latency than the ICL U scores.

He ran GB4 two times like this I chose the higher score to compare, links below:

Link 1

Link 2

I then compared this to the ICL U score:

Link to ICL U score

Compared score link

Yes, I know one is running Windows (6700K system) and the other is running Linux (ICL U) but this is the best I could do for you for direct comparison. Not scientific at all of course. So I typed the sub scores into excel made a graph and this is what I got.

Overall I got an average percentage change of:

ST: 24% (rounded to 1 decimal place)

MT: 14.3% (rounded to 1 decimal place)

Nice!

 

[Dayman1225]

I have found an m3-7Y30 that was also running Ubuntu albeit a slightly older release, updated graphs below:

Average difference:

ST: 15%

MT: 15.5%

Link to compared results

M3 score by it self

 

[.vodka]

Let's do some crude and probably horribly wrong numbers to compare these two

@csbin's 6200U @ 2.8GHz sample vs that Icelake 2.6GHz sample, both running Linux. It's GB4 4.2.2 vs 4.3.0 though...

ST
i5-6200u: 4001/2800 = 1.4289 points per MHz
Icelake sample: 4151/2600 = 1.5965 points per MHz

1.5965/1.4289 ~ 11.72% ST increase over Skylake on average.

Not too bad, that's in line with the 5-10% average increases we've been getting since Nehalem.

 

[Gideon]

Well Zen2 or Zen3 really need to be good, to compete, unless Intel fails miserably with 10nm clocks

 

[Gideon]

I have found an m3-7Y30 that was also running Ubuntu albeit a slightly older release, updated graphs below:
Link to compared results

What were the clocks the m3-7Y30 was running on? It's base clock is 1 Ghz

 

[Dayman1225]

What were the clocks the m3-7Y30 was running on? It's base clock is 1 Ghz

Link to where you can find this

adding .gb4 to the end of the Geekbench score link brings this up

 

[Gideon]

Link to where you can find this

adding .gb4 to the end of the Geekbench score link brings this up

Thanks!
The closest I could find (also ubuntu) is this:
https://browser.geekbench.com/v4/cpu/compare/10417578?baseline=10445533

But it clocks minorly higher, and has similar scores to m3-7Y30.

Overall looks impressive (though we obviously need more datapoints). There are also 2 things to keep in mind that somewhat inflate the Icelake Score. 1. It has very good bandwidth numbers (quite a few tests are sensitive to it) and it's AES throughput is phenominal.

It's still pretty good, in whatever way you slice it. Now we obviously need more results (and also other benchmarks) to get a better idea overall.

 

[TheGiant]

Thanks!
The closest I could find (also ubuntu) is this:
https://browser.geekbench.com/v4/cpu/compare/10417578?baseline=10445533

But it clocks minorly higher, and has similar scores to m3-7Y30.

Overall looks impressive (though we obviously need more datapoints). There are also 2 things to keep in mind that somewhat inflate the Icelake Score. 1. It has very good bandwidth numbers (quite a few tests are sensitive to it) and it's AES throughput is phenominal.

It's still pretty good, however way you slice it. Now we obviously need more results (and also other benchmarks) to get a better idea overall

Now we need to get that released. With turbo like 4GHz it can be very nice. The AES improvement were announced and long time expected.

 

[Nothingness]

https://browser.geekbench.com/v4/cpu/compare/10436977?baseline=10445533

It doesn't look anywhere near 30%. More like 10-20% depending on workload, which matches what it was rumoured to be. It's also going to need to maintain that IPC in high clocks (memory).

I picked against a i8550u running Linux which does ~4800 @ 4 GHz
http://browser.geekbench.com/v4/cpu/compare/10449968?baseline=10445533
That gives 1200 / GHz vs 1600 / Ghz, a 30% increase.
Isn't Kaby Lake Intel latest uarch ?

EDIT: Silly me, I forgot to sort the results...
So that is 1350 vs 1600 so ~20%.
Sorry for the mistake

 

[jpiniero]

Well Zen2 or Zen3 really need to be good, to compete, unless Intel fails miserably with 10nm clocks

The max clocks are definitely going to be worse than Intel's 14 nm, hence why Intel is sticking with Coffee Lake for desktop next year. Whiskey Lake U can turbo to 4.6, so if they manage that, that would be a decent upgrade with the IPC gain for mobile.

 

[yuri69]

So it got 50% bigger L1D, 100% bigger L2 and 30% bigger L3. Plus they throw a lot of crypto units to the mix.

This is the way they spent the transistors the 10nm brought to them.

 

[Dayman1225]

So it got 50% bigger L1D, 100% bigger L2 and 30% bigger L3. Plus they throw a lot of crypto units to the mix.

This is the way they spent the transistors the 10nm brought to them.

L3 is the same size

 

[TheGiant]

So it got 50% bigger L1D, 100% bigger L2 and 30% bigger L3. Plus they throw a lot of crypto units to the mix.

This is the way they spent the transistors the 10nm brought to them.

If it does more IPC and can clock around 5GHz (what is IMO the biggest issue now) I don't care