The Intel Atom Thread

Page 210 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

IntelUser2000

Elite Member
Oct 14, 2003
6,841
1,395
136
That's an improvement of 12.5%, that's not that much (especially considering the skipped Gen 10).
That's incorrect reasoning.

You got 2x gains going from 24EU Gen 9 to 64EU Gen 11, while the bandwidth increased 55%. That's because things don't scale linearly.

Bandwidth increases only by 33% while the shader throughput increases by 77% for Elkhart Lake.

If you want double the performance in a balanced uarch, you need to double the bandwidth, the fillrate, and the shader throughput.

You agree yea?
 
Last edited:
  • Like
Reactions: mikk

IntelUser2000

Elite Member
Oct 14, 2003
6,841
1,395
136
It means Gen11 is making better use of available bandwidth.
Something like that.

Prescott Pentium 4 was not faster than Northwood.

Yet on Celerons, it was 25-30% faster per clock.

Same thing with current CPUs where Core class chips are seeing 15-20% gains but Atom-class chips are getting 30%.
 

moinmoin

Golden Member
Jun 1, 2017
1,613
1,509
106
That's nice and all. My point was that whereas the platform balancing optimizations are all commendable the architecture improvements appear to be negligible. That the Cores are even worse doesn't make it any better.
 

IntelUser2000

Elite Member
Oct 14, 2003
6,841
1,395
136
That's nice and all. My point was that whereas the platform balancing optimizations are all commendable the architecture improvements appear to be negligible. That the Cores are even worse doesn't make it any better.
Architectural improvements(larger caches, tiling) are why it can perform greater than specs suggest.
 

IntelUser2000

Elite Member
Oct 14, 2003
6,841
1,395
136
Gemini Lake Refresh J5040 gets 95 points in Cinebench R15 1T. That's a 15% improvement. That's a 3-4% higher than specs suggest. It must not having been reaching boost as reliably.

Improvement will be less in MT. Seems to be 5-7%.
 

DrMrLordX

Lifer
Apr 27, 2000
15,788
4,773
136
Gemini Lake Refresh J5040 gets 95 points in Cinebench R15 1T. That's a 15% improvement. That's a 3-4% higher than specs suggest. It must not having been reaching boost as reliably.

Improvement will be less in MT. Seems to be 5-7%.
Hmm, do we have any idea what kind of average clock it's hitting in ST and MT?
 

vstar

Member
May 8, 2019
29
13
41
First 10nm Tremont SOC?

I think this is the first news of a Tremont SOC (other than Lakefield) to come out this year.

I didn't see anything else about the technical specs for the P5900, but I am curious to see what the ST/MT clock speeds are.
 
  • Like
Reactions: coercitiv

Bouowmx

Senior member
Nov 13, 2016
975
366
116

Up to 24 cores, 2.2 GHz, 27 MB LLC (4.5 MB per 4 cores), 2x DDR4-2933, 32x PCIE 3
 
  • Like
Reactions: vstar

IntelUser2000

Elite Member
Oct 14, 2003
6,841
1,395
136
Looking at Lakefield die, I estimate each Tremont cores to be 0.7mm2 without the 512KB L2 cache.

So its actually comparable to ARM cores with similar uarch/performance using 7nm process.

It's only their Core cores that are very large. Speculation that it must be larger to allow for high clocks could very well be true. The two AVX-256 units in Sunny Cove are almost the size of one Tremont core along with its 512KB L2 cache.
 
Last edited:

Thala

Senior member
Nov 12, 2014
962
306
136
So its actually comparable to ARM cores with similar uarch/performance using 7nm process.
It is close to impossible to design an x86-64 CPU with the same relative performance in the same power and area range compared to ARM...so i pretty much have doubt in your claim.

But yes, depending on your cell library, the high performance cells can be larger - this holds in particular for SRAM cells. In the past for instance you had different track heights. For N7 you can mostly chose different Vth cells and not different track heights.
 
Last edited:

TheGiant

Senior member
Jun 12, 2017
659
262
106
It is close to impossible to design an x86-64 CPU with the same relative performance in the same power and area range compared to ARM...so i pretty much have doubt in your claim.
relatively impossible to tech calculated output or the current and "D+1" software connected to it?
I doubt ARM can replace x86 if AMD and Intel deliver
 

ASK THE COMMUNITY