Then its all about the cache so to say :p:Quote:
Originally Posted by Macadamia
That also explains the "unrealistic" 25% increase from Phenom to Phenom 2 in this test. Not even to talk about the 100% change to the 7000 series Xeons.
Then its all about the cache so to say :p:Quote:
Originally Posted by Macadamia
That also explains the "unrealistic" 25% increase from Phenom to Phenom 2 in this test. Not even to talk about the 100% change to the 7000 series Xeons.
The first 2 AMD Phenom II that will be released on 9th of january will be :
Source here and transalted here !Quote:
1. AMD Phenom II X4 940 Black Edition, Part Number : HDZ940XCGIBOX with 125W TDP, on Socket AM2+. The default speed will be 3000MHz with 8MB Cache.
2. MD Phenom II X4 920, Part Number : HDX920XCGIBOX with 125W TDP, on Socket AM2+. The default speed will be 2800MHz with 8MB Cache.
:)
That's what I thought, and obviously that is what is indicated, but I think you both are talking about the performance numbers in the linked benchmark.Quote:
Originally Posted by Macadamia
:confused:
why so much confusion about the zdnet article... numbers just fit fine...
Queen:
Dunnington:
12core: 57432
1core: 5635
Scaling: 10,2
the only odd thing are the julia benches. :p:
BTW, a Chinese website showed a 3.2Ghz QX9770 getting 24k while a 3.5Ghz Q9300 (6MB Cache) @ 3.5Ghz OC get 20k.Quote:
Originally Posted by Shintai
More food for thought? :p:
Nah, that fits the cache theory :DQuote:
Originally Posted by Macadamia
Yes, the core has gone some optimizations such as fiing coldbug and allowing high clocks(dunno if it is due to 45nm or an actual march fix). IPC-wise there is not that much, except 3x bigger L3 and lower latencies(?).Quote:
Originally Posted by Chad Boga
Do you have some info to link on that? It would be nice to know more about the IPC changes.Quote:
Originally Posted by Calmatory
Does anybody know if AMD still uses SOI in 45nm.? It wasn't mentioned in the press release. The lack of cold bug is pretty fishy :)
Yes it is still SOI.Quote:
Originally Posted by kl0012
Yes they do(use SOI).SOI is even planned for high perf.(read server/desktop) parts on 32nm while bulk and highK is for GPUs and low power CPU parts(mobile/netbooks).Quote:
Originally Posted by kl0012
i would really...but don't know if its worth the trouble if only 2 peoples are interested :)Quote:
Originally Posted by JumpingJack
I'm really interested,but those phenomII results from that blog are kinda fishy so it would be pointless for you to waste your time if we can't compare the results.Thanks anyway :).Quote:
Originally Posted by JumpingJack
You clearly misunderstood me.Bulk is for GPUs and highk is for low power CPU partsQuote:
Originally Posted by LOE
<-respectively.Quote:
while bulk and highK is for GPUs and low power CPU parts(mobile/netbooks).
Err... why can't 32nm SOI HP use HKMG too?
Weren't IBM-AMD working in that direction?
Whether the GPUs have HKMG is up to timeframe, since I think HKMG does not need a big transition in terms of design.
Quote:
Originally Posted by STaRGaZeR
I think you need to calmly wait and see, like for all of us there is an NDA on any numbers being published.
Try to remain focused on the main context of the thread, which was Phenom2 is hitting well over 5GHZ chilled. When the numbers surface detailing performance we will start the debate all over gain.
Nothing much, based on rumours AMD saying "up to 25 %" IPC improvements for Shanghai over Barcelona(NOT for Deneb over Agena, however!). Some leaked Deneb tests whow approx. 5-10 % increases over Agena(C1 AFAIK). It is hard to say how much it is faster "on average" because it is solely up to the tested apps. E.g. I could test only with programs which yield biggest gains and claim that "+15 % increases" whereas usual desktop apps(E.g. WinRAR, divx/video en/decoding, photo editing, audio en/decoding etc) show less than 5 % increases on average (NOTE: I am not claiming that those apps would show less than 5 % IPC improvement, that is just a simplified example! And I have no clue about the IPC improvements on different apps, other than that it should be from 5-10 %, quite average I'd say, sometimes less, sometimes more.).Quote:
Originally Posted by G0ldBr1ck
I'd say that +7 % IPC improvement is slightly pessimistic one. I'd look forward to 8-10 %, but it varies between apps so much that it is impossible to prove/tell/check anyway.
I have been seing this statement around the web quite a bit, not sure what its based on but it sounds way out there..........
Quote:
Beyond that, from a clock-for-clock performance perspective, optimizations have been made to the CPU pipeline, including improvements to branch prediction, pre-fetch mechanisms and TLB (translation look-aside buffer) optimizations. In total, along with larger cache sizes (now 512KB L2 cache per core with a 6MB L3 cache) and higher clock speeds, AMD claims the new Phenom II X4 should offer a 30 – 40% performance increase over all.
Branch prediction - now that was one of the things CPU Queens is supposed to be testing!
(But it ended up loving cache more than BP. :p:)
As much as I want to believe it, I'm thinking that is just a wet dream. Now, it would be nice if I was proven wrong in January but I'm not holding my breath.Quote:
Originally Posted by G0ldBr1ck
Any clue on what has been done for the branch predictor? Core architecture as great BP, for example Core architecture can predict branches perfectly up to 64 loops, while K10 can only predict perfectly up to 9 loops.
I think that "AMD claims the new Phenom II X4 should offer a 30 – 40% performance increase over all." does mean that with higher clocks there can be up to(At least thats what they are saying when/if the claims don't meet with the reality, obviously :p:) 30-40 % performance gains(for same price?), not that IPC improvements would be so great. Don't try this, Shi....tttt. :)
WHY everybody respect the NDA from AMD and nobody respects NDA from Intel??????????? :confused:
BTW, for anyone having trouble with trolls or just someone who isn't exactly trolling but serially upsets you, this forum does sport Annoyance-Be-Gone Technology that is quite helpful for staying calm.
---
Yes, SOI and HKMG are not exclusive. I wonder if AMD would see even further benefits to their already spectacular 45nm process if they utilized HKMG in conjunction with their existing SOI goodness.
it seems AMD has rewrote the book when it comes to SOI.........they may end up having no use for HKMG on desktop CPU's.
How many people saw the results? How many enthusiast people care about AMD? People are after Nvidia and Intel everywhere. :p:Quote:
Originally Posted by vbo