Printable View
Over at toms hardware they have an excellent article displaying the efficiency of all the processor architectures from 2005. It compares Intel architectures against it's self and AMD and vice versa.
http://media.bestofmicro.com/I/O/298...al/overall.png
An excellent read
So when Bulldozer comes out, how will they define a 'core'? ;)
Pentium 4, you so crazy! The efficiency gain between NB & Conroe really is quite astonishing and impressive.
Efficiency is the wrong word for it really - NB was designed for high frequency and Conroe was designed for IPC - comparing either of them by the other's metric is going to show them up badly. NB was also available in frequencies up to 2x regular Conroe chips, so the overall gain in speed wasn't as dramatic as this chart shows.Quote:
Originally Posted by Terbinator
After having a quick look at the chart in table one, it seems a Sandy Bridge core is around 28% faster than a Phenom II core. TBH,it is not as large as I expected. It seems to be significantly less than the IPC difference between the Athlon 64 and the Pentium 4. It really does make think at times why people whinge so much about the IPC of the Phenom II TBH!!
Anyway,like kalniel stated before the chart does not take into consideration whether a CPU is designed for higher or lower clockspeed targets.
So AMD barely reached the IPC of Conroe with Thuban - an architecture 3 years newer? That pretty much says everything you need to know about just how good Conroe was! That said, it also gives you a good impression of how well AMD have done tweaking K8 in later iterations, to be able to make up the IPC despite having one less int pipeline.
This is going to make a similar comparison to Bulldozer very interesting - if AMD's switch to 4-issue can approach the gains Intel got from P4 - Conroe, they could get very close to Sandy Bridge with it. That said, K8 was always targeted at lower clock speeds & higher IPC than Netburst, so they probably don't have as much leeway as Intel had in that regard.
Would have liked to see a few done at 1 GHz or something, with a Pentium 3 or P3-S, Pentium-M, Athlon XPs, and a Northwood/Williamette Pentium 4.
Interesting to see just how outdated the Pentium 4 has become, now we are back up to 3GHz+ again that inefficient core really looks slow, the Athlon 64 architecture has aged better.
I'm actually surprised at just how much IPC gain both companies have squeezed out of their recent architectures. Intel have made 25% from Conroe to Sandy Bridge: AMD have squeezed 20% from Windsor to Thuban, but of course Windsor was a fairly late K8 architecture and I assume they made IPC gains between Clawhammer (2003) and Windsor (2006) ... all without having to make significant changes to the core. I think most people would be pretty chuffed with a 25% IPC gain from iterative improvements...Quote:
Originally Posted by kingpotnoodle
If I remember clock for clock actually got slightly worse - the latencies increased, but they gained clockspeed and DDR2 controller.Quote:
Originally Posted by scaryjim
I knew AMD weren't doing great on the IPC front, but I'm quite shocked at how far down the list they are.
Effectively, even with my years-old Yorkfield processor, I'm still doing better than I would be with a new AMD quad.
I probably should've known this already, but I haven't actually looked too carefully at any of the recent AMDs, having no occasion to actually buy them. Scary how far ahead Intel are.
Nah, that was just the change from DDR to DDR2 memory at the same frequency.Quote:
Originally Posted by kalniel
The cores certainly improved.
http://www.anandtech.com/show/2139/3Quote:
Originally Posted by DanceswithUnix
Was what I was remembering - which actually concerns the 65nm chips, my mistake, not the 90nm.Quote:
Originally Posted by anandtech
Quote:
AMD has given us the official confirmation that L2 cache latencies have increased, and that it purposefully did so in order to allow for the possibility of moving to larger cache sizes in future parts.
It's all well and good looking at IPC but it's like comparing car engines based purely on torque, ignoring fuel economy, power and how fast the engine can spin.
Just like looking at Clock speed is like comparing engines based purely on how high they can rev.
Sure, but we're nerds and we like pointless benchmarks ;)Quote:
Originally Posted by badass
Yeah, the slower cache to allow for a size increase that never happened. Not their best move :)Quote:
Originally Posted by kalniel
I was thinking about the 130 to 90nm shrink, with SSE3 & improved DDR interface.
And in full nerd mode, I would say this is more like BHP per litre (ie work done for a given capacity). Another benchmark useless on its own, but I rather like it :DQuote:
Originally Posted by kalniel