Maybe AMD can finally be quiet about their "extreme bandwidth"?
Quote :
An Itanium processor is implemented in 8M 65nm CMOS and measures 21.5×32.5mm2. The processor has four dual-threaded cores, a system interface and 30MB of cache. Quickpath high-speed links enable peak processor-to-processor bandwidth of 96GB/s and peak memory bandwidth of 34GB/s.
Quote :
QuickPath whiz and analyst David Kanter is more willing to tackle the Hypertransport debate based on information he uncovered last year.
He tells us, "It looks like Tukwila's QPI links are running at 4.8GHz, which is about the same speed as Hypertransport 3 (maximum speed of 5.2GHz). Realistically, Intel will pack quite a bit more bandwidth on - because they are using 4+1 QPI links (4 to talk to other processors and 2 half links for I/O), compared to the 4 HT3 links that AMD will be using in future MPUs (that's right, no HT3 in the MP version of Barcelona). What's most impressive about Tukwila is the memory bandwidth - it has the same bandwidth as a full 4 socket Opteron system, all in one socket.
Dear Jebus!!!!! The same bandwidth as a 4P AMD Opteron system? Imagine this as a 4P system. That would equal 4x4P Opteron systems or 64 Quad core Opterons but only using 16 cores.
Now I wounder what kinda virtualization/scale you can get with that. Any expected release date for it?
So with QPI just on the horizon, I wonder how AMD's going to spin their products in their favor? I was trying to come up a few advantage against Intel's QPI and Nehalem, and couldn't find any...
I personally think AMD jumped on the IMC bandwagon too soon and got their heads all big and puffy. While they did that Intel was developing their IMC and version of directconnect but wanted it to be faster, stronger and more powerfull.
This is like taking a Honda Civic tuned to 300HP and pitting it agains the 2007 Lingenfelter(not sure if I spelled that right) Corvette that pumps out 1200HP.
And Bob I agree I am not sure why they didn't go to 45nm but don't worry. There will be a 45nm Nehalem for servers that will have quad channel DDR3. MMMM..... quad channel....
It says Itanium or IA-64 based architecture rather then the Core 2 X86 based architecture. They probably have no need to go 45nm on the Itanium architecture for some reason or another and are just using 65nm.
Sounds interesting, as this is the same interface Nehelem is supposed to be using, maybe we can get similar bandwidth with that too.
It says Itanium or IA-64 based architecture rather then the Core 2 X86 based architecture. They probably have no need to go 45nm on the Itanium architecture for some reason or another and are just using 65nm.
Sounds interesting, as this is the same interface Nehelem is supposed to be using, maybe we can get similar bandwidth with that too.
Well according to wikipedia.org the high end desktop Nehalem will have 1x Quickpath which will be targeting a bandwidth of 24GB/s-32Gb/s. The server Nehalem will have 4x Quickpath so just double the numbers.
Also the high end desktop version of Nehalem will support tripple channel DDR3 as to be able to fill and use some of that delicous bandwidth where as the server one will have Quad FB-DIMM2. I can't wait. I want it to be November now!!!!!!!
Message edited by jimmysmitty on 01-31-2008 at 12:35:46 AM
It says Itanium or IA-64 based architecture rather then the Core 2 X86 based architecture. They probably have no need to go 45nm on the Itanium architecture for some reason or another and are just using 65nm.
Sounds interesting, as this is the same interface Nehelem is supposed to be using, maybe we can get similar bandwidth with that too.
Wrong logic. It is *never* a good idea to stay on a larger node for economies of scale. Have you seen the size of that monster? 21x32mm? Yields would be significantly better at a smaller node.
The current Itaniums are on 90nm. The real reason it's on 65nm is because 45nm is just out. This is a new design. You never/shouldn't put a new design on a new process until the process is rock solid. That's for Nehalem and Silverthorne later this year. There probably wasn't room to debug yet a third processor in minimal pilot lines... and run full-bore Penryn production.
Plus there just isnt' a high-demand for this chip. Nobody is buying them. (much). MIght as well use them to give the 65nm plants something to do until they make the conversion to 32..... =)
I personally think AMD jumped on the IMC bandwagon too soon and got their heads all big and puffy. While they did that Intel was developing their IMC and version of directconnect but wanted it to be faster, stronger and more powerfull.
This is like taking a Honda Civic tuned to 300HP and pitting it agains the 2007 Lingenfelter(not sure if I spelled that right) Corvette that pumps out 1200HP.
And Bob I agree I am not sure why they didn't go to 45nm but don't worry. There will be a 45nm Nehalem for servers that will have quad channel DDR3. MMMM..... quad channel....
On the contrary, I think AMD made the right choice of jumping on the IMC bandwagon. Through that, and a successful K8 architecture (thanks to Jerry Sanders), AMD was able to secure a foothold in the lucrative server market. But in reality, IMC is only needed when you're going 4P and above, and that market is really niche.
Wrong logic. It is *never* a good idea to stay on a larger node for economies of scale. Have you seen the size of that monster? 21x32mm? Yields would be significantly better at a smaller node.
The current Itaniums are on 90nm. The real reason it's on 65nm is because 45nm is just out. This is a new design. You never/shouldn't put a new design on a new process until the process is rock solid. That's for Nehalem and Silverthorne later this year. There probably wasn't room to debug yet a third processor in minimal pilot lines... and run full-bore Penryn production.
Plus there just isnt' a high-demand for this chip. Nobody is buying them. (much). MIght as well use them to give the 65nm plants something to do until they make the conversion to 32..... =)
I'm really curious, what's the yield on those baby? I'm guessing one chip per wafer...
This is like taking a Honda Civic tuned to 300HP and pitting it agains the 2007 Lingenfelter(not sure if I spelled that right) Corvette that pumps out 1200HP.
Just to let you know, my 2002 Civic Si turbo (338hp) not only beat my brothers '03 Viper that he tuned up quite a bit from its stock HP of something like 533hp it's probably something about 800+ now not only in mile time, but also in 0-60, 60-0, and cone test....and just so that he wouldn't get pissed, i let him drive both.
so please dont bash my civic please
and for some reason i dont know how much i trust this article, its making AMD to sound about as competative as Via in the x86 market, and for some reason i just don't believe that even AMD would sink that low.
Just to let you know, my 2002 Civic Si turbo (338hp) not only beat my brothers '03 Viper that he tuned up quite a bit from its stock HP of something like 533hp it's probably something about 800+ now not only in mile time, but also in 0-60, 60-0, and cone test....and just so that he wouldn't get pissed, i let him drive both.
so please dont bash my civic please
On Man, I think I saw your civic once. Silver with red trim?
Just to let you know, my 2002 Civic Si turbo (338hp) not only beat my brothers '03 Viper that he tuned up quite a bit from its stock HP of something like 533hp it's probably something about 800+ now not only in mile time, but also in 0-60, 60-0, and cone test....and just so that he wouldn't get pissed, i let him drive both.
so please dont bash my civic please
and for some reason i dont know how much i trust this article, its making AMD to sound about as competative as Via in the x86 market, and for some reason i just don't believe that even AMD would sink that low.
Not to get off topic but Vipers suck. Most every Vette can smoke one. In fact look up the 01 Lingenfelter 427tt Vette. Twin turbo 800+HP and 850+ tourque. 0-60 in 1.92 seconds and the 1/4 mile in 8.95 seconds. Doubt yours can top that. Now I personaly prefer the Ford Contour SVT as it can take a BMW325i on stock and smoke it and be pumped up to 700+HP on a little 2.5l V6.
But my meaning was simple. The Civic with 300HP is AMD and the Lingenfelter Vette( http://www.fast-autos.net/vehicles [...] win-Turbo/ ) is Intels new QPI. AMD would basically lose their "better memory bandwidth" claims to this if their specs are true.
Message edited by jimmysmitty on 01-31-2008 at 03:51:14 AM
not going to bash your civic im sure it does what you say it can, but i bet your brother has fists of ham and 2 left feet. I romped on a ferarri in my mustang on a road track does that mean my cars faster no just means the driver was a bit afraid to ruin his 200k toy. Plus who cares what itanium does, anyone here running it I work in the it field and have never seen one with the countless servers ive seen none have been itanium.
Ps vipers do suck there brakes are terrible just terrible like trying to stop a train with brakes you get on a bicycle.
We have an itanic at work. They wont let me near it.
Maybe it's because I laughed at the boss for buying it. It spends a lot of time in x86 emulation mode. A P3 would have been faster.
Dont expect too much from the QPI on desktop, the fsb is still holding it's own. The IMC wont be much help either, since core2 has so much cache, and does prefetch so well.
Then again, you can bet Intel will make it work better than AMD could ever do.
