backtop


Print 79 comment(s) - last by PrezWeezy.. on Apr 3 at 1:57 PM


"Torrenza" systems will accept both multi-core, accelerators, or "Fusion" processors
AMD sheds more light on accelerated processing projects

This week at CeBIT 2007, AMD revealed more details about its "accelerated computing" platform, codenamed Torrenza. AMD's goal behind Torrenza is to create a platform where application-specific processors can interact cost effectively and offer better performance than a general purpose CPU, while remaining compatible with off the shelf platforms.

AMD guidance revealed this week that future processors will also have integrated "accelerators" embedded into them. A Torrenza system will have at least two sockets, and both will accept accelerators and accelerated CPUs.

One accelerated-processor project on AMD plate, slated for 2008 under the codename Fusion, and combines a dedicated GPU or GPU accelerator onto the same package or even the same silicon die as the main CPU. AMD has already set the ground-work for Fusion processing with its Stream Computing initiative -- utilizing ATI-based graphics adaptors for heavy number crunching. 

Other Torrenza ready projects are also coming to light.  Clearspeed announced its CSX600 math-coprocessor plug-in last year, with the stated intention of creating a socket plugin version for Torrenza.  Los Alamos National Labs is currently building the world's fastest supercomputer, Roadrunner, with Opteron and Cell processors on the Torrenza platform.

Torrenza is not just locked within the compounds of the CPU sockets. According to AMD, Torrenza systems will accept accelerators in a PCI-Express interface too, allow for multiple application specific accelerators to access system memory and processor functions directly.  Mercury systems announced a PCIe plug-in accelerator late last year.

While Torrenza is well on its way to seeing daylight, Intel is also working on its own open architecture platform. Notorious for keeping its CPU platform a closely guarded technology, Intel indicated that it was working on a competitive technology to AMD's HyperTransport, dubbed CSI, allowing direct CPU and memory access.

Intel guidance suggests the company will announce its Torrenza competitor sometime in mid-2008.


Comments     Threshold


This article is over a month old, voting and posting comments is disabled

By Viditor on 3/19/2007 9:35:28 AM , Rating: 2
quote:
I think torrenza could potentially have an impact on certain markets, but I am not sold on the idea that it will have a major impact anytime soon on most markets or end users

Soon is relative, but there are many things that Torrenza offers for end users...for example a PhysX coprocessor.

You need to remember what Torrenza and Fusion actually are...

Torrenza = cHT...that's it. The coherent HT link to the cache just like other CPUs get in a multi-chip board has been opened up to coprocessors.

Fusion = Torrenza on-die. It's basically including the graphics coprocessor into the die itself.

These sound fairly simple, but if you think about the stages you bypass by doing all of this, you can see just how powerful this simple process will be, even for main stream (though more so for servers!).


By KillerNoodle on 3/20/2007 1:03:17 PM , Rating: 2
quote:

Fusion = Torrenza on-die. It's basically including the graphics coprocessor into the die itself.

Correct, but the Fusion's main target will most likely be for the scientific industry, while they will need graphics it will not be #1 in all cases. Co-processors in the form or Math, Physics, or even Cryptographic co-processors are going to be the top sellers.

What is even more impressive is the way the "co-processors" or "Accelerators" are implemented. Be it Just Torrenza with the socket, chipset, or add-in implementations (Add-in including the HTX[HyperTransportX? a slot for HT interaction?] or the PCI-ex slot add-in) the options will provide for a number of routes to implement these accelerators. Then with the added Fusion it provides that final jump to the on die accelerator.

I fear that the Fusion implementation is the worst if it is unchangeable. It will hinder upgrade-ability since you have to buy a new processor and co-processor at the same time instead of one or the other. Think of when AMD made the switch from DDR to DDR2. It required a new processor instead of just new memory and motherboard = costly. Then again these aren't going to be aimed at the budget sector for quite sometime but one would hope that they look to the future and plan on where they want to go with the series.

Example of what fusion might entail for product options
AMD F6400P AMD F6600P
AMD F6400M AMD F6600M
AMD F6400C AMD F6600C
AMD F6400G AMD F6600G
AMD F6400G2 AMD F6600G2
AMD F6400_ AMD F6600_
With each of the trailing letters representing a different co-processor and the 6XXX numbers are representing the speed of the general CPU. This will get very troublesome, let alone the implementations and the different API that each co-processor will need.

I foresee Headaches ~:-/


By ZmaxDP on 3/20/2007 2:49:57 PM , Rating: 3
I think we're undervaluing the Torrenza part of the equation here. I thing that Fusion has a lot of possibilities, but I doubt that AMD or Intel would be interested in having more than one or two versions of a given processor grade. It is likely that the Fusion strategy will be used on things where there is a predictable high demand. One likely candidate is system on a chip. Another is a processor with a major FP boost. But I don't think they'll make a list of every conceivable possibility for a processing core and then try to produce every possible combination thereof.

That is what Torenza is for. You can premier a new core variation as a single die that connects vie Torrenza. If that particular core sees high enough market demand to justify it then AMD can transition it to the Fusion strategy. Otherwise, they can keep making money of niche markets for the remainder of the options.

In other words at worst you'd see:

AMD 6400 General Purpose
AMD 6400 FX Gamer Edition (physics, AI co-proc)
AMD 6400 SC Scientific Calculations (FP co-proc)
AMD 6400 CR Cryptography (who knows)

Point being, it is more likely that they would identify target markets and create a fusion CPU for the market. The particulars (2 physics co-processors, two AI co-processors, and 4 AMD K10 6400 Cores) would be covered in the specifications. Still a lot of choices, but more self-evident. Even then, it is likely the cryptography variant would just be Torenza based. Same goes with anything that might need upgrading separate from the general purpose CPU.

I just doubt that Torrenza is going to vanish into the mists and that Fusion will rule all...


"Well, there may be a reason why they call them 'Mac' trucks! Windows machines will not be trucks." -- Microsoft CEO Steve Ballmer

Related Articles
IBM's Next-Generation Supercomputer
February 28, 2007, 1:52 AM
AMD Reveals More "Fusion" Details
November 17, 2006, 10:01 AM
Intel Responds to AMD's "Torrenza"
September 27, 2006, 9:30 PM













botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki