backtop


Print 37 comment(s) - last by outsider.. on Jul 26 at 6:07 PM


AMD's "Torrenza" platform mock up
AMD opens up the Opteron architecture to other microprocessor R&D companies

Today AMD unveiled what it calls the evolution of enterprise level computing, called Torrenza. The new platform, says AMD, will utilize next-generation multi-core 64-bit processor that have the capability to work alongside specialized co-processors. DailyTech previously reported that AMD was considering working with co-processing design firms such as ClearSpeed, to develop and design platforms that would be able to utilize specialized processors for specific duties alongside the general host processor in a traditional Opteron socket.

With Torrenza, AMD has designed what it calls an open architecture, based on the next wave of Opteron processors, which allow what AMD calls "Accelerators." Using the add-in accelerators, a system will be capable of peforming specialized calculations similar in fashion to the way we use GPUs today.

Because of its flexibility, the HyperTransport protocol allows a multitude of co-processor designs that are already compatible with systems on other platforms. For example, with Torrenza, specialized co-processors are able to sit directly into an Opteron socket, and communicate directly with the entire system. During the conference, Cray Inc. noted that it had worked with AMD to design a system where a system can contain even up to three different co-processors, all dedicated to specialized tasks.  All three processors would communicate directly with Opteron processors and the system chipset harmonously. The open-ended nature of Torrenza will allow companies to design specialized processors to plug-in and work with Torrenza-enabled Opteron systems.  Although AMD acknowledges many of these applications can run off PCIe and other connection technologies, Torrenza emphasizes HT-3 and HTX in particular.

AMD representatives said that because of the archicture, Torrenza allows very low latency communication between chipset, main processor and co-processors. According to both Cray and AMD, applications can be written in a way where all the variouis processing architectures are recognized and are fully usable. Torrenza-aware applications are on the way said Cray, but the company did admit that developing them was very much "rocket-science".


Comments     Threshold


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

interesting
By hwhacker on 6/1/2006 9:53:36 PM , Rating: 2
This certainly would be interesting if it could be intergrated down to the desktop level, not to say what could be done of the grand server level isn't exciting.

We know both ATi/Nvidia have at least dabbeled in the idea of a gfx socket, or an AIB with set ram, but a socket so one could keep their current card, but change the gpu without paying for the RAM,PCB, etc all over again. It would be interesting if GPUs/PPUs etc could use this tech, and we could just put them in a socket. I wonder where all the memory bandwidth would come from though...Would it use system memory, or would we use the pci-e cards just for the next-gen of G-RAM? I dunno. :P


RE: interesting
By cnimativ on 6/2/2006 12:10:39 AM , Rating: 1
uh, i doubt ati/nvidia will use that socket because:
1) memory architecture that ati/nvidia uses is different to that of amd's.
2) ati/nvidia are higher bandwidth memory (gddr2/3) compare to amd.
3) chips design for that specific sockets will still be reworked to fit in intel's system.


RE: interesting
By mino on 6/2/2006 12:55:40 AM , Rating: 2
Yep.
However HTX Graphics cards would be cute. :)

About the lack of PCIE peripherals:

AFAIK to implement PCIE on your own designs isn't a "rocket science". But close to it.
Even HT1.0 is easier to do (at least the "low speed" 400 to 600MHz 8bit variants).

HTX for desktop :) we all need 10Gb infinipath for our downloads !!!

;)


RE: interesting
By PrinceGaz on 6/2/2006 1:19:11 AM , Rating: 2
You won't see a GPU socket on a mobo because the tight manufacturing tolerances needed for routing high speed wide (256-bit) memory buses to dedicated memory sockets is not practical, and system memory bandwidth and latencies still lag far behind what even mid-range GPUs use.

What is needed is some way of providing a fast bus to a graphics-core that has its own high-speed memory. The ideal situation would be some sort of card that combines the GPU with the dedicated memory, which could be inserted into a compatible slot on a motherboard.


RE: interesting
By Gnarr on 6/2/2006 6:04:50 AM , Rating: 3
quote:
What is needed is some way of providing a fast bus to a graphics-core that has its own high-speed memory. The ideal situation would be some sort of card that combines the GPU with the dedicated memory, which could be inserted into a compatible slot on a motherboard.


Also known as a "Graphics Card", wich is inserted into a compatible PCIe or AGP slot...


RE: interesting
By LCC2286 on 6/2/2006 11:52:59 AM , Rating: 2
LOL

I'm sure he is talking about GPUs with embedded memory. $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ $$$


"Nowadays you can buy a CPU cheaper than the CPU fan." -- Unnamed AMD executive

Related Articles













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