Print 39 comment(s) - last by Viditor.. on Sep 22 at 6:07 PM

Torrenza technology connects the industry says AMD

Today AMD announced that it plans to push its Torrenza technology into an open space, creating a unified socket-system that will be used by several top tier manufacturers. The Torrenza platform also allows companies to develop their own processor cores and other related processors, and use them in the main CPU socket. For example, a system can have two Torrenza-enabled CPU sockets, with one containing a real CPU while another socket containing a specialized application specific processor.

Several manufacturers including Dell, Cray, Fujitsu Siemens, IBM, Sun and HP have pledged their support for Torrenza, saying that AMD is doing just what the industry needs. In fact, IBM's chief technology and company fellow said "as a leader in the open movement, IBM applauds AMD for taking this step and always welcomes partners that take an open and collaborative approach to innovation."

DailyTech previously covered AMD's Torrenza technology, which the company talked about in June of this year. AMD mentioned during a previous conference that Torrenza would allow manufacturers to develop products for the x86 platform easier, without having to create proprietary platforms. The platform will also allow products that currently exist on the PCI Express platform to move over to HyperTransport and AMD's Direct Connect Architecture. According to AMD:

Through the Torrenza Initiative, the AMD64 computing platform is opened for industry-wide innovation, such as connecting non-AMD accelerators to AMD64 systems via HyperTransport technology links. Torrenza supports a range of integration innovations from interconnections leveraging HyperTransport, to co-processors accessing HyperTransport, to plug-in co-processors that directly harness the speed and communications delivered by HyperTransport.

Many companies appear to be pleased with the direction that AMD is taking. The company has been doing extremely well recently, gaining large design wins as well as creating important partnerships. Both AMD and Dell announced earlier this month that Dell has already begun making AMD-based desktops available.

Comments     Threshold

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

Future Motherboards?
By fk49 on 9/21/2006 2:58:02 PM , Rating: 2
I can imagine that a decade or so of this technology could lead to motherboards that are nothing more than I/O, BIOS, and 10 or so sockets.

In those sockets, you could put anything from CPU and GPU to a sound card or memory.

Wouldn't that be cool?

RE: Future Motherboards?
By Kuroyama on 9/21/2006 4:09:53 PM , Rating: 2
I'm guessing that on the GPU side this would only be used at the lower to mid end, because GDDR memory is faster than regular computer memory. Whether the socket is worth its cost at this end of the price spectrum is not so clear.

RE: Future Motherboards?
By Spivonious on 9/21/2006 4:52:03 PM , Rating: 2
But video cards, soundcards, etc. are far more than chips. What about memory requirements, external connectors, multi-chip solutions?

Also, who but the enthusiast is going to swap expansion chips? Most people I know are uncomfortable plugging in a PCI card.

It's a neat idea but I don't think it's going to catch on.

RE: Future Motherboards?
By jtesoro on 9/22/2006 10:05:41 AM , Rating: 2
While it seems that there is some benefit of having a second socket for CPU-like devices, I think that PCI/PCI-E slots already allow for general purpose hardware expansion. I'm not too sure myself if this technology will catch on.

Not good, AMD...
By Jeff7181 on 9/21/2006 10:58:02 PM , Rating: 2
In an era where people value low cost components, which are provided due to interchangeable parts, a proprietary (for all intents and purposes, it's proprietary) design such as this is nearly doomed from the start. The ONLY place I expect to see this is in high end workstations and servers, like an Opteron box from Sun.
AMD should take a lesson from IBM's failures due to proprietary designs back in the 90's.

RE: Not good, AMD...
By Heron Kusanagi on 9/22/2006 12:00:32 AM , Rating: 2
I do not think that this design is at all proprietary. It is open. Please do make the distinction clear.

I will think that such a platform can do good for all types of system. Sockets for physics processors, general processors or even low end video processors, PCI-e and Direct Connect for high end video cards, dedicated sound cards and other add ons.

A Dirty Little Secret to Consider
By Dfere on 9/22/2006 11:22:18 AM , Rating: 2
Is that a lot of companies make a lot of money nowadays on patents. Regardless of the discussion above, even if AMD makes this a truly open standard and it becomes such, my guess is that the money with an array of already designed patents based on hypertransport protocol is where AMD is betting the future.

This is what IBM has primarily evolved into (from a manufacturer to a patent("IP Firm") and consulting services firm). My other guess is that ATI acquisition plays into this somehow (Hypertransport based or improved HypT graphics processing and devices for such to license to manufacturers or other tech companies.) I wouldn't be surprised with Torrenza if AMD gets out of the manufacturing end of things in the long run, and instead designs them.

I think divesting of manufacturing assets is a bad focus for extreme long term strategy for a company, but most manyfacturers have reaped the rewards of outsourcing manufacturing to say, China currently. This means better profitability now, and better board salaries, now.

This makes a firm more flexible as wage standard variations get exploited (Mexico, China, Thailand, India, then most likely Africa), but ultimately these firms risk losing a competitive edge in basic manufacturing, which can detract from future ability to create patents.

RE: A Dirty Little Secret to Consider
By mino on 9/22/2006 3:40:23 PM , Rating: 2
Workers cost of cutting-edge FABs i completely unimportant.
Especially compared to an experience of skilled worker teams.
The amount of money better skilled workforce can save you in this industry usually far outweights their salaries.

Therefore I see no reason ousource CPU production.

And this stands almost forever - the farther in the future, the less the difference the price of a worker in China and EU will be.

In all regards
By Regs on 9/22/2006 8:51:44 AM , Rating: 2
This is a long-term strategy of course. AMD's main concern is FAB production as it seems to be holding them back with a lot of their innovative ideas. Like their next CPU socket and design. After all, they did hire a few well known guys who have extensive experience in FAB production so at least we know AMD is just not sitting back.

Just sell...
By JeffDM on 9/22/2006 1:37:56 PM , Rating: 2
An FPGA chip and a development kit for the chip. Seriously. If the thing can be reprogrammed in-socket then it might be a boon to hobby hardware developers and encourage more group-development in the hobbyist level, possibly reducing the barriers to entry. The cost of making PCI-eboards on a custom level is sometimes prohibitive to hobbyists, if it's an off the shelf chip that goes into an existing socket, that saves money.

Otherwise, I'm not certain what this can do. I need to see more information about the platform and the currently planned commercial products to see if it's something I should care about.

Just another proprietary bus
By DallasTexas on 9/21/06, Rating: -1
RE: Just another proprietary bus
By OrSin on 9/21/2006 12:39:41 PM , Rating: 2
True but how else can it be done. AMD is not charge lisencing for it. Free and to thier specs is the best we can hope for. The way it set up IBM could build cell processor that could work in the systems. I can see even GPU being used here (ATI merger might actually do something). Image in 20 years thier is no PS5 console but a ps5 chip you plug into main board has it boot its software and play console games. That could be sweet.

RE: Just another proprietary bus
By DallasTexas on 9/21/06, Rating: 0
RE: Just another proprietary bus
By othercents on 9/21/2006 12:59:44 PM , Rating: 2
The problem is IEEE is slow to respond. Plus I don't believe that IEEE has ever made a processor socket standard. This was originally driven by Intel and then when Intel made their socket proprietary AMD had to create their own. Now I think AMD is taking the right approach to opening it back up to the competition.

I would love to see the day that we can take a motherboard and put any processor and co-processor into it. Then 2 years later just upgrade the processor and co-processor without having to buy a new motherboard. I think overall we can get to a point where motherboards don't have to be replaced everytime there is a new processor.


RE: Just another proprietary bus
By Araemo on 9/22/2006 9:17:43 AM , Rating: 2
What is so good about the IEEE specifically? Why not let this come from ISO? Or any one of another set of standards bodies?

Just because it is tied to a for-profit corporation(which I believe the IEEE is, actually) doesn't make it bad.

AMD's hypertransport consortium is a nice setup. It isn't quite as open as the IEEE, since noone can kick AMD or the other founding members out, and it is slightly 'viral' like the GPL (Any member has to disclose and license all hypertransport-related patents to the entire consortium)... Hypertransport v. 1.0 was somewhat closed, AMD and a couple other companies did it in-house. The current version has been collaboratively developed by an open standards body open to anyone. It just has some restrictions about joining that Intel wouldn't like. ;)

Torrenza is just an extension of this, I believe. They may open it slightly further by allowing license-free usage by non-consortium members.. but that is just speculation on my part.

"Open" as in open source software(As in, freely modifiable) is not necessarily a good thing for hardware - you don't want anyone to be able to go in and change the spec(But keep the name)... If hardware isn't interoperable, what good is a spec?

Open as in specs and patents visible and given freely.. is always good for development. I don't think Torrenza/Hypertransport counts as proprietary anymore, since it is not individually privately owned or developed.

RE: Just another proprietary bus
By FightingChance on 9/21/2006 12:56:10 PM , Rating: 2
For Intel to use this would be admission that they are following instead of leading, and this is unacceptable in corporate terms. Just like the integrated memory controller on Athlon64's or the Intel BTX standard; both great things but neither company can readily submit themselves to using their direct competitor's ideas without losing face.

RE: Just another proprietary bus
By MrPickins on 9/21/2006 1:15:38 PM , Rating: 2
What about x64 and EM64T?

RE: Just another proprietary bus
By dgingeri on 9/21/2006 1:22:55 PM , Rating: 2
What about x64 and EM64T?

Intel had their hand forced on this one when the customers said they wanted it. They never would have if it weren't for that, just like cooler, lower clock speed, more efficient chips.

RE: Just another proprietary bus
By Drexial on 9/21/2006 2:32:45 PM , Rating: 2
that was a natural evolution. you cant say efficiency and advancing standards are following in footsteps.

RE: Just another proprietary bus
By dgingeri on 9/21/2006 1:21:09 PM , Rating: 2
For Intel to use this would be admission that they are following instead of leading, and this is unacceptable in corporate terms. Just like the integrated memory controller on Athlon64's or the Intel BTX standard; both great things but neither company can readily submit themselves to using their direct competitor's ideas without losing face.

AMD doesn't use BTX because it is far too difficult to arrange due to the integrated memory controller and the setup of the chipsets. I saw an article on that on Anandtech about a year ago. Personally, I don't like BTX for one more reason, the arrangement of the expansion cards and the necessity of the riser cards make it far too difficult to make a decent case with the expansion I would need. I have yet to see a BTX setup that can handle the 4 hard drives, 2 optical drives (DVD+/-RW and CD-RW/DVD), 2 video cards (sli) with doublewide coolers, the TV card, and the sound card I have. As soon as I see that, I may have to reconsider, but until then, I consider BTX a bad idea.

RE: Just another proprietary bus
By Araemo on 9/22/2006 9:23:14 AM , Rating: 2
BTX doesn't 'require' riser cards at all.

A full size BTX case is very similar to a full size ATX case(Just with the motherboard on the opposite side, and a lot of components on the motherboard in mandated positions and orientations).
It is only in the mini- and micro- BTX sizes that riser cards are used.

RE: Just another proprietary bus
By Viditor on 9/22/2006 6:04:30 PM , Rating: 2
It's not the riser card but the distance of the traces required for BTX. Because the memory is placed so far from the CPU, the traces must be longer which is a problem with an on-die mem controller. In addition, the angle is different, and the traces for memory (from an on-die controller) must be exactly the same length...

RE: Just another proprietary bus
By finalfan on 9/21/2006 1:27:58 PM , Rating: 1
Still there is people thinking on die controller is AMD's invention. Intel used it on Timna back in 2000 along with a on-die graphics controller. The on die memeory controller is the direct reason Timna project was cancelled since it uses RDRAM. AMD followed and was successful just becuase Intel has matured the DDR and DDR2 memory market so AMD can be sure there is no risk to use that memory technology.

RE: Just another proprietary bus
By MonkeyPaw on 9/21/2006 4:20:17 PM , Rating: 5
The problem with Tinma is that it never made it out the door. I did some digging, and I believe Intel planned an IMC all the way back in the 486 days. Also, Intel wanted RDRAM to be the future, but AMD and some chipset companies (SiS) showed that DDR was a better product. Rambus came under fire, and Intel adopted DDR. It's true that AMD didn't invent the IMC, but they made a successful one. As great as it is to innovate, execution is a bit more critical, isn't it?

RE: Just another proprietary bus
By formulav8 on 9/21/2006 4:20:55 PM , Rating: 2
Intel never released the cpu, and noone really knows how far along they made it either, and thus, they were not the first. AMD was if you compare it in x86 terms.

AMD was also first to use DDR, and it was them that so-called 'Matured' DDR, whatever that really means.

So, then, by your logic, it would be Intel is the follower.


RE: Just another proprietary bus
By finalfan on 9/21/2006 6:09:10 PM , Rating: 1
My point is simple:

1. Intel had the technology first
2. Intel failed using IMC becuase it was and still is the leader and explorer of new memory technology
3. Intel's adoption of DDR matured that market and lowered the price dramatically so AMD can follow with an IMC safely
4. Without Intel's adoption of DDR and DDR2, AMD's IMC would have had the same fate as Timna

RE: Just another proprietary bus
By PrinceGaz on 9/21/2006 10:38:17 PM , Rating: 2
You are totally incorrect when you say Intel adopted DDR first.

AthlonXP based systems were being built around chipsets supporting DDR memory long before Intel made the move from RDRAM to DDR at the high-end (and Intel was still using SDRAM for mid-range and budget machines).

VIA KT266A based mobos (Athlon/AthlonXP with DDR memory) were selling like hot-cakes when Intel's best offerings were i845 (SDR) and i850 (RDRAM). It was third-party AMD chipset manufacturers and not Intel who brought DDR to the mainstream.

RE: Just another proprietary bus
By mino on 9/22/2006 3:25:31 PM , Rating: 2
"AMD followed and was successful just becuase Intel has matured the DDR and DDR2 memory market"

Are you joking us or what ???

Try remember who was so heavily promoting DDR back in its infant stage...

Also AMD could have run DDR even to 2008, they simply DO NOT need DDR2. Only reasom AMD has gone DDR2 is market wanted, nothing else.

RE: Just another proprietary bus
By Tsuwamono on 9/21/2006 5:23:47 PM , Rating: 1
AMD didnt use BTX because it doesnt have a heat problem like intel. Mr prescott running at 70 degrees thats slower then my San Diego at 29 degrees both on stock cooling

RE: Just another proprietary bus
By hstewarth on 9/21/06, Rating: 0
RE: Just another proprietary bus
By Targon on 9/21/2006 3:43:42 PM , Rating: 2
You need to understand that a GPU can talk to the rest of the computer in different ways. Sure there might be a series of GPUs that fit into the current socket as a co-processor, but the same GPU could also be made as a part of a video card that works with PCI Express or whatever comes next when it comes to expansion slot technology.

The AMD/ATI merger is one possible way for AMD to make their platform better than what you might be able to get on an Intel based platform. Since Intel has their own chipset and GPU divisions, I don't see how you think that AMD is trying to set themselves up as some sort of monopoly. Intel could hop on board and make their processors socket compatable with AMD processors if they really wanted to.

AMD has opened up a number of technologies to be used by others, and it's only the idiots at Intel who are at fault for not being willing to use them.

RE: Just another proprietary bus
By Samus on 9/21/2006 8:29:47 PM , Rating: 2
it's not proprietary because its open. they're completely opposite things.

what didn't you understand about the term 'open'

it also isn't a bus, it's based on hypertransport, which is also 'open' and free. even intel uses hypertransport, an amd 'invention' (although they joint developed it with many partners, much like this socket platform will be)

RE: Just another proprietary bus
By dgingeri on 9/21/2006 12:51:58 PM , Rating: 5
by definition, an 'open standard' is one that anyone can use, just like HyperTransport and AMD64 instruction sets, anyone can use it and anyone can design around it. AMD is simply submitting a way that these things can be done.

'Proprietary'is a closed method that can only be used by the initial designers and/or someone that licenses the technology from the original designers, like Intel's interface for their processors and the interconnects to their gigabit NIC's in the 875 chipset. Intel is all about Proprietary, not any sort of open standards.

AMD calls this 'open' because they are allowing anyone to use it, including Intel, without cost. Intel only allows others to use suck technology with a very costly license. Sure, Intel can attach a Hyperlink interface to their chipset and put a Torrenza socket on their motherboards without any trouble from AMD, I just doubt they'll do it until the technology starts becoming more common. They wouldn't even make an AMD64 compatible chip until the Opteron actually showed that it would be more powerful than anything they had at the time.

I'd love to see this become widely available. Dedicated processors for physics and 3d rendering beyond the video card would help in many industries, and maybe even work better for games.

I'm afraid you're getting mixed up as to what 'open' means.

By rushfan2006 on 9/21/2006 1:13:18 PM , Rating: 2
I'm very glad you made that point about what open means. Its one of the most commonly mis-used terms in the technology/IT realm that I see on a consistent basis. A large portion of the populace have this thought that "open standard" means that only something that a community of people who freely invested their time and graciously contributed their talents to produce. And as such a single source being a company, even a for profit company or a single individual who even originally sold whatever it was for profit but changed his mind and now allows open access for all doesn't mean "open".

This is one major basis why I don't get dragged into most "open source" -v- "non/proprietary source" debates anymore.

Open source is an idea, a concept of community sharing with no single owner claiming rights to said project. Its a simple concept really, thus I don't understand why some get it confused.

Anyway..good post.

RE: Just another proprietary bus
By Spoelie on 9/21/2006 12:52:30 PM , Rating: 2
HT is open, and fronted by the Hypertransport Consortium . Yes originally during development (when it was still called LDT, lightning data transport) it was an AMD technology but that's not the case anymore. ccHT (cache coherent) was closed and AMD proprietary, but I thought they threw it or were planning to throw it out in the open.

RE: Just another proprietary bus
By dgingeri on 9/21/2006 1:15:36 PM , Rating: 3
Oh, I forgot to mention, relying on IEEE to come up with standards is a very bad idea. They have not done well with wireless networking protocols (leaving a lot of holes and taking forever to actually get a standard defined) and TCP/IP (with far too many holes in DNS standards) to be taken seriously anymore. They have become more of a political organization than a technical one in the past decade. That is why PCI and HyperTransport are governed by dedicated groups, rather than the IEEE.

RE: Just another proprietary bus
By ghackmann on 9/21/2006 5:42:17 PM , Rating: 2
They have not done well with [...] TCP/IP (with far too many holes in DNS standards)

TCP and IP are not IEEE standards. DNS is not an IEEE standard either. And DNS isn't even part of TCP or IP! (DNS sits of top of TCP and/or UDP, but is not required for either to operate.)

RE: Just another proprietary bus
By MonkeyPaw on 9/21/2006 4:43:22 PM , Rating: 3
It's pretty harsh to call this proprietary, since it's based on HT--something that has been developed jointly by the HT consortium. Of course AMD is going to develop systems built around their products, it's why they are in business! Right now, most companies only look to AMD and Intel for a vast majority of their CPU needs, so I only see this as adding choice where there previously wasn't one. If nothing else, AMD is trading away a socket where you might normally put another Opteron. Torrenza technology lets companies put whatever they want in the other socket, which makes it quite a bit different than proprietary. AMD still has to develop the connection, since they have to develop platforms for their CPUs anyway.

So if you are seeing this as AMD's attempt to offer a proprietary socket, then you are missing the point. What this technology will do is make the CPU secondary in purpose, and the "torrenza CPU" to become the application-specific workhorse. From the sound of things, it will allow companies to develop a single purpose product without having to deal with the complexities that AMD and Intel have to deal with, especially PCI/PCIe busses. Physics processors and GPUs already stand to benefit greatly from this technology, as these products like to work at the clock cycle level.

Intel--or any other company for that matter--is more than welcome to join the HT consortium and adopt the same technology. AMD is simply leading the way. This does allow AMD the ability to control the direction of development, but it appears they are asking what bigger companies want. This is why AMD is succeeding in many places, as they are trying to give people what they want!

RE: Just another proprietary bus
By Viditor on 9/22/2006 6:07:44 PM , Rating: 2
It's actually not proprietary...AMD have opened up the license to all manufactureres, and it's no longer controlled by them but by the HT Consortium.
BTW, Apple still own the license to the iPod connector...AMD have given away their license to the consortium.

"Folks that want porn can buy an Android phone." -- Steve Jobs
Related Articles

Copyright 2016 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki