backtop

Print E-mail del.icio.us 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 Darkskypoet on 3/18/2007 7:54:03 AM , Rating: 5
...if you've been watching AMD over the past year or so, the potential of Fusion and Torrenza has been quite clear for a while.
(I posted this a couple days back, in a different forum... got blasted for it to... sigh. Perhaps not all applicable to Fusion / Torrenza, most is.)

****
I truly applaud Intel's 'Athloning', the Athlon with c2d. However, it's really not that revolutionary of a chip... You make a P3 with a far smarter / efficient decode engine, and a fatter / wider set of RISC pipes, and voila a fast short pipelined chip. Kudos, and see Barcelona will be the same, with some other tricks up its sleeve.

These chips are made possible thru very very sweet production techniques. All Intel's core dev teams have done is incremental optimizations of a short piped RISC chip... Kudos, but it is manufacturing prowess and money that are their greatest assets. If both companies were still at 90nm, AMD would be far more competitive with Intel even with a c2d chip. You'd see relatively similar clock ceilings, and not the runaway over clocking Intel has granted the many of us (myself included) with any c2d. In fact I would say AMD has a far better 90nm process then what Intel had. You have to when you can't just go build a Fab out of petty cash.

C2D Wins, but don't think its because the fundamentals behind the chip are truly special, its simply a lot of money, and a lot of medium delta re-spins. Many different cores came and went as mobile processors to come out with the c2d. This time they managed to fix the last of the bottlenecks, and make a truly balanced and properly functioning chip. Remember Pentium M? Honestly a far better processor then anything Netburst gave us, however its achilles heel: High Bandwidth Media transformation / encoding. This wasn't truly fixed until the last iteration of this line: the C2D.

Next;
Fusion: The CPU world as we know it is heading for change. AMD with the acquisition of ATI in their fusion concept can drop 2 R600 cores, and 2 Barcelona+ cores onto 1 die, and drop two of these into a much maligned Quad FX mainboard (yes, or pretty much any Dual or Quad socket Opteron mainboard as well).
Wonder why 45nm and fusion is so special for them?

Anyone see the terraflop in a box performance? Realize then we are talking roughly double that performance plus 4 cores of Barcelona+ goodness. Bring your 8 core C2Ds to the table, we'll race for pink slips.

Fusion isn't about graphics... It might be in a desktop consumer market targeting OEMs or others :). But as has been pointed out ad nauseum, it's about servers. With Servers, it's about stream processing, and specialized processors / co-processors. There is a reason AMD opened up Hyper Transport technology; more chips made for their platform, more options, more specialized blades, and clusters / nodes.

The future is not simply who has the fastest General Purpose CPU, the future is in what platform can I leverage to be most efficient with my workload? The one for which I can only drop in Intel C2D derivatives? Or the one that I can drop some stream processors, cell derivatives, encryption copros, etc. into?

I for one can't wait to see the latest server benchmarks with AMD's “Close to Metal” SDK, and a few R600 boards, or soon multicore R600 Stream processing boards. Can a 6ghz quad core C2D match that floating point monster? Anyone who remembers IO cards can attest to the evolution of boards to parts of a piece of silicone... Can't wait to get my dual quad core fusion chips. 4 Barcelona+ cores, and 4 R600+ cores in two sockets. All those folding PS3's better watch out.

Fusion is only about graphics for consoles, set top PC's, etc. Its really about creating heterogeneous cores with which to customize performance. It's not necessary for business PCs. Business PCs will have full system on a chip designs. OEMs will love this. AMD is already leading there as well CPU/NB on a chip.. add video... and come on, like an SB is hard to mold into the equation. Give me a break.
(Note: Intel created, then abandoned a system on chip years ago... but not quite the same beast we speak of here, [Timona or some such?])

All of which is so much easier with the purchase of ATI!

Now onto the difficult short term reality. AMD needs cash. Yes Fanboys, and Girls... Send your cheques to Dr. Ruiz care of... You get the picture. The delays, I bet you anything, are due to cash issues. FABs are expensive! Slow 65nm rollout; lack of cash... delayed R600; cash, and possibly frictional adjustment issues with integrating two firms of that size. AMD needs money now more then ever. However, I think Intel is praying no one buys them as an investment opportunity. Give AMD a solid financial footing, and a hands off investment group... and Intel will continue to lose market share. Furthermore, AMD will be more dangerous to them then ever.

The other side of the coin:

I notice everyone jumps on the new $266 QuadC2 promised to be at a store near you in the near future. Does anyone else but me realize that Netburst chips still make up a terrible amount of Shipped Intel CPUs? As bad as AMD's ASP is, Intel hurts there own inventory value with ever price gouge. This is why AMD is not doing as badly as it could be. I also would hazard that the coming of cheap C2Qs coincides with enough of Intel's CPU sales being Core based that completely torching the old inventory won't wound them as badly as it would today. However, with Allendale, and cheap cheap cheap C2Ds they are hurting themselves, as well as AMD.

C2Ds aggressive pricing is one of the main reasons their revenues are down. Last I checked nearly 50% of Intel's shipments were not C2Ds... So 50% is getting given away so Intel can fight a price war with AMD. Deep pockets? Yes. Happy shareholders? Maybe a stretch. AMD needs to stay in the middle of the sandwich, which they are doing, until Barcelona hits server side at the least. Then ASPs can marginally increase for AMD, whilst they fight intel on the low and mid end where OEMs buy millions of chips.

AMD still sells every chip it makes. Period.

End result: AMD will retain, if not gain market share, until Intel depletes its massive inventories of Non-C2D chips. Then the pain comes for AMD, unless Barcelona can be fast enough in certain high markup segments to boost ASPs while AMD matches Intel in the mid to low end cheap 65nm X2's made by Chartered, vs cheap 65nm C2D (and derivatives) made by Intel.

45nm..... Is the next massive advantage for Intel, but again... CPUs are not sold anywhere near cost, so undercutting the sales of 65nm C2Ds in price by trying to leverage a slight cost decrease from 45nm (per die)Penryn, will just serve to hurt their ASP even more. So 45nm will more then likely be high end. This means we have to ask hat will AMD have high end at that point? Can they leverage it to raise ASP, while still competing in the low to midrange price war?

In reality, the game is changing, and AMD has been one to move before Intel on game changing solutions. Raw manufacturing capacity has always been Intel's strong suit. C2D is not a deviation of this, C2D is a company spending billions on a design team to keep working till they got it right. It took many P-M iterations to get to C2D. Arguably something that wasn't to far out of their reach when the first P4 rolled out to the first stupid buyer who honestly thought 1300mhz P4 > 1000mhz P3...

AMD brought SIMD instructions to FP processes before Intel.
AMD Licensed Alpha EV-6 bus topology, better then Intel.
AMD brought On die MC before Intel
AMD produces and sells Stream Processing Cards, w/SDK (faster then Intel's General purpose CPUs)
AMD Went the C2D route initially (Short and wide, while Intel went Marketing with Net Burst.
AMD Adopted a more advanced Point to Point Bus topology with HT. Intel still hasn't.. (Want to count the years between Athlon MP and when Intel Finally fixes its multi cpu platform?)

Night / or Morning to you all. :)




By BLOfelt22 on 3/18/2007 9:31:52 AM , Rating: 2
Very well thought out and equally well worded post. I can't imagine why anyone would flame it but I guess everyone is entitled their opinion. Only time will tell and I do believe that AMD has been in worse situations before. We'll just have to see.


By just4U on 3/18/2007 2:36:48 PM , Rating: 2
I have a question. Amd has been able to license Intel instructions and such over the years right? IF HT is a open standard can't Intel just use that eventually? Why complicate things further by creating your own thing when there is already a standard in place?

Just curious here!


Moderated
By Darkskypoet on 3/18/07, Rating: 0
By Darkskypoet on 3/18/2007 7:27:22 PM , Rating: 5
No its not.

Seriously, Intel has a lot of its 'mojo' from its design / implementation prowess. Taking tech from an 'inferior' firm after downplaying pretty much every tech brought forward from said 'inferior' firm hurts that image.

However not having any response to the currently available, and shipping HT tech (allowing for other firms to create chips to drop onto Opteron boards) leaves Intel out of a new and growing market. Thats not fanaticism!

I am saying the question is which does more damage: Waiting for another year plus to have a standard to bring to market, thus letting HT gain mind share and development dollars. Or, simply adopting the HT spec, thus giving AMD more legitimacy.

Neither is desirable vs Intel having had a competing standard 2 years ago.

(I am starting to become amused by your consistent down rating of decent posts. You really ought to stop attaching fanboyism / fanaticism to any post you for some reason do not seem to like)

Intel has lost ground, this is not speculation. Or are you unaware that AMD server market share has been growing? Especially in terms of 4 and >4 socket implementations.

Are you also unaware at the industry support now behind AMD HT, something that even 4 years ago would have been almost impossible due to AMD's lack of Server penetration. Sigh.

zsdersw please do not read any Intel is dying, etc. crap into my posts. Relative change in position between the two companies has been favoring AMD.

This necessarily shows Intel as losing some of the immense power it had relative to AMD in the past. HT, is part of this. Or are you saying it is not?


By zsdersw on 3/18/2007 7:40:24 PM , Rating: 1
First, I cannot rate your posts down because I've participated in this story's comment section.

Second, I am not addressing the question of image. I was addressing your apparent belief that Intel will necessarily lose market share whether they use HT or their own technology; that there is no hope for Intel to either increase market share or hold what they have.

Third, don't raise issues that I'm not arguing with you about.


By Darkskypoet on 3/18/2007 10:30:16 PM , Rating: 2
Well then, as I am new here, I apologize. As I am not a neophyte, I also realize the existence of multiple accounts per IP is allowed, so I am not going to pursue that any further.

Secondly, I am not saying will, or won't. I am saying have, and is still happening. Nor am I saying that they cannot take back market share, or that there is no hope. Please, quote the no hope wording from the post. I am saying that they are losing, and are behind in this venture because they have nothing comparable out.

I am saying quite clearly that to adopt HT after maligning AMD tech so much is bad for image, and secondly that coming to market so late hurts the initial uptake of this technology by other CoPU vendors.

Either way, adopting late, or coming to market late hurts market share. This should be elementary. I fail to see how "no hope" is read from that statement.

There is a reason AMD is losing enthusiast market share, because Barcelona is late. If I, aparently an AMD fanatic, am building C2Ds for myself, and clients, (left, right and center) obviously in the market in which I dwell, C2D is the leading chip. However, if AMD had Barcelona right now, it would be a much more difficult choice.

Since it isn't; we can see what happens when you come to market late with a competitive technology. You lose market share. Thankfully in this case for AMD, it is much simpler to create a CPU that doesn't rely on massive industry support for success. Whereas a new platform (like an HT competitor) requires such support, and as such needs more lead time to again become competitive, and to be developed for.

(For your benefit, more lead time, and more time in general does not equate to "no hope". Just wanted to clarify that)


By Viditor on 3/19/2007 9:17:07 AM , Rating: 2
quote:
When your posts are absent of any balancing good news on the Intel front, the tone of your posts quickly becomes one of anti-Intel, pro-AMD

Or it a post that is correct...often times there IS no balanced good news from Intel or AMD (that we know of...).


By zsdersw on 3/19/2007 10:13:47 AM , Rating: 1
There's always a way to balance it out.


By gonchuki on 3/19/2007 12:30:22 PM , Rating: 3
may be it's me, but i see it balanced taking into account that he states and remarks AMD's fault for being late in the market with Barcelona.
You are just trying to obscure things and point your own fanatiscism with those comments.


By zsdersw on 3/19/2007 2:39:08 PM , Rating: 1
That is the only example of balance.


By Yawgm0th on 3/18/2007 3:39:49 PM , Rating: 2
Using a technology developed by your main competitor isn't a great PR move, especially with fanboys waiting to jump on a little piece of information like that.


By Darkskypoet on 3/18/2007 7:54:02 PM , Rating: 2
Exactly... However, my post pointing such out was moderated down... Sigh. So I replied with gusto :)

I always thought the posts that were moderated down, were full of AMD is L33t & Intel Sux0rs... but it seems otherwise intelligent posts share that fate as well on this board. Sigh.


By zsdersw on 3/18/2007 8:02:30 PM , Rating: 1
A fancily-worded version of fanaticism is just as likely to get voted down just as much as a less eloquent version.


By Darkskypoet on 3/18/2007 9:16:45 PM , Rating: 2
So somehow pointing out the surprisingly obvious fact that AMD has been / is taking server market share from Intel, and part of that is due to HT and the working consortium behind it, is fanaticism? That is rather funny. I think your down moderating of intelligent factual posts is more akin to fanaticism, then my noting fairly obvious trends.

You disagree then that Intel adopting HT, late, will do no harm to their 'superior' mind share? Also, that by being a late adopter they would then have more ground to cover development wise in a shorter time frame?

Or that the delay in adopting a new point to point, somewhat open standard for interconnect has done no harm to their higher end server platforms?

Really? Let us know what you think about this. I mean it is one thing to call out 'fanaticism!!', and yet another to post an alternative for the readers to evaluate.