backtop


Print 44 comment(s) - last by Spivonious.. on Oct 2 at 1:31 PM

Intel will open up its server platform

This week at IDF, Intel made an official announcement on its response to AMD's Torrenza technology. AMD made waves earlier this year when it announced that it would open up its Opteron platform to the industry, allowing other manufacturers to create and develop add-in components that communicate directly with the system processor and memory. Going beyond that, AMD also mentioned that Torrenza would allow companies to create accelerators or co-processors that could be used directly in an Opteron socket.

Intel said that like AMD, it also plans to open up its chipset platform technology. The move would be an unprecedented move for Intel, as it has been guarding its platform for the longest time. Intel's primary goal is to introduce an alternative to AMD's HyperTransport. The technology would allow devices to communicate on a much faster pathway than PCI Express alone could muster. Interfacing directly with the front-side bus (FSB), devices will be able to communicate directly to the processor and or other accelerators. Non-Intel chips will be able to plug into a Xeon socket for example, and work parallel to the main processor or processors.

With the introduction of an open FSB platform, Intel will also be making a move towards integrating memory controllers directly onto processors. This is something that AMD has been doing for several years with the original Opteron processor. DailyTech previously reported that a number of large companies were already partnering with AMD to create accelerator and other co-processors. The decision to open up its platform has propelled AMD into the enterprise market in very large way. It will be interesting to see what Intel's move into an open space will do for the industry.

Currently, the technology is expected to be introduced sometime in the next one to one and a half years. Some analysts speculate that Intel will show off an open FSB specification in 2008 on Itanium, and on the Xeon sometime in 2009. Reports say that Intel is currently working with several companies to create co-processors -- they too would be able to plug directly into a Xeon or Itanium socket.

This week, Intel also announced several updates to its product family. The first being that Kentsfield will now be called Core 2 Quad, which it promised to ship one million units before AMD could ship a single one. Other interesting developments from Intel include two new 45nm fab locations as well as a strong push into tera-flop computing research.


Comments     Threshold


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

fsb
By Lazarus Dark on 9/27/2006 10:33:44 PM , Rating: 2
I don't think this will work until intel moves away from its current fsb system. Or am I wrong? would this be able to compete with hypertransport or be a poor substitute?




RE: fsb
By HopJokey on 9/27/2006 10:46:14 PM , Rating: 2
This could pave the way for when Intel introduces it's point to point bus (CSI) with it's next generation x86 product, code-named Nehalem (CSI is also planned on it's next generation Itanium product, code-named Tukwila). First Nehalem based processor should ship sometime in 2008 if all goes planned.


RE: fsb
By Jharne on 9/28/2006 6:54:57 AM , Rating: 2
CSI has been a moving target for some time, Intel need to deliver it this time.


RE: fsb
By Motley on 9/28/2006 12:53:19 PM , Rating: 2
No, they need to keep waiting and improving it, and only release it 6 months before it's really needed. I don't really want a whole new architecture for the measly 2%-4% speed bump it'd give today. I'd rather they spend their efforts like they have, concentrating on the 20-30% speed bumps without a massive architecture change.

Granted, eventually the large speed bumps without an artitecture change will become more and more difficult, to the point where the greatest gains could be gotten from an architecture change, but that time isn't today.


RE: fsb
By Justin Case on 9/28/2006 4:48:26 PM , Rating: 2
Performance improvements are always needed. 10% faster systems can mean 30% higher profits, in some businesses. There's no such thing as "6 months before it's really needed". It's "needed" yesterday.

I fail to see how frequent updates are a bad thing. Between a company that releases a 30% faster product after 3 years, and another that releases a 9% faster product every year, I'll take the latter. It might be 0.5% slower in the last year, but it kept me ahead of the game in the previous two.

The problem with Intel is they keep changing sockets, changing chipsets, changing memory types, etc.. And that's what AMD finally figured out, with their "stable platform" initiative: if they commit to a platform for 5 years, that gets them support from the enterprise segment. Most companies (apart from a handful of IT mega corporations) couldn't really care less about Torrenza. But every company likes to know that the servers they buy today will be upgradable in 3 or 4 years, and won't have to be completely replaced because mighty Intel decided that everyone should jump onto a new, untested, more expensive, and sometimes slower platform.

That is why AMD is growing in the server space, and is likely to keep on growing (unless they seriously screw up). Intel is going to need at least another 16 months to catch up in terms of interconnect technology (which is the major issue in the enterprise server space), and that's assuming its management doesn't shoot itself in the foot, as it did so often in the recent past. Netburst is what happens when you let the marketing department run a technology company.


RE: fsb
By Dactyl on 9/30/2006 5:50:03 PM , Rating: 3
You only get a big benefit by changing your architecture if your old architecture sucked (e.g., Netburst).

AMD can't just pull a rabbit out of its hat and get a 40% performance increase/40% power decrease by changing its architecture, because its architecture is pretty good.

It will have to get the increases the hard way (smaller processes, moving to quad core, K8L extensions, on-die L3 cache). This is very similar to how Intel improved the performance of its P4 chips before it finally got rid of them (smaller processes enabled higher clockspeeds, dual core with Pentium D, added HT, boosted cache sizes). (and I mean no disrespect to K8 by comparing it to P4!)

That said, as we move into the future, with smaller and smaller processes and better interconnect, radical new design approaches (e.g., mini cores) might start to make sense. I doubt mini-cores would have made sense when CPUs were being manufactured at 180 nanometers, like the first P4s.

Evolutionary improvements to those new architectures will not only make them better (faster, more performance per watt) the chipmaker will also learn how to further optimize future processors. If a chipmaker completely turned its back on evolutionary improvements, it could end up making radical new designs with many of the same minor flaws as in old designs.


RE: fsb
By Viditor on 9/28/2006 11:48:22 AM , Rating: 2
quote:
First Nehalem based processor should ship sometime in 2008 if all goes planned


First CSI for Itanium (Tukwilla) should be the end of 2008, first CSI for Xeon (Nehelam) is slated for early 2009. That said,
1. I don't believe they have worked out the bugs yet, so it's still not a lock...
2. The proposed open platform model is for FSB, not P2P...so it wouldn't work on CSI platforms.
3. Your point is quite relevant but for the opposite reason...if the new open platform is adopted, then it won't be ready until CSI is near completion (rendering it useless).

My guess is that Intel is wisely not putting all of their eggs in one basket this time. They are adopting this model in case CSI doesn't work...JMHO


RE: fsb
By Phynaz on 9/28/2006 12:39:08 PM , Rating: 2
quote:
2. The proposed open platform model is for FSB, not P2P...so it wouldn't work on CSI platforms.


Nope. The proposed open platform model is based upon PCIe.


RE: fsb
By Viditor on 9/28/2006 7:59:56 PM , Rating: 2
quote:
The proposed open platform model is based upon PCIe

From Anand's article:

"In another example of a string of Intel following AMD's lead, Gelsinger announced extending Intel's FSB license to FPGA manufacturers like Xilinx so that companies can produce other chips that can work alongside Intel processors with a direct FSB connection to the MCH"


RE: fsb
By Phynaz on 9/29/2006 9:21:16 AM , Rating: 2
The FSB isn't open. PCIe is open.

From the same article:
quote:
Intel and IBM got together and proposed an extension to PCI Express that would offer both higher bandwidth and lower latency. Since PCI Express is already an industry standard embraced by all sorts of manufacturers, the evolutionary move to another PCI Express based interface makes a lot of sense and is more likely to gain traction than requiring vendors to produce HTX compatible solutions.


RE: fsb
By Viditor on 9/29/2006 11:47:54 AM , Rating: 2
Notice the difference there...

"Intel and IBM got together and proposed an extension to PCI Express..."

"Gelsinger announced extending Intel's FSB license to FPGA manufacturers"

Intel is hoping to do both...but the PCIe proposal is not going to directly connect to the cache and would have a much higher latency...


RE: fsb
By Phynaz on 9/29/2006 2:25:05 PM , Rating: 2
Note your original post where you stated the proposed open platform is FSB.



By peternelson on 9/28/2006 2:35:32 AM , Rating: 2
So, if I wait until 2008 I can have a spec which will let me develop products to talk to an Itanium! (ITANIC).

Just it won't work with any CURRENT Itaniums or Itanium 2s. Oh and people aren't exactly rushing to buy those either.

So, maybe I can wait until 2009 to get a spec to talk to Xeons.

Hmmmmm, I can CURRENTLY make products to talk to AMD over hypertransport on HTX connector. Or I can do the torrenza socket thing that works NOW. There are at least two chip in socket products for opteron 940 for a while based on xilinx and altera fpgas.

Oh and sometime Intel will catch up with integrated memory controller.

As someone who want to make hardware accelerators this Intel announcement doesn't do it for me. I will stick with the AMD standards based approach rather than Intel vapourware. The AMD architecture for multi socket systems eg 8 sockets interconnected by HT links, 4 cores per processor, and using some sockets as acceleration coprocessors sounds like a good sweet spot to me for HPC apps.





By crystal clear on 9/28/2006 6:32:55 AM , Rating: 2
With FULL RESPECT for your opinions,I would say brushing off
the INTEL initiative is a bad business decision.
Better to keep all your options open.


By peternelson on 9/28/2006 8:32:46 PM , Rating: 2
Thanks.

You're right of course, either AMD, Intel, IBM or others could go splendidly bust and life would go on.

If I want to go with Intel's architecture I have to wait 3 years or longer (to get the spec, design my product, manufacture it).

If I want to go with AMD's architecture I can take existing product and adapt it to an available specification.

Even if I'm ignoring Intel for now, any product I come out with now will be obsolete in 4 years time as the pace of improvements continues. Therefore in 3-4 years time I can look at the market. Is there a large installed base of torrenza capable systems and users? Is it cheaper to build in support for the AMD or Intel standards? Does the market for acceleration on Intel demand my attention.

You are right that we can't ignore Intel. I had no intention of sticking my head in the sand in a fanboyish way. In terms of making products, Intel can only offer us PCI-Express interconnect, which is pretty good, and is cross platform. It's the stuff that really pushes boundaries that demands anything better.


By FITCamaro on 9/28/2006 11:04:15 AM , Rating: 2
Intel could have done an IMC. They choose not to because it would mean requiring a new socket and new platform. Their attitude is just that its better to have a FSB because it allows more flexibility. And they're right. That doesn't mean FSB is better than Hypertransport though or vice versa. Hypertransport is faster and more efficient no doubt. But it does cause limitations on flexibility and upgradeability.

I'm saying this though as an AMD fan who's currently on an Intel system because at the time, it was the faster of the two platforms. And thats what I care about. Performance. I don't care who's name is on it.


By MonkeyPaw on 9/28/2006 12:24:27 PM , Rating: 4
Well, I think Intel likes to sell chipsets, personally. Yes, they could have introduced an IMC and an HT-like protocal a while back, but that would mean no more "chipset upgrades." Think about how many times a new Intel CPU is incompatible with a current chipset, or that the current chipset requires a revision. Intel is designing both the CPU and the NB, so how can they not anticipate these problems? The only other company who has frequently "revised" their chipsets for new CPUs is VIA! No, if Intel introduced the IMC and HT today, there would no longer be a need for so many different chipsets and so many chipset revisions. They probably couldn't charge as much for the "Centrino" platform either. Instead, we get a new CPU which is often incompatible with current chipsets, then we get an "ideal" chipset launched shortly after. This move burns enthusiasts, but it also encorages OEMs to sell their old inventory with old CPUs.

Before you dismiss my comments as fanboy, consider Intel's x86 SMP solutions. The opteron is pretty simplistic and highly scalable thanks to the IMC's ability to feed each socket, and HT's ability to allow fast communication between CPUs (and most other devices, for that matter). Intel, on the other hand, offers a rather elaborate shared-bus to run even a 2-socket server/workstation. While the results are very convincing, it's still the hard way to do things, and it makes the industry dependent on Intel's solution, which is frequently answered as I mentioned above. If Intel is planning to make "their HT" connect to yet another dedicated northbridge, the scalability problems will not go away. I can see Intel using HT to allow NB-to-NB connections to increase scalability, but that comes at the cost of yet another Intel chip on the board, and it adds to complexity and power demands. It's still not the ideal solution, and at least not as good as a direct connection to the CPU itself. Should Intel adopt a similar IMC and HT as AMD already has, they would eliminate many of their current scaling issues immediately. Until they resolve this, it will be AMD's CPUs going into most of 4-socket and up configurations. AMD has put the server industry on its ear because they have given them what they want. Intel is still not heading in the right direction, IMO.


By encryptkeeper on 9/28/2006 5:54:49 PM , Rating: 3
I work for a PC distributor, and my big problem with Intel is their chipsets. Take a look at Intel's site to see what CPUs work with what motherboards. Socket 775 has been around for about a year now, and its STILL almost impossible to tell what board will work with what processor. Like the difference between the 630 and 631 processor. Alot of the lower end boards will *not* work with the 631. Same thing with the D940. When they went to the D945, it wouldn't work with all of the 945 chipsets like the D940 did. We are constantly pulling our hair out over this crap when we try to configure systems for our customers. I don't care about who has the absolute fastest performance because most of the time you either don't need whatever's the fastest. Processing power is moving so quickly, you'd have to be an obsessive jerk to try and get the absolute fastest thing out there, which I'll admit is Core 2 Duo.

But take AMD. They have a good range of chips still available (single core Intel chips are getting to be a bitch to find) and they have a great unified socket. You take *any* AM2 chip and it's going to work on *any* AM2 board. AMD has even said that with on-chip memory controllers AM3 processors will work on AM2 boards. Holy hell! Think about that! A next generation processor that you can use without having to replace your whole machine??? I'll take AM2 any day over Intel. Plus, they're AMD, they'll gain back whatever lead they lost in terms of processing speed.

With that in mind, think about why Dell decided to go with AMD processors. I'm sure that part of the decision was that their kit designers bitched about how hard it was to configure an Intel system compared to an AMD, and the execs just broke down and gave in to their demands.


By Rob94hawk on 9/28/2006 10:02:12 PM , Rating: 1
Well said. Despite Intel's deep pockets and plethora of resources they are either too dumb or too arrogant to simplify things for the consumer.

Intel reminds me of the US government, all these Ivy Leauge educated people running the show yet they are some of the dumbest people on the planet.


By JeffDM on 9/28/2006 11:49:03 AM , Rating: 2
Wait, are you saying that I can buy a 4x4 (or Torrenza) system today?


By Viditor on 9/28/2006 8:10:29 PM , Rating: 2
quote:
Wait, are you saying that I can buy a 4x4 (or Torrenza) system today?


Two different things...Torrenza is AMD's open license for the cHT (coherent HT) connections which allows third parties to connect coprocessers directly to their own system ram and directly access the cache of any CPU.
4x4 is the dual socket for AM2 which also allows for multi-GPU slots.
It is possible that 4x4 will allow Torrenza chips (i.e. one X2 CPU and one physics coprocesser on a single consumer board with 8 Ram slots).


By peternelson on 9/28/2006 8:24:07 PM , Rating: 2
You can buy accelerators using the same technology in opteron 940 sockets already eg from DRC using Xilinx.

Such companies are likely migrating products to AM2 and/or socket F now as well as incorporating latest Xilinx Virtex 5.

AMD often make announcements during Intel's IDF so maybe we will hear more about 4x4 this week.



By hstewarth on 9/28/2006 9:20:14 AM , Rating: 1
Think about - an Open platform on AMD or Intel is not possible now Why because AMD desires the cpus to be architexture in their style and Intel desires in their style.

Eventually the only common thing between could be original instruction set that Intel evented a decade ago.




By FITCamaro on 9/28/2006 11:07:23 AM , Rating: 3
Sorry but please for the love of god learn to spell.

architecture not architexture
invented not evented


By hstewarth on 9/28/2006 2:46:09 PM , Rating: 1
Better yet, this site allow uses to edit there requests. Sorry for any confusion.


By stanwalters on 9/28/2006 9:52:34 PM , Rating: 3
this site allow uses to edit there requests

No, try "this site allows uses to edit their requests"


By JeffDM on 9/29/2006 6:04:04 PM , Rating: 2
No, try "this site allows uses to edit their requests"

Why would the uses edit their requests?


By Spivonious on 10/2/2006 1:28:26 PM , Rating: 2
lmao!


By Viditor on 9/28/2006 9:07:19 PM , Rating: 2
quote:
an Open platform on AMD or Intel is not possible now

I'm not sure what you're saying here...
Hypertransport is now an open platform that is governed by the HT Consortium, not AMD.


AMD's HyperThreading
By subhajit on 9/27/2006 10:02:01 PM , Rating: 2
"Intel's primary goal is to introduce an alternative to AMD's HyperThreading"
I think you mean Hyper Transport.
It would be nice if they of them can come up with something that works for both platform. But I think that is out of the question for now.




RE: AMD's HyperThreading
By peternelson on 9/28/2006 2:39:29 AM , Rating: 2
Well, others have adopted hypertransport. eg Cisco.

That option is available to Intel, but they seem not to want to go down that road :-(

Agreed it would be better for us if there were common standards, but Intel is not cooperating.


RE: AMD's HyperThreading
By Hare on 9/28/2006 4:19:32 AM , Rating: 2
I believe Apples G5 systems also had hypertransport?


RE: AMD's HyperThreading
By JeffDM on 9/29/2006 6:00:02 PM , Rating: 2
The G5 Power Macs did, but it's between the memory controller and the I/O chip. The G5 chips themselves don't use HyperTransport.

http://developer.apple.com/documentation/Hardware/...
http://developer.apple.com/documentation/Hardware/...


RE: AMD's HyperThreading
By EclipsedAurora on 9/30/2006 1:27:09 PM , Rating: 2
U pointed out a good point.
Actually I think the move for both AMD and Intel are just an answer to IBM's Cell BE achitecture.

For most Cell BE (as well as Sony's PS3) system, the FlexIO is technical superior over anything Intel and AMD can offer even within forecastable futures. I think both AMD and Intel must do something to defense for their products.


They're serious now...
By Beh on 9/28/2006 12:36:39 AM , Rating: 2
I guess Intel doesnt get serious until they start to lose sales. Where's the innovation without competition?

I love capitalism.




RE: They're serious now...
By nerdye on 9/28/2006 1:18:40 AM , Rating: 2
I agree entirely, the netburst architecture was very expensive and underperforming the a64 up until the core 2 duo on the desktop came out just recently (yet the intel northwood was burly, and mobile is a different beast of course). Intel seemed quite intent on selling p4's so long as they would sell, but amd put a small fork in the road of their plans. Now I don't doubt the power of money in research and development in the cpu department as we watch intel dominate with conroe, and the expectations of quad core in nov 2006. We watch intel benefit from 65nm and prepare for 45nm as we sit idle waiting for the peoples champ, amd to return to performance greatness, it must start with a smaller manufacturing process...


RE: They're serious now...
By Tsuwamono on 9/28/2006 7:39:38 AM , Rating: 1
I believe AMD may have dropped the ball a few months ago when they saw conroes preliminary benchmarks. Im sure right now they are all working on something to counter conroe and probably kick the crap out of it since AMD seems almost to have a personality in these kinds of things. I think AMD is going to wait alittle bit longer so they can out do intel by the same margin or higher that conroe out did the A64 single cores.


RE: They're serious now...
By ShapeGSX on 9/28/2006 8:58:54 AM , Rating: 2
It takes years to design a microprocessor. And it usually takes a year to go from tape-out to a chip that can be shipped.

If you get caught off-guard by a competitor in this business, your reaction to that will come at least 2 (if you already have a good design to work from) to 3 years later.

That is why good processor road maps are so important.

Intel was clearly caught off guard by K8. Look at how long it took them to come back with a solid competitor.

K8L is essentially a finished product. And it was very close to finished when the Conroe benchmarks started coming out, too close to have been affected by the Conroe news. If AMD was truly caught off guard, AMD's reaction to Conroe can't be K8L. This is not to say that K8L can't compete with Conroe, of course. It may indeed.

Of course, Conroe is still in its infancy. :)


RE: They're serious now...
By ajfink on 9/28/2006 3:25:45 PM , Rating: 2
AMD has already announced that it has completed the primary design phase of the architecture that is going to completely replace the K8 line. That was months ago, in fact. However, we won't see that for years.


The introduction of an open FSB platform
By crystal clear on 9/28/2006 5:20:00 AM , Rating: 2
"It will be interesting to see what Intel's move into an open space will do for the industry."

This is the best opportunity for Venture capital Funded,
High Tech Start ups would be looking for,to make it big
in the computer business.
Its these small startups who with a small highly motivated & ambitious staff ,with their do or die attitude/fight for survival attitudes come with some facinating products.
Markets for such products is assured so financing comes easily at better terms.
If you are looking for a challenge/opportunity you got one
waiting for you to take.




RE: The introduction of an open FSB platform
By rcc on 9/28/2006 12:53:03 PM , Rating: 2
I feel as if I just got done reading haiku.

Crystal Clear???


However, there are indeed many opportunities inherent in both companies offerings.


By Spivonious on 10/2/2006 1:31:58 PM , Rating: 2
quote:
I feel as if I just got done reading haiku.


LOL! Priceless!


What about the unified FSB?
By JeffDM on 9/28/2006 1:44:57 PM , Rating: 2
Some analysts speculate that Intel will show off an open FSB specification in 2008 on Itanium, and on the Xeon sometime in 2009

I thought that Itanium and Xeon was supposed to share a socket well before then.




RE: What about the unified FSB?
By Viditor on 9/28/2006 9:04:17 PM , Rating: 2
quote:
I thought that Itanium and Xeon was supposed to share a socket well before then

Intel has cancelled that part of the design spec...


"If you can find a PS3 anywhere in North America that's been on shelves for more than five minutes, I'll give you 1,200 bucks for it." -- SCEA President Jack Tretton

Related Articles
Intel Pledges 1M Quad-Cores Before AMD Ships 1
September 27, 2006, 12:13 PM
Intel Shows Off Teraflop-on-a-Chip
September 26, 2006, 2:49 PM
Intel Talks 45nm Production
September 26, 2006, 2:19 PM
Intel "Kentsfield" Named Core 2 Quad
September 26, 2006, 1:53 PM













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