Home > Intel > Intel to Introduce CSI to Counter AMD’s HyperTransport Intel to Introduce CSI to Counter AMD’s HyperTransport Submitted by Mica Schunarez 2006-09-11 Intel 29 Comments Intel is getting ready to introduce a chip communications technology called Common System Interface (CSI) and also an integrated memory controller that mirrors an approach central to recent successes of rival Advanced Micro Devices. About The Author Eugenia Loli Ex-programmer, ex-editor in chief at OSNews.com, now a visual artist/filmmaker. Follow me on Twitter @EugeniaLoli 29 Comments 2006-09-11 4:56 am tmack Intel must really feel like Microsoft now. Always playing catchup to its competitors… 2006-09-11 7:43 am MamiyaOtaru “Always playing catchup to its competitors…” Except they passed them with Core 2, and with CSI (assuming it works) will just be piling it on. It’s a little bittersweet really. I used AMD for years and was glad to see them provide some competition. I jumped ship for Core 2 (simply better) but hoped they could maintain some sort of edge to prevent Intel from becoming complacent. I figured Hypertransport could let them hang on, with a big advantage as the number of cores increased. It seemed they’d continue to do well in server space, with a chance to get back in the desktop race with 65nm + SOI, but with Intel going quad core, and adopting a Hypertransport-alike, I worry about them. We the consumers need the competition of two companies striving to add new features and lower prices. CSI itself is a shining example of that, with Intel motivated to add it as a response to Hypertransport. I just hope AMD will have a counter so things don’t slow down again 2006-09-11 11:57 pm abraxas Except they passed them with Core 2, and with CSI (assuming it works) will just be piling it on. Hardly. CSI won’t be introduced until 2008 in Itanium processors. It probably won’t make it in x86 processors until another year. Don’t think AMD is standing still. They are revamping their DC technology and it will be introduced before CSI. 2006-09-11 5:06 am Eric Martin I prefer the name …. Dancing with the Stars ! 2006-09-11 8:32 am HelloWorld82 If Intel copies AMD direct connect architecture, they could at least make it compatible to HyperTransport (3.0!). The consumer would really beneficit from it: he would be able to connect intel’s and amd’s processor together. Sadly, I guess this will not happen. Intel mark strategie is “try to run AMD down”. 2006-09-11 9:54 am John Blink Yeah I would have prefered Intel joining the Hypertransport Consortium. http://www.hypertransport.org/ 2006-09-11 3:48 pm CaptainPinko Hmm, but would that help them integrate their Itanium line with x84-64 line? I think that is their main driving force behind the CSI. Sorry, it if it is a stupid question since I don’t know the specifics of HyperTransport. 2006-09-12 12:03 am rayiner They’re not copying Hypertransport. They’re implementing something based on the same principle (fast uni-directional serial-ish links), but many other busses are implemented the same way, and its not like AMD was the first one to it. 2006-09-11 11:06 am chrish You can really see the difference in corporate culture, or at least the image they’re trying to project… compare the amazingly dry and meaningless (hey, it could refer to ISA or PCI or USB) “Common System Interface” to the EXTREEEMEE!!1! “HyperTransport”. Intel = boring, corporate monolith going for the “enterprise” customers. AMD = hip, with it underdog going for the “enthusiast” customers. Except, of course, they sell into the exact same markets (with the apparent exception of Apple hardware, now that Dell seems to be using more AMD parts). – chrish 2006-09-11 11:17 am 1c3d0g That’s because Intel doesn’t have time to f–k around with “cool sounding names” etc. And it’s the reason why currently AMD is getting its ass smacked by Merom/Conroe/Woodcrest. 2006-09-11 1:13 pm sbenitezb “That’s because Intel doesn’t have time to f–k around with “cool sounding names” etc. And it’s the reason why currently AMD is getting its ass smacked by Merom/Conroe/Woodcrest.” They have enough time to f–k around with “stupid sounding names” though. BTW, People often seem to like monopolies, but don’t realize that without AMD around, probably we would be running 486 at the price of thousands. 2006-09-11 4:19 pm NotParker “AMD is getting its ass smacked by Merom/Conroe/Woodcrest” Currently Intel only produces about 5% of it total chip output as Merom/Conroe/Woodcrest. The rest of its output is still crappy Xeons, P4’s, 8xx and 9xx chips that suck up electricity like its free. That means most of AMD’s production is a better choice than 95% of Intels production Also, Woodcrest is only working in 2P motherboards. Anyone who wants a 4P or better motherboard should buy Opterons and avoid Xeon’s like the plague. 2006-09-12 12:05 am rayiner No, the reason AMD is getting its ass smacked by Core 2 is: 1) AMD isn’t on 65nm yet; 2) AMD can’t keep a chip dev team together to save their lives. The team that did K7 and K8 are no longer in tact. The K7/K8 core has had a great run, but the basic design has changed only moderately since 1999 — it’s time for a new core, and AMD doesn’t seem to have one ready. 2006-09-12 3:34 pm k00k The basic design of the K7/K8 changed only moderately? I dunno about you, but the K7 was an elegant solution for its time, and it did beat Intel’s NetBurst at one thing: proving that performance and efficiency could come hand in hand. And you do realize that the design for the current Core and Core Duo processors stem from the Pentium M–which, in turn, is essentially an improvement over something ‘old’ too–yes, the P6 core and its ilk. Look at it this way: this only tells us that one need not completely change to another design paradigm just to keep competitive. (NetBurst vs. everything else comes to mind) This can be seen in Intel’s P6, Yonah and all its offspring, as well as ATI’s R300 and the designs that came from it, such as the R420. More than anything, this only tells us that from a common-sense POV, ASIC design still remains more evolutionary than revolutionary–and i’m willing to bet that AMD still has a lot of things going for it with the things that were spurred by the K7 and K8. Don’t count AMD out ‘just because’ they only have the K8 as of now. Not just yet. Their recent Dell win is more than enough of a statement that says AMD will still continue to survive and even thrive. Edited 2006-09-12 15:36 2006-09-12 9:12 pm rayiner The basic design of the K7/K8 changed only moderately? Yes. The K7 -> K8 transition was much more conservative than the Yonah -> Core 2 transition. K8 is a very conservative evolution of K7. It was a new core in the sense that the circuit layout was completely redone, but the major architectural changes was the widening of the integer pipeline to 64-bit, deeper issue queues, wider L2 path, and the IMC. Core 2, on the other hand, is almost a completely new core, with deep changes throughout the OOO and memory units. AMD is not going to catch up to Intel just by tweeking K8, at least not in integer performance. Core 2 is just a deeper and wider design than K8, and to compete with it, AMD is going to need to go deeper and wider. It’s not entirely clear that K8 can be scaled further. Edited 2006-09-12 21:12 2006-09-11 1:50 pm jamesd If Intel quad core cpu had this technology when it ships, then intel would be much more competitive. But instead they increase core count by 100%, front side bus by 30%, and see the majority of benchmarks increase by 30-50%. That is not exactly scalalible. This is why all the reviews are focusing on the home market. I would suspect that Intel quad core chips just dont have the ability to process high throughput transactions. sure they are fine for simple video games that work on the same 4GB of data. But when you have 2GB of data coming into the system per second, over various gigabit networks and multiple sata and sas/scsi , in addition accessing memory, all in one shared front side bus, the quad cores will sit spinning waiting for data. Perhaps with this new technology Intel will have a chance to catch AMD in server market, but I have a feeling that technology will sneak in and keep Intel from winning the battle. Yes CSI seems to answer the Memory bandwidth problem, allowing it to have direct connections to the memory, but there doesn’t seem to be a CSI connection to IO busses. AMD opteron chips give you a seperate link to the memory like CSI does, but AMD opterons give you 3 other links to other cpus and IO busses. Sure today Intel/CSI can make do with a single link to the ram and another to the IO buss. But will they be able to handle 10 gigE connections allong with multiple sata II disk controllers at 300MB/s each. Sun already has a system with 48 sata II controllers and 4x gigE network adaptors on board, with a few open IO slots for future expandibility how well do you think intel’s CSI + FSB will cope with that much data. Sun’s niagaraII chip that is being tested, has two 10gigE on die, in addition to on board memory controllers and hypertransport. Because AMD and Sun see the light, they know if there is no data to process, the cpu is just spinning its wheels. Intel is just now relising this fact and will take a small step in rectifying it, in 2008 when CSI enabled chips ship. But is it enough? 2006-09-11 3:41 pm slashdev Didnt Intel already release a “counter” to hypertransport years ago? I cant remember the name of it. But Intel kept saying that “HyperTransport wasnt enough bandwidth, it wont last, etc, etc”. 2006-09-11 10:48 pm atomicplayboy Maybe you’re thinking of PCIe? Intel’s been waving hypertransport off for a while now, saying it’s a niche market and all. They also said that 64 bit instructions on x86 weren’t needed/wanted and that they were going to stay focused on the ghz, rather then increase the cores in their cpus. Intel’s strategy these days seems to be to insult the competition, fail, and then copy the competition. They may have the performance edge right now, but they’ve been playing catch up for so long, I don’t see how they can keep that edge for very much longer. 2006-09-11 3:48 pm bnolsen I’m still mad at AMD for sitting on its rear end. I mean they pretty much had Intel technologically they released the Athlon64 series. Now it’s making them tons of money. So the question is, what did they do with the past 3 years? It seems when the A64 released they dropped R&D in favor of minor maintenance activities. Since the A64: – Modify onboard memory controller for dual channel memory. – Leverage existing A64 design to add dual core. (follow intel) – Add SSE3 instruction compatability. (follow intel) – Up the hypertransport speed from 1600 to 2000MHz. – CPU frequency scaling (follow intel) – Modify onboard memory controller to support DDR2. – Add hardware virtualization support (follow intel) AMD’s release of the A64 was an industry leading event. What happened to them?? Where’s their clear *cool* roadmap? Or now they just decided to wake up, and it’s 2 years too late? 2006-09-11 10:59 pm atomicplayboy – Leverage existing A64 design to add dual core. (follow intel) This is just simply wrong. AMD had dual core well before intel did, not to mention that they designed the AMD64 architecture with core scaling in mind. This will be their biggest advantage if it gets to a core battle with intel. I think AMD is trying to appeal to the server market at the moment, and not create any massive architecture changes. They may not have a *cool* roadmap, but they’ve got a clear and consice one, which can not be said for Intel. I like seeing the underdog ahead of the major manufacturer as well, but if this keeps the money coming to AMD for more R&D, then I’m happy. 2006-09-12 6:15 pm phoenix I’m still mad at AMD for sitting on its rear end. I mean they pretty much had Intel technologically they released the Athlon64 series. Now it’s making them tons of money. So the question is, what did they do with the past 3 years? I’ve never understood this attitude. Why do people consider tech companies to be failures if they don’t release new, updated, better versions every 3 months? What ever happend to building a good product, that will last? Make small changes, incremental improvements, but without requiring wholesale changes to the underlying support hardware? For instance, with AMD CPUs, you can plug in the latest and greatest CPU (with a few exceptions) into a first gen mobo. Every minor speed bump in Intel-land requires a new chipset, a new socket, and new RAM. Modify onboard memory controller for dual channel memory. – Leverage existing A64 design to add dual core. (follow intel) AMD had dual-core CPUs before Intel did. Intel released a so-called “dual-core” CPU that was nothing more than two separate CPUs on the same PCB that plugged into a single CPU slot. Yeah, that’s great tech, there. AMD released a true, designed-to-be dual-core CPU long before Intel ever did. – Add SSE3 instruction compatability. (follow intel) Considering Intel designed SSE1, SSE2, and SSE3, how do you expect AMD to release them first? You’re also forgetting about 3DNow!, 3DNow! Pro, and the other 3DNow! release from AMD that Intel doesn’t support. Don’t forget that AMD brought out x86-64 before Intel, so it’s not as one-sided as you think. 2006-09-11 3:52 pm slashdev http://news.com.com/2100-1001-270578.html http://www.findarticles.com/p/articles/mi_m0EKF/is_32_47/ai_7710588… _QUOTE from news.com_ Intel’s Grilley praised the move toward making HyperTransport a more open standard rather than a company’s proprietary technology, Grilley said, but Intel believes HyperTransport will run out of steam too soon. “HyperTransport is…a good five-year solution” for connecting chips within a computer, Grilley said. “We just think there needs to be a 10-year solution. What we want to do with the 3GIO is to go one step farther.” _END QUOTE from news.com_ So here it is 2006….Five years later. Hypertransport is still here and doesnt look like its going away at least for the next 4-5 years because single HT connections have been getting faster and you can add as many single HT connections as you need for bandwidth. Now, what happened to 3GIO? did it change its name to PCI Express or is this CSI… 2006-09-11 9:34 pm ronaldst AMD is in major trouble. Intel has clearly better CPUs in every dept. This isn’t a good time for AMD. Better speed. Cooler to boot. They’ll loose gamers by the bucket. How can AMD turn this tide of bad luck around? 2006-09-12 12:11 am rayiner AMD isn’t really concerned about gamers. They’re $$$ come from the server market, and the low-end computer market. A 65nm Athlon X2 shrink will be good for the latter (being *substantially* smaller and thus cheaper than Core 2), while Opteron still has the high-end. Remember, the high-end buys the infrastructure more than it buys the chip. Opteron has a very mature infrastructure based around HT already in place. Intel doesn’t have the same thing, yet, for Core 2, and it’ll be a year or more before the level of maturity is comparable. If AMD has a new core lined up for Q1 2008, they might lose the enthusiast market, but should be able to hold out on the server front. 2006-09-12 6:24 pm phoenix AMD is in major trouble. Intel has clearly better CPUs in every dept. Intel has a better laptop CPU (Pentium-M, Core, Core2), yes. Although for low-end laptops, the Turion64 is a better buy. Intel has a better dekstop CPU (Core2), yes. Although the Sempron may be a bit cheaper. But AMD has the better server CPU (Opteron). Nothing Intel has comes close to a multi-CPU setup using Opterons. For single-CPU servers, a high-priced Xeon setup may be better at some things than a less expensive single-CPU Opteron server. But put a dual-core CPU in, or use multiple CPUs, and even the Core2 systems can’t keep up. And when you consider that server CPUs have the highest margins, and make up the bulk of AMD and Intel’s revenues, I really don’t see how AMD can be “in trouble”. 2006-09-11 10:04 pm atari05 I’m not saying they don’t need to be thinking and coming up with some numbers, cause lets face it..numbers talk but would you rather AMD, the company we have all come to love, release some medicore technology to gain some speed and performance and then later suffer for it ala Intel? Sure I would love to see AMD come back quick with a “HA!!! CONROE?, We got <killerhardwarehere> but at the same time, I trust AMD not to release crap to its customer base. I rather see them wait, come at me with some good tech that will be scailable and last like they did when they hit us with AM64. Like someone else mentioned ..its been 5 years HT is still around and is a good solution and its performed EXTREMELY well. That to me is a better solution than the intel “Hey, this will last us a generation..but…uhh….how are we going to shoe horn this tech into the next gen”. Don’t get me wrong, I think conroe is impressive and I give Intel some “props” for a fast display of product turnaround but still….I think people are jumping the gun. I for one will wait until next year to give AMD the grim sticky note of death. 2006-09-11 11:23 pm kscguru From the article: > That means CSI won’t arrive in high-end Xeons until 2009. Wow. Amazing how a product 3 years away from market suddenly gives life back to Intel. Because, you know, nothing could possibly happen during those 3 years. 2006-09-12 6:50 am Anon Why do Intel NEED to make an alternative transport bus? Hypertransport is a standard all can use. Intel have just pulled a 100% Microsoft like tactic. Ignore a standard, make some other useless incompatable one. 2006-09-12 9:16 pm werfu Actualy, AMD is far to be in trouble. Fanatics have cried for a new architecture when Intel launched the Core and so the AM2 is born. It may have left a lot of people somewhat disappointed, but the move AMD has made with the AM2 plateform is more strategic than the big bang Intel has made with it’s Core architecture. With it’s AM2, AMD has planned the way for a much bigger evolution comming slowly in 2008, the AM3, which promise a lot more than the current tech bump. AM2 only give DDR2, some new frequency on the processor side, and introduction of the new virtualisation tech (Pacifica I think, not sure). I guess we can be sure the 65nm SOI is near of production and they shall support higher hyper-transport and RAM frequency, pushing forth on the performance. Still, This will only be a small “upgrade” compare to what will come with AM3. You can already guess that with this new socket, a brand new revision of hyper-transport will come, with even higher clock, a brand new memory controler will integrate DDR3, and you can already say it, AMD will kick Intel’s butt on the multi-core battle, having a better inter-connect architecture. Also, on the power consumption side, Intel may have momentary the lead, but it will not last long, AMD going to the 65nm soon. We can expect that the R&D on the Turion side will bring back some success on the desktop and server too. And after all, TCO may be important, but not as much as the actual efficiency to get the job done, even if it mean an higher electricity bill at the end of the month.