Transmeta Will Help AMD Make Code-Morphing Chips 73
Mark Imbriaco (and company!) writes: "This story at News.com talks about how AMD is working with Transmeta to ship developers systems using the processor instructions from their upcoming Sledgehammer chip -- apparently Transmeta is working on a version of their code-morphing software that supports this instruction set. In return they seem to be getting a license to make chips using parts of the Sledgehammer design. If it's true, it's a pretty cool step for Transmeta since their other products to date have gotten a mostly lukewarm response over the past couple of months."
Re:Cool For AMD As Well (Score:2)
Yeah, I'm thinking the same thing. Your Subject: line is prophetic - "Cool for AMD".
The only real barrier I see AMD having in the mobile market is heat. PowerNow (that thing where they underclock/undervolt the CPU during times of low load) should help considerably, but what I'd really like to see is how much of TMTA's "we run super-low-power and super-cool" technology AMD can l33ch.
As I see it, the world doesn't need a laptop emulating Sledgehammer anywhere near as much as it needs a laptop with a cool-running 1.2 GHz Athlon at a fraction of the cost of a PIII.
(Of course, if Sledge lives up to its billing, it'll be faster emulated on a Transmeta chip than Intel's solution anyway, so I could be missing the forest for the trees here. I gotta admit, a cool-running Sledge laptop would also kick mucho blue man ass.)
Re:SMP that the OS dosn't see ? (Score:2)
I can't guess one way or the other... but your prediction does fit nicely with the multiprocessor capabilities of the Hammer line. Could be damn interesting if you're right. (And a great way to sell lots of CPUs per box... if you're a chipmaker, you stop caring if PC makers are selling less boxen when you're regularly selling more than one chip per box.)
What are they really doing? (Score:1)
Re:Cue the Peter Gabriel... (Score:1)
--
Re:Cue the Peter Gabriel... (Score:1)
However, I'd like to be corrected if I'm mistaken. :)
Re:Cue the Peter Gabriel... (Score:3)
Oddly enough, so are the Blue Men. [apple.com]
Re:Cue the Peter Gabriel... (Score:2)
#include "disclaim.h"
"All the best people in life seem to like LINUX." - Steve Wozniak
Re:Sever Market (Score:1)
Well... this case is a bit different... Usually AMD has created chips that are *compatible* with Intels offerings... This time they aren't. I don't know, but i'm guessing that this time Intel patented their instruction set or otherwise protected it in order to make sure no AMD's arise on that front for many years.
So it's just a matter of which architecture developers follow. That said. I'm betting AMD will turn out the big winner, in light of they're not looking to abandon their platform as Intel is. The Itanium looked really powerful on paper a few years back when Intel was saying it'd debut at 600 or 700 MHz, but now AMD's at 1.2 GHz, the P4 is at 1.4 GHz, et al...And as much as Intel wants it, the market they're going after is actually quite small. Yes, high margin, but still incredibly small compared to the market which brought them their current riches.... And their chip will be the least estabilshed brand. So while AMD can eventually "float" the higher end market, intel's offering basically end up being stuck in a sink or swim predicament...
ANd if that happens... What'll AMD decide to charge for their chips if they've clearly bested intel?
1186.20 on RedHat 7.0 with default kernel (Score:1)
.technomancer
Re:Sever Market (Score:2)
Re:Sever Market (Score:1)
Merced/Itanium/IA64 hasn't been subjected to that competition at all. Intel's just been taking their time designing it, but already, the chip is proving difficult to run past 600 or 700 MHz. Thats the same speed that it was expected to arrive at years back, when we were using 233, 266 and 300 MHz Pentium II's.
Yes, more instructions per cycle, but the race between Athlon and Pentium is pushing x86 performance higher than Intel ever expected. So, for all but the most memory intensive operations, the x86 may be able to win out over Itanium just based on clockspeed. The expectations of the Itanium haven't been increased to reflect market conditions.
If that makes sense?
Re:SMP that the OS dosn't see ? (Score:3)
> involved here ?
Sure. What you're talking about is impossible.
You assume that software is inherently fully paralellizable, and that's very far from the truth. A huge amount of work has been done on automatic parallelisation, and the results are not very promising. Automatic parallelisation is difficult to implement, takes a lot of computation, and in the end it only works for certain kinds of programs (programs which are dominated by loops with predictable structure).
Many programs are inherently serial and cannot be parallelised. Automatic parallelisation also tends to rely on static analysis, which is basically impossible to do at the level of binary machine code.
The Stanford Hydra processor does a form of automatic parallelisation of binary code using thread-level data speculation, but there's plenty of evidence that the speedups are not that great without support from the compiler.
If your hardware supported thread-level data speculation (none does at present), then I could believe that code morphing might be able to produce significant speedups using parallelisation, but the code morphing process would be much slower and the speedups would be much less than proportional to the number of processors.
Re:Umm.. (Score:1)
Hmm.. (Score:5)
AMD isnt' redesigning Sledgehammer to include Transmeta tech. They're using Cruoses as development simulators to get developers to port their code to the Sledgehammer architecture *before* the silicon hits the shelves. Today, this is NOTHING MORE than a way for AMD to ship fast enough simulators so that ppl can start coding for the Sledgehammer.
In the long term however - it'll allow transmeta to develop Sledgehammer compatible chips - but that's a long way off. (BTW, Is it only me that thinks that they targeted their chips at the mobile market as an afterthought "Oops guys, we can't get this to run fast enough. What to do?" "Hmm.. we'll call it a mobile chip.")
-henrik
Re:Cue the Peter Gabriel... (Score:1)
Now THAT'S the ad campaign AMD needs to get "mindshare". It would beat Intel's Blue Men, and DEFINITELY the guys in the shiny clean suits with the pseudo-rock playing in the background.
The 'Intel Inside' melodic bong-bong-bong-bong! is kinda catchy, though.
--Just Another Pimp A$$ Perl Hacker
Re:Cool For AMD As Well (Score:3)
Assuming that the Sledgehammer code can run (decently) on existing Transmeta chips, then they may beat Intel to market on the 'next generation 64-bit x86 successor', in a laptop version. This leaves AMD to concentrate on a desktop/server processor without worrying about mobile concerns (heat/power/size).
It would also give transmeta a shot in the arm since they would be in the position of offering something that no one else had (a laptop ready version of Sledgehammer).
Re:Feasible performance? (Score:1)
(Or not.)
Cheers.
Re:Take that Intel (Score:3)
Unfortunately it seems like businesses treat Intel and AMD like Microsoft and Open Source. Will this help change anything?
"Why don't use Open Source/AMD instead of Microsoft/Intel?"
"Well, we have been doing pretty good with Microsoft/Intel and and we don't see a reason to change."
"Can't you see that Open Source/AMD is better, fast, and cheaper?"
"Yeah, but change is bad..."
Umm.. (Score:2)
SMP that the OS dosn't see ? (Score:3)
This means that you could have a machine that Linux or Win2K recognizes as simply 2 way SMP ( nearly liner scaling at that level ) but in reality it's 16 CPUs in gangs of 8. Even with the overhead from code morphing being high ( and I have no evidence that it is ) this could still perform like a 12 CPU dream machine.
Then again I may be totally wrong and this "prediction" has no connection with reality whatsoever.
Anyone care to enlighten me as to what is really involved here ?
Re:Sever Market (Score:3)
What, never heard of the Compaq Alpha?
Check some of the examples at Polywell [polywell.com]. You can find more with some simple searches, like Yahoo [yahoo.com]
Relevent story on The Register [theregister.co.uk].
Oddly enough, Polywell indicates they are _shipping_ Athlon DDR systems [polywell.com].
--
+++ Out Of Cheese Error +++
+++ MELON MELON MELON +++
Re:Umm.. (Score:5)
And Transmeta got it. The AMD deal has allowed Transmeta to produce the only things they've been competitively successful at shipping: Press Releases and Slashdot articles.
Re:SMP that the OS dosn't see ? (Score:1)
Of course, I don't have as wide of a field of knowledge in this area as some of the other people in this forum, but this is my guess
Sounds like AMD needs a fast Sledgehammer Simul. (Score:1)
It sounds like AMD's software simulator for the 64 bit Sledgehammer was too slow to use
Imagine what a developer gets on their desk - a hardware box that looks and feels like a 64 bit Sledgehammer, except the BIOS initializes the CPU a little differently and it's a bit slower than the real thing, which doesn't exist yet. Sure beats the hell out of simulating everything, including the I/O, which is the other alternative.
Random acception in stories (Score:1)
Slashdot needs to get it's act together, and the moderators will probably troll me for insulting them, but it's true
Re:Release date for Sledgehammer? (Score:1)
AMD's design, as reported so far, unlike Intel's design, does NOT use 32 bit emulation in the sense that your post implies.
From the latest I've seen, AMD's chip runs 32 bit code natively, on a pair of 32 bit cores that are capable of being operated together in 64 bit mode. There is no redesign of the emulation facilities because the Sledgehammer doesn't use it. That will be one prime advantage of the Hammer series of chips over Intel's 64 bit chips.
AMD & Transmeta's goal is to get a good, fast Sledgehammer emulator into the market well in advance of the actual Sledgehammer chip.
Please read the articles more carefully before posting.
Re:oh well (Score:1)
Re:Sever Market (Score:1)
I would say that's unlikely. An Alpha will outperform a higher clockspeed x86 chip. It's more than just more instructions per cycle. 64 bit instructions help just about everything. I haven't looked at the spec for the instruction set, but I'm guessing that Intel has dumped a lot of the crap that the x86 instruction set had. There are instructions for BCDs in the x86 instruction set. Does anyone use them these days? Not likely. Even if they were used they would probably be slower since Intel's strategy lately has been to optimize the simpler instructions in imitation of the RISC architecture. Optimized code uses a bunch of little instructions rather than one slower equivalent instruction. However, the other instructions are just dead weight. Getting rid of the crap leaves more chip real estate for optimizing the simpler, more general purpose instructions.
"Homo sum: humani nil a me alienum puto"
(I am a man: nothing human is alien to me)
Re:Hmm.. (Score:2)
There's no reason why Transmeta sledgehammer compatible chips are a long way off. First off, the simulator chip could be completely sledgehammer compatible, albeit a bit slow on actual 64-bit operations, since the Transmeta chips apparently currently have 32-bit ALUs. Second, extending the Transmeta chips to have 64-bit ALUs probably isn't that expensive. The MIPS guys have reported how much complexity it added to the R4000... it wasn't bad.
Re:Hmm.. (Score:1)
-henrik
Re:Take that Intel (Score:1)
"Yeah, but change is bad..."
Oh please. I've used AMD in my home system since the early K6's, but I still won't put them in a server. Why?
I have no problem with AMD CPUs, but nobody seems interested in targetting the server market with them.
So Happy (Score:1)
Shot in the Arm (Score:2)
Unfortunately, as all know, the recent failure of Transmeta to really secure a company willing to openly use its technology (with the one exception of Sony) has hurt it in the public eye. I'm willing to bet that, at least on some level, irrepreable damage was done a week after launch when compaq pulled out, and IBM chose to rethink their position.
However, a new major backer, especially another semi-established chip company, might just be the major backer that T needs to go from in the dark chip company to what it really should be, a publically watch company with a great force in the future.
Sever Market (Score:3)
If amds chips will perform as well as intels chips do in servers this will very good for consumers, since amds chips cost 30% less than intels do(usually). This will allow smaller companies to have much faster servers at a better price. Seems like a win-win situation, well except for intel(which will still get the high-end market share,for now anyway)
Sledgehammer is one of AMD's most ambitious projects to date. The chip will process data in 64-bit chunks, rather than in 32 bits like AMD's Athlon processor. Sledgehammer also will allow computers to manage more memory than current PCs and servers. The chip will compete against the long-awaited Itanium processor from Intel.
Finally a 64 bit processor, and with amd the server chips might be affordable.
More Hype? (Score:2)
Let's just hope that more hype like this about Transmeta doesn't totally finish them off.
Ha (Score:2)
Re:Feasible performance? (Score:1)
oh well (Score:1)
Re:Take that Intel (Score:1)
It only makes sense for these two underdogs to join together to take on Intel, but they've got a lot of catching up to do in the public eye before they become a serious threat.
Release date for Sledgehammer? (Score:2)
Sledgehammer, does that mean a later release date for the chip? How
does that affect the likelihood of uptake for the new chip?
I don't know what to make about this story. My gut feeling is that
AMD is casting about in its 64-bit strategy after not getting the
support it had been counting on from Microsoft.
Re:I had to restore a Sony Crusoe laptop yesterday (Score:2)
Considering I can restore my old NCR 486-66 from a parallel port 2x CDROM drive (image is 685M, gzipped) plus install Windows 95 manually in a little under two hours, I'd say it came down to either operator incompetance or a CDROM drive banging against bad media.
maybe i'm wrong (Score:1)
- transmeta releases its chip. Sony, hitachi and others think "interesting! just build up something". IBM awaits.
- IBM refuses to implement the TM in its laptops, adducting (and being right about) performance problems. In the underground, develop his code morphing engine.
- Transmeta, without the IBM support, now search someone to catch in the idea, just not to fight alone between giants.
Adding code morphing capability to AMD Processor should reduce the overheat who often afflicts the AMD processors. So everyone (TM and AMD) have something to learn in this partnership.
Re:Take that Intel, eh? (Score:3)
--
+++ Out Of Cheese Error +++
+++ MELON MELON MELON +++
Memory Access Still a Dilemma (Score:5)
The latter "bit" is no small matter, and the potential of having a bunch of CPUs on one chip doesn't make these issues go away.
These issues are essentially why AMD (and Cyrix and others) haven't had SMP systems; it's costly to construct the logic necessary to let multiple CPUs play on the same set of memory buses without trampling on one another, and the tradition of AMD/Cyrix being "low end" was just not compatible with spending the money to build that.
I'm still skeptical that there will be any massive movement by AMD towards SMP. And the introduction of some "cool code morphing" from Transmeta doesn't do anything to simplify or otherwise resolve this.
I would think that there could be some pretty slick results from an AMD/Transmeta technology transfer; it's just that SMP doesn't seem high on the list of "obvious cool things."
Interesting possibilities... (Score:2)
Ok, let's say Transmeta and AMD go all out on this partnership and exploit their respective strengths...
The last possibility is really interesting. I bet AMD and Transmeta have thought of this too.
An innocent question (Score:1)
The other side to competition (Score:3)
This is an example that should be followed more widely. The strength of each company is being utilized here, and the end result is better for everyone: AMD, Transmeta, and the consumer.
This is also an example of why most of the people reading this site despise Microsoft so much. When was the last time MS worked together with another company to make a better product? It's so much easier for them to simply buy the competitor, or crush them with their own product, haphazardly cooked up with a flurry of resources. MS is unfortunately in the position where they can ignore QA and true innovation due to their almost gauranteed market share for any product they introduce.
May more companies follow the lead of AMD and Transmeta! Oh, and Happy New Year!!
---
Re:Hmm.. (Score:2)
Your comment doesn't fit the text of the article. AMD clearly is seeking a fast simulator for the x86-64 instruction set. Whether Transmeta licenses LDT or not is a separate issue. And whether or not you think Transmeta's technology is new or not, it certainly is a dandy way to build a fast simulator for an x86-related instruction set.
Re:SMP that the OS dosn't see ? (Score:1)
--
Re:SMP that the OS dosn't see ? (Score:1)
This would make a massive number of Crusoe chips linked together a bad joke.
This is just a development tool (Score:3)
We're talking about a modest number of development systems here, not a production product.
Re:Hmm.. (Score:2)
Not sure if you are joking, so I'm answering seriously. It's clear to me they had the mobile market in mind all along; their design is tiny and dissipates a very small amount of heat, and they have a feature for dynamically changing the clock rate. Existing "mobile" chips have a feature for idling the CPU to save power, but Transmeta can actually slow the clock rate down. More on this, plus cool infrared photos of heat dissipation, here [transmeta.com].
I would seriously love to have a desktop computer with two or four Crusoe chips in it. I dream of having a computer running Linux quickly yet as quiet as the Atari 520 ST we used to have. (The Atari had no cooling fans, and no hard disk; if the floppy disks were idle, it was silent.) If a single 600 MHz Crusoe runs about as fast as a Celeron/300, two of them ought to be plenty for reading my mail and such. And it should be possible to cool them with just heat sinks.
steveha
Re:SMP that the OS dosn't see ? (Score:1)
This is certainly true, but then there are a fair few things of this nature which, if code is tailored for, can be massively optimised.
The example that springs to mind is Quake-style rendering routines... If this code was suitably analysed (at copmile time i guess), a multi-processor array could safely parallelise the code and gain a linear performance improvement.
The point about thread-level data speculation is valid, and there again it is a matter of annotated compilation. Roll on next-generation languages for parallel systems. Code morphing is not the answer...
Cue the Peter Gabriel... (Score:2)
#include "disclaim.h"
"All the best people in life seem to like LINUX." - Steve Wozniak
Take that Intel (Score:3)
Now AMD is taking market share away from Intel, their chips are better and Intel is the one who can't manufacture anything in quantity to save their lives. To top it off, the AMD design is better.
Who does Transmeta turn to in that situation. It isn't hard to see that Intel isn't going to want to help them politically, but since AMD is a better choice anyway they go with the underdog.
Sweeeet.
--
Cool For AMD As Well (Score:2)
Slightly off topic, I wonder if Sony would consider creating low cost Playstation2 development systems with Transmeta morphing software...
Bryan R.
Feasible performance? (Score:2)
So... there's no way of telling how well your code is actually performing? Nice step forwards with regards to translatable code and stuff, but you wouldn't exactly develop a 3d engine on a 386 these days...
Also, what are the implications for the Transmeta chip? If interoperability is one of the USPs, then a 15-year old performance isn't good. However, the article also mentions:
by two girls, actually. (Score:1)
i bet i get negative Karma for that here, but it speaks well of the coming year for me.
Re:SMP that the OS dosn't see ? (Score:1)
Re:Feasible performance? (Score:1)
REAL STORY (Score:2)
Re:Take that Intel (Score:1)
I am not just talking about the Server market. How many desktop PCs at your office use Intel microprocessors? How many use AMD?
If I was unsure of using AMD on not as mission critical applications, such as desktops, why would I even think of trying AMD on a server when it becomes available. My point is the rut that a business can get into where the benefits have to be much higher than would logically require a change. If there was a way to reduce that rut, all new and better technologies would benefit. I don't know if it is more based on the human fear of change or the risk factor.
Re:Release date for Sledgehammer? (Score:1)
If your Karma is greater than 25, you post on
Yep. I do. In fact, this is my second account. My first, with a 4-digit ID#, had 98 karma when the caps came in, and only seeing the number go down depressed me. When this one goes up the handful of extra points to hit 50, I'll create another one.
Re:Memory Access Still a Dilemma (Score:1)
Re:Hmm.. (Score:1)
Re:Sever Market (Score:2)
Scratch that. Try to put that in a 2-processors system.
Scratch that. The P4 is not able to do SMP. Pity, isn't it?
SMP is kind of a must-have in servers doing work, and I mean lots of work. The P3-Xeon is sort of the best thing out there in the x86 word (but sadly it only scales up to 8way AFAIK. Compare that to UltraSparc, whose hardware could theoretically scale up to 64k processors, and routinely scales up to 64).
More of the same: when you go to the computers manifacturers sell as "servers", the CPU cost is actually tiny when compared to the whole system. Usually you have custom-built motherboards with nifty monitoring features, custom-tailored cases to go with the custom motherboards. You have to have ECC RAM, actually lots of it, and the controllers that go with that. Two redundant power supplies, willya? And then a battery-backupped RAID controller, or RAID will be no good at all. And of course the disks, all of them UW160SCSI, and usually you have a support contract with replacement clauses for them. That is, unless you go with shared fibre-channel or whatever else. As an afterthought, redundant NICs also add some.
In all of this, the CPUs weight actually less than 10% of the total hardware cost for your typical x86 "server".
If you also add redundancy (you need to double the server to get decent availability of course), and costs because of downtime (no x86 server comes with hot-plug CPUs or memory AFAIK, and only a few have hot-plug power supplies, while most have hot-plug HDDs), well in these conditions you'll find that alphas or sparcs or mips actually come pretty cheap, bang for buck.
Re:Sever Market (Score:2)
Data General have been shipping 32-way Xeon boxes for some time now, in the form of their AV25000 [dg.com] server. They're also about to release the 64-way AV35000 [dg.com] shortly. Obviously, that all comes with redundant power, cooling, storage etc. They run DG/UX, or NT, but if you run NT, you can only use 4 of the CPUs at once :-) You can always run multiple system images on one box, but each NT
image can only access 4 CPUs. Naturally, DG/UX
can use the whole machine...
Re:Sever Market (Score:2)
Re:Sever Market (Score:2)
Sorry.
Eagerly... scared (Score:1)
Well, I have to admit I love Transmeta for how they are stirring things up in the CPU market. At the same time, having been closely affected by their recent recall [slashdot.org], I shudder at the thought of their name being put right next to "code-morphing chips"...
Me:I just wanted to compile the 2.4 kernel...
It:Here you go - (hands me win32)
Yes, I'm just kidding, but... hey, the possibilities are endless...
Re:Feasible performance? (Score:2)
This is referring to their existing simulator, not the Transmeta one. And at any rate, for the purpose of optimizing your software (especially server software where you don't need to make the kind of subjective evaluations about screen refresh being fast enough like you do in a game) you don't need a processor that runs at full speed, you just need to be able to profile your software to identify which loops are eating up too many cycles, etc.
Re:Feasible performance? (Score:1)
Feasible performance? Oh yeah... (Score:2)
So... there's no way of telling how well your code is actually performing? Nice step forwards with regards to translatable code and stuff, but you wouldn't exactly develop a 3d engine on a 386 these days...
Actually, this is the performance of the systems they sent out about a year ago. Their hopes are that the Transmeta based systems will be faster and make up for that 'speed' problem.
Bryan R.