Slashdot Mirror
AMD Announces August Release Date for Barcelona
An anonymous reader writes "Rumors said the release wouldn't be until late Q4 but an August ship date is now promised for AMD's quad-core chips. They're only releasing up to 2.0 GHz processors at first, with the top speed devices coming out later in the year. 'AMD's Barcelona puts four cores on a single slice of silicon, an approach AMD calls native quad-core, and the company has argued that Barcelona will outperform the Xeon 5300. The only problem: that comparison soon will become obsolete. Intel's second-generation quad-core server processors, Harpertown a server member of Intel's Penryn family, will arrive this year, too, with the promise of better performance, lower power consumption and lower manufacturing costs by virtue of a manufacturing process with 45-nanometer features. AMD is only just now moving to a 65-nanometer process.'"
Lowering the costs of manufacturing may be nice, but I don't think it's a particularly significant factor in their pricing model.
Thanks, anyways.
AMD is only just now moving to a 65-nanometer process
That's a nice thought, except it's totally wrong. All their Brisbane core X2 chips are on 65nm now, and have been for quite awhile.
And when is it gonna be released outside of Spain? Good thing we have eBay to mitigate this regional bullshit.
...of the launch event for "Barcelona".
It's so good, they persuaded Freddie Mercury to come back from the dead for it!
Moving to native quad core has a lot of advantages and I'm actually excited to see how well this CPU will perform. Critics that claim that AMD lags behind in the process size would do well to note that AMD has ALWAYS lagged behind Intel in that category, and, yet, has managed to not only survive, but prosper.
This is my sig.
Wow, it'll be 6 or 7 usable ghz of computing power! WOW! We're a whole 1.4% of the way to that 500ghz CPU power that we were talking about in yesterdays article! Can you smell it? The future is coming!
*snicker*
AMD's HyperTransport is as far as I know a major advantage in "big boy stuff", like physics simulations. Does intel's memory transport still suck in 8-cpu configurations?
I hope they manage to avoid the problems that hit Intel's Core 2 Duo, and squash some of their own bugs while they're at it. Or is it too late to handle those problems at this stage in the manufacturing process?
"What do you think?" "I think 'What, do you think?!'"
I'm sorry... I couldn't understand anything in this article at all...
I managed to gather "intel ^&^T& iPhone *&^&*^ iPhone &*^*&^* 45nm *&^*&^ single slice&^(*&^(& iPhone"
Games can be specialized to use 4 or more cores.
Servers will really use it.
Mr. PC enthousiast who likes to rip DVDs and do other things in the meanwhile can do with 2 cores.
I'm a multitasker who converts audio and video and downloads a lot while intensively browsing the internet. I see no need for me to go more than dualcore. If you are like me; better yet use the money on more happy HD-space, quiet cooling and memory.
If you mod this up, your slashdot background will turn into a beautiful sunset!
It seems to me that AMD is always a step behind in implementing a new generation of SSE instructions and that causes them to lose a lot of benchmarks, especially games. If SSE(x) instructions have to be implemented in an OS, aren't they a type of standard that any chip manufacturer can implement? Or are they created by Intel and only through licensing/agreements available for the likes of AMD to implement?
It is unfortunate, because the only reason Intel killed off the poorly designed Pentium 4 was the strong competition, and the only reason you can buy a Core 2 Duo for under $150 is the competition from AMD. AMD is showing that it costs them more to make worse chips than Intel - since they have no choice, they are selling them at price points that often make them a great value, especially if you don't need the best performance per watt. But AMD isn't making enough profit to do the development they need to catch up. Note to IBM: buy out AMD and set them on the right path before it is too late. There is great potential in their CPU designs, but without more cash and technology to speed up development to compete, they will die a slow death.
Hello! Oka- [gulp, nauseated expression] New teeth. That's weird. So where was I? Oh, that's right -- Barcelona!
"The problem with socialism is eventually you run out of other people's money" - Thatcher.
I can understand that there would be a difference between chips designed for multiprocessing, but if I wanted to have a single-chip workstation, would there be any difference between the Barcelona and the Phenom?
I'm interested in building new workstations for my company, and the Phenom chips look great except that they don't exist, and won't, for at least six months. Why not build Barcelona workstations, though?
Thad Beier
I love Mondays. On a Monday, anything is possible.
I don't think the battle is only on the 45 vs. 65nm arena. There are other interesting technologies in the package that deserve some consideration. Barcelona will include Nested Page Tables (NPT) technology, which could potentially give a significant performance boost to memory intensive applications running on virtual machines once the hypervisors start supporting it.
Intel will also be coming out with a similar technology called Extended Page Tables or EPT, but AFAIK their timeframe is early 2008.
AMD is using SOI, which as I understand, produces some benefit over Intels process at a given process size. If the benefit offsets 65 v 45, or comes close, I don't know.
I do know that AMD sales have persisted in high-performance applications, simply because AMD's memory and IO architecture remain *much* metter than Intel's. Intel with Core 2 finally seriously had a competitor in terms of performance (i.e. very good floating point), but it will be interesting if Barcelona essentially matches the performance in terms of floating point and such, but maintains (or widens) the memory performance, it could be an interesting time for AMD once more.
XML is like violence. If it doesn't solve the problem, use more.
moderators seem to be very harsh today
this should be moderated funny imho (the tomatoes part made it funny)
on another note i hope this chip is killer because it deserves to be !
i saw this chart which iirc eventually meant every core on the amd can be controlled
frequency wise and vcore wise although i'm not sure about the latter
AMD are always behind with SSE instructions because they're developed by Intel, and Intel release the required information to use the instructions (but not necessarily to implement them) far too late for AMD to even consider adding them.
I also tend to find that, until recently, most math-intensive benchmarks seem to be tuned for Intel's Pentium 4 series of CPUs, which have strong SSE performance, but utterly crap scalar floating point performance, and using SSE instructions that only Intel's CPUs had. Compared to AMD's CPUs, which have great scalar floating point, and so-so SSE, to the extent that SSE is sometimes slower than scalar floating point. Obviously not in cases where SSE lets you paralellize things more, but in cases where developers are using SSE to do scalar math.
The problem is that Intel will not implement anything AMD comes up with unless AMD can convince enough software companies to switch their products to the AMD developed standards. For example way back, they created 3dnow! instructions and had minimal success even though there were some obvious performance benefits. Intel never implemented 3dnow and instead went with SSE. It's pretty obvious who won that battle. Now AMD did have a win with AMD64/x86-64, but progress there has been slow and Intel still doesn't want to admit defeat by relabeling the supported instructions as EMT64. If AMD wants to truly win the instructions standards battle, they would have to focus more effort on gaining market share and software partnerships first.
Another thing to note is that while they do typically lag behind one version of SSE, they did implement a number of improvements to SSE instruction performance in barcelona. Hopefully these improvements will shorten the gap on SSE heavy benchmarks (optimized encoding software, games).
The early Barcelonas will be designed for dual socket servers, and they'll be released at reasonably low clock speeds. If you want to make an 8 core workstation that isn't super fast on single threaded tasks, then they'll be a great deal. If you want a single socket system, you'll be spending more money for lower speeds than if you waited for the Phenom processors.
-- The act of censorship is always worse than whatever is being censored. Always.
When have you ever seen a game benchmark changed by the difference between SSE2 support and SSE3 support? From what I've seen, most game developers don't even consider using the new SSE instructions for a couple years - both waiting for AMD to support them and waiting for people to replace the vast majority of older Intel machines that don't support them.
Even when chips do support SSE type instructions, they rarely produce as drastic a performance improvement as the chip manufacturers hype would imply. Writing a program for SIMD parallelism (like SSE) is just as hard as writing it for a multi-core processor, and works for far fewer workloads.
-- The act of censorship is always worse than whatever is being censored. Always.
Stop making excuses for AMD's poor performance and lack of product. it's no good saying "but but amd will have this and this out SOON" that's irrelivant. the only thing that matters is what you can buy on the shelfs right now, and right now intel is far better, so stfu and stop crying.
The only OS support for SSE(x) is saving/restoring the registers while doing a context switch.
Do you even lift?
These aren't the 'roids you're looking for.
http://www.sun.com/emrkt/innercircle/newsletter/04 07feature.html
8 hyperthreading cores running 8 threads each, with each core having 2 ALUs and 1 FPU.
That's 64 concurrent threads, 16 ALUs, and 8 FPUs. And probably only needs a 150- or 200-watt power supply. There's a reason why Sun is getting something like $20K per UltraSPARC T1000 or T2000 rack-mount systems and can't keep up with demand...
AMD and Intel have cross-licensing deals that handle the instructions that each company creates... these deals go way back to the mists of x86 time. So, for AMD to implement SSEn there is no legal problem. Ditto for the reverse. Except, the "problem" is that Intel w/80% of the market can pretty much dictate what instructions will survive in the market -- with the big exception of x86-64, and potentially some of the new virtualization stuff.
Now, about releasing chips in a timely manner... the trick is Intel doesn't have to tell anyone about the new instructions until they are well on their way to being in Intel CPUs. AMD finds out about these things at the same time as software developers get the promotional material from Intel. There's no way for AMD to release chips with these functions at the same time as Intel - they have to wait until the next moderate chip revision.
Does it matter? Usually not. Most software lags instruction changes by years. The exceptions are typically where performance really counts. For example, video encoders picked up on SSE2 pretty quick, since it provided dramatic improvements for their code.
I've written video processing code using SSE, and it makes a substantial speed difference. I haven't looked at the numbers recently, but my recollection is about a 30% reduction in run time. Beyond having to learn more assembly language instructions, writing for SIMD is not terribly difficult. Writing for a multicore processor requires learning about threads and paying attention to data timing, which I find quite difficult.
Contribute to civilization: ari.aynrand.org/donate
Yea, video processing code is the poster child for SIMD - a 30% runtime improvement there over no SIMD is quite reasonable. On the other hand, the difference between SSE2 and SSE3 for the same code is probably somewhat smaller. In other areas, SIMD doesn't help at all, or only helps if you use techniques that are much more complex than multi-threading code.
Another interesting development is GPUs as general purpose SIMD processors...
-- The act of censorship is always worse than whatever is being censored. Always.
There is a diffrence. AMD has to pay to license Intel patents...
http://news.com.com/2100-1040-257059.html
http://news.zdnet.co.uk/hardware/0,1000000091,391
So everything AMD invents can be adopted by Intel for free. But AMD has to pay license fees to Intel. Pretty indefensible really, since Intel has a much larger market share, and much higher proft margins.
echo -e 'global _start\n _start:\n mov eax, 2\n int 80h\n jmp _start' > a.asm; nasm a.asm -f elf; ld a.o -o a;
Another interesting development is GPUs as general purpose SIMD processors...
Yes but your problem there is fetching data back from the GPU (if that's not where it's going to end up). Most GPU benchmarks use artificial datasets to make their case. When you put real data in there -- data that has to be loaded and returned to main ram -- GPUs are often much much less efficient to optimize for than SSE.
"AMD does get to collect royalties from Intel for any patents Intel might adopt."
"So everything AMD invents can be adopted by Intel for free."
??
I think that's a misprint. It should be "AMD doesn't get to collect royalties from Intel for any patents Intel might adopt."
echo -e 'global _start\n _start:\n mov eax, 2\n int 80h\n jmp _start' > a.asm; nasm a.asm -f elf; ld a.o -o a;
No one has voiced it yet, but AMD's 65nm process is a failure. It's 65nm parts overclock worse than processors at 90nm process and that's probably why AMD are still producing all there high end parts at 90nm
This is exactly right. One of my best friends is an engineering manager at Intel, and he's been there for 9 years now, and he survived the 1000 manager cut recently. A couple of years ago, when AMD was kicking Intel's butt, he said "just wait about 3 years". I thought he was just blowing smoke, but he was right. He recently told me that when AMD burst onto the scene with the Athlon and started trouncing Intel, Intel took it very seriously. There were many meetings about what to do, and they had people do reasearch.
He said they were in a large meeting with some higher-ups, and he saw a presentation that outlined the move to 45nm. He said after that meeting, nobody there was worried at all about AMD because they knew it was a matter of time. They knew AMD didn't have the financial or research resources that Intel had, and THAT is what it takes to get ahead. So they bided their time, did what they needed to do, and trounced them.
He said there was also talk about how brash AMD was, and how the execs would drive their flashy cars around, and be just blowing money right and left. The Intel execs were more subdued because it wasn't new money to them. Tortoise and the hare. There was some back and forth, but it will be a LOT harder for AMD to compete now than it was for them to come up. They woke Intel up, and they aren't stupid. My friend acknowledges, and said everyone at Intel does as well, that AMD coming onto the scene was a great thing for Intel. It made them get off their lazy asses. Not to mention what it has done for the computing industry.
My beliefs do not require that you agree with them.
Well, SSE matters much in HPC code. Intel compiler has quite decent vectorizer which helps a lot. And writing vectorizeable code is not that hard, actually.