Slashdot Mirror
Transmeta Astro Processor
simpl3x writes "Apparently, Transmeta's next generation processor was demonstrated to some folks the other day at Comdex. Tom's Hardware was at the demo and they had this to say: "The new Transmeta Astro was faster in every demo that we saw than the Pentium 4m 1.8GHz chip that was in the Sony GRX." Cnet had some information on the processor also . I just ordered a tablet to play with, though I ordered the Fujitsu which has a P3m (the Compaq has a bad screen according to the reviews). I certainly wish that something like this were available, and i do hope that the manufacturing goes smoothly. Mo options, mo better."
I was fortunate enough to see this cpu in action at a press demo and I must say I was sadly dissapointed...
The lack of sse2 support greatly hindered this chip in any fps demo, where it was brutalized by the p4 (I'm sure even an amd athlon could beat it under those conditions!).
The 'code morphing' technology also uses an astonishing amount of ram, up to 64mb in some cases, so linux users who need all that ram for gnome should steer clear of this chip. I also noticed that compared to a p4 based system, it was quite unstable, requiring a reboot in windows98se after just 2 hours of demonstrations. I have also heard, from reliable sources, that boards using this chip can only run at agp 2x, which again can hinder game performance.
For power desktop use forget about using this chip, although I'm sure for student or 'dumb terminal' use this chip is suitable.
Stanley Feinbaum, professional journalist and master debater! God bless the USA!
From the CNET article "It will consume less power than the company's first Crusoe chips, the TM 5000 series, but offer substantially more performance, said Chief Technical Officer David Ditzel."
Wow... and according to tramsmetazone the thing was running at 500 mhz for the demo (against a speedstepped pentium) WOW.
According to the CNet article (which actually had more information in it), the Astro will have lower power consumption than their current and competing chips. There is also mention that this is to compete directly with Intel's mobile chip. So I guess the answer is, that yes, they will be doing laptops. Of course, I thought that was what they were aiming for all along.
The Cnet article said the chip would be lower than Intel's offering. The big thing was that they doubled the intructions per clock cycle, while keeping the lower power consumption that makes it desirable for portables (Tablet & Notebooks).
"Give a woman two glasses of wine and some pad thai, and they'll agree to just about anything." the Sports Guy
Again, as mentioned in the CNet article, "Transmeta executives, though, indicated that the company would beat Banias in price." The Banias is Intel's soon-to-be-released next mobile chip. It should be at least fifty dollars less than Banias, but twenty dollars more than Celeron.
the astro does have an integrated chipset. Transmeta is not a big enough company to have many different chipset models developed so they ussually go with the cheapest chipset model with the lowest power consumption.
agp 2x was chosen for power reasons, as 4x would have pushed the (relatively underpowered) chipset and cpu too far.
Stanley Feinbaum, professional journalist and master debater! God bless the USA!
Are you a troll? ...lack of sse2... ...uses an astonishing amount of ram...64mb... ...unstable... ...only run at agp 2x...
It doesn't even run X86 natively!
64mb or ram costs 15$ The price difference between the P4 and the transmeta will easily be more than that. Buy more ram!
It hasn't even been released. Kernel 2.5 isn't all that stable, but no one complains because it is a testing/prototype.
The speed of the agp bus has been shown to be inconsequential to the performance.
The rumor is that the demo chip is running at 500Mhz at the moment. Comparing that to the 1.8ghz P4 suddenly doesn't seem so out of proportion does it? I gaurantee you it will be running at at least 1ghz when it's finally released. The final board for it (not the notoriously shoddy reference boards) will perform better as the memory bandwidth will probably be improved.
What if I had done the same review of the Itanium 6 mo. before it was released? It was running at 400Mhz, couldn't run X86 software as fast as a 266, and was practically an unstable toaster oven.
Karma Clown
SSE2 is used by so few apps that its not very useful for 99% of the stuff people are doing. As far as agp support only being 2x, thats the motherboard's (well northbridge's) responsilibility, not cpu. And besides, agp is a joke, anything more then 2x gives you ZERO performance increase.
Lawyers, MBA's, RIAA? A jedi fears not these things!
For one, there simply wasn't much of a market in emulating non-x86 architectures. The available software and knowledge base, particularly for embedded design, is much higher in x86. For the systems that must have non-x86 architecture, generally they'd just get the real system rather than an emulation "hack", as often these other architectures are chosen to get the maximum reliability or performance characteristics that those markets require.
Also, because CISC breaks down to multiple smaller operations easier than RISC, CISC is much easier to get performance advantages with code morphing.
I was lucky enough to meet Linus at the last Bay Area Linux User Group (LUG) picnic, and he was still at tranmeta. He hinted at a new product that would use low power as the big selling point. I wouldn't be too worried about good linux support in the chip with Linux around Transmeta.
The article (reading it would be good) mentions DVD playback and other typical benchmark-y things.
Okay, I'm not sure what you heard about the Cyrix 6x86, but here's how it works. Intel has more than once released the specifications for various of their instructions so that compiler writers, assembly programmers, etc could use their CPUs. Even SSE2 had to be documented *somewhere* for them to be used. So, there shouldn't be any reverse-engineering needed. Even if there was, a 100 or 200 instruction system is hardly going to be anywhere near the magnitude of hundreds of api functions, with every so many weeks meaning yet another whole slew of things to add. You can't modify hardware nearly that quickly to thwart off cloning. The design costs alone would run the company into the ground. So, the bases for the majority of the instruction set (up to 686 and probably most of the Pentium 4) is well known. No problem.
:) So, it's far from a minor issue.
As for power consumption, the three major power consumption parts of a computer are the monitor, the cpu, and memory (the constant refreshing). Laptops already use an LCD screen to avoid the huge monitor hit. Reducing CPU power *is* a big deal as well. Switching to a memory type that didn't require constantly refreshing would be a big save too, but all known memory of that type is horridly slow (cmos, flash memory, etc), so you can't fix that real well. So, focus on the CPU is a huge deal. Especially when all excessive is wasted energy, and then a fan on top of that is even more.. Not to mention having a laptop that burns your lap.
Consequently, it multi-processor transmeta systems will outperform single processor Intels dissipating the same amount of heat. This also translates to higher reliability. If the memory busses are done correctly, having inexpensive multi-processors may alos provide significant performance enhancements over a single CPU. (for example, if memory bottlenecks dominate then multiple simple processors that are stalled witing on memory will ustilize every memeory fetch perfectly, whereas a pipelined single processor will waste a large fraction of the memory fetches making it slower).
A schematic of the current trends look something like this.
.......ioo......
|...........i.t..
|..........i.t..
|..........it...
H.........it....
E........it.....
A........i......
T.......i......
|.....io..o.....
|....io.........
|___i____________
Speed--->
o = Transmeta
i = pentium
t = former trendline
Some drink at the fountain of knowledge. Others just gargle.
Those computers with palladium will all have a way to turn the thing off. When turned off they would be just like any non-palladium machine and just like other non-palladium machines they won't be able to view palladium content without the hacker intervention.
Hmmm... Pie...
Lack of SSE2 is a bummer but unless you're doing content creation or playing new games it won't matter.
Lack of SSE2 will not matter for any games, new or otherwise, because none of them use double-precision floats in any speed critical code paths. It'll hurt you if you want to render with Lightwave on your laptop, or run fluid dynamics simulations, or whatever. Not games.
Will this new processor let me have my laptop on my lap without burning my penis like this guy did
09 F9 11 02 9D 74 E3 5B D8 41 56 C5 63 56 88 C0
Here educate yourself:
6
/. moderation is so uninformed.
http://murl.microsoft.com/LectureDetails.asp?59
Note the date.
some more [in german] links for the lazy ones.
Ok, I havn't read the linked articles. But I really think that you are wrong about Transmeta abandongin the code morphing. The things I've read in the past about their next generation cpus, which I assume is the astro, is that the maximum theoretical throughput would be doubled from 4 ops to 8 ops. That would ofcourse be impossible to do, without going the VLIW + codemorphing path of previous generations.
From what I recall reading the CPU has hardware MMU and exceptions that follow the x86 design. The processor is designed only to execute different native instructions.
So I doubt it could be easily used to emulate a PPC.
Tom
Someday, I'll have a real sig.
I don't like the whole "desktop replacement" idea. It should come as no surprise that sales of such units have been accelerating, because that's the only thing being offered by the big name brand retailers.
Subnotebooks have yet to be fully and properly put on the market in North America. My local superstores don't carry anything like them.
Sony's picture book is, well, a toy. It's a nice toy, but it's not a simple subnotebook.
The Fujitsu Lifebook was intriguing and had good press, but the delay in delivery of the Crusoe 5800 really hurt--they were simply unavailable in North America last Winter. People were on 3 month waiting lists for crying out loud.
How much more evidence of a demand for subnotebooks will it take to convince the major retailers to put these things on shelves? No gimmicks. No fancy new designs. Just a small notebook computer with long battery life. And for heaven's sake don't clip the battery in half to cut down on the weight. The 4 or 5 hour battery life of the Crusoe is an honest-to-God selling point.
Switching to a memory type that didn't require constantly refreshing would be a big save too, but all known memory of that type is horridly slow (cmos, flash memory, etc), so you can't fix that real well.
It's called SRAM, and it's wicked fast. You may be familiar with it since it's beed used in caches for decades. The S stands for Static, in constrast from Dynamic styles of logic (i.e. DRAM which includes SDRAM, DDR, etc.)
The reason why we can't use it for main memory is cost. DRAM is easy to integrate into extremely dense layouts, it's basically one transistor per bit in a grid arrangement. SRAM usues about 6 transistors per bit and is not nearly as easy to arrange into nice regular patterns like DRAM.
SRAM requires more power than DRAM. When the DRAM is not being refreshed, the charge is held in a capacitor next to that one transistor to store a bit, this is why it needs to be refreshed, the charge will leak out of the capacitor. SRAM has power constantly flowing through its transistors, it doesn't need refreshing because the data is encoded in which pathway the power is flowing through.
Flash memory and EEPROM would take less power because they retain data when power is cut, but it has a shorter lifespan and is much slower. Magnetic core memory would also take less power for the same reason.