Slashdot Mirror
Intel's New Architecture Too Late?
rts008 writes to tell us that TG Daily has an interesting interview with Randy Allen, AMD's vice president of the server products division, about (among other things) AMD's recent stellar fourth quarter numbers. From the article: "Responsible for that shrinking lead is especially AMD's server products group. Intel's CEO Paul Otellini recently acknowledged that Intel had to give up market shares to AMD and will likely be forced to hand over more shares until the next generation of server chips arrives. [...] AMD's Randy Allen explains in this conversation with TG Daily why he believes that Intel will need much more than a new processor to be able to slow AMD's growth."
Another perspective here:
1 3.html
http://www.thestreet.com/_yahoo/tech/semis/102631
I heard from someone that the Pentium-M is better than any of AMD's offerings for mobile CPUs; is there any truth to this?
No its incorrect. the AMD mobile CPUs still outperform pentium M and their offspring currently the core duo and core solo chips. So the AMD ones are still out in front.
The only difference is the AMD laptop chips use more power, but you can always plug in somewhere and recharge so really this is no disadvantage.
I know that the Pentium M wins in battery life. Additionally, if it means anything, January's issue of Maximum PC compared an AMD Turion 64 MT-28 with an Intel Pentium M 740 using some benchmark tests, and concluded that the Pentium M was better overall.
I don't mean the current Core-branded Yonah processors, but rather the forthcoming Conroe and the cores to follow. When it ships it will beat the X2 clock-for-clock and consume less power, and suddenly the things that have been novel about the K8 won't be. Instead of AMD having a clear leg up on Intel (the Pentium D is rather embarrassing at this point), AMD will have to compete on price with Intel having a clear advantage in process and volume.
You don't say in what way, so I will. IIRC, The Pentium M is ever-so slightly better in integer ops, but it gets creamed in floating point. Yes, this is fixable.
No
You are giving a very incomplete answer. AMD has a line of very low voltage K8 chips called Turion (yes, the name's questionable, but that's nothing new when it comes to brands in the processor world). The most efficient run at about 25W, but unlike with the P-M, this figure includes AMD's on-die memory controller, while Intel defers that extra power cost onto the board logic. Also, one of them (I can't remember which) reports wattage at peak value, and the other at typical value.
The only way you are at all correct is that Turion adoption was slow, because IIRC laptop manufacturers, for whatever reason, do their redesigns at the beginning of the year, and they missed this opportunity last year (or the year before, whichever) because they couldn't introduce the Turion in time.
Roughly, Turion laptops get 3:30 to a comparable Pentium M laptop's 4:00.
Uh, Sun have a 8-core cpu for sale right now. See their t1000/t2000 servers.
Face it fanboys, a laptop with an AMD chip is a piece of stinking shit.
Except when it comes to FLOPS, cache size, i/o speed....?
That said, anyone who buys a laptop and doesn't play games on it is better off sticking with Intel. There's no better chip when it comes to running the usual suspect win32 apps. But your 3D games will suffer.
May the Maths Be with you!
> The only difference is the AMD laptop chips use more power ...
No they don't. They have a higher quoted TDP, but that number is an engineering choice, AMD typically quotes a higher TDP for the same power consumption. AMD is crrently shipping two versions, one of which has a TDP of 25W that under typical use comes in at about 17W (slightly better than the equivalent Intel part even though the AMD part has its memory controller counted in that while the Intel part doesn't). Practically this means no difference to battery life given that the rest of the system is the same (except the northbridge, of course). And you get better performance with a Turion 64 processor for everything except video transcoding. When we see 64 bit optimised encoders, I expect that to change too. It is pissing me off how long it is taking to get hand-optimised AMD64 routines for tight inner loops in various common algorithms.
I do remember, Intel lists only 75% of the TDP while AMD tells the truth and lists the peak value.
Quote taken from: http://www.businessweek.com/the_thread/techbeat/a
I particular getting dual core onto the desktop first isn't going to be as big a coup as AMD thinks. Business generally doesn't seem to want dual core desktops yet and is certainly unwilling to pay a price premium for it.
Hmm...it was actually Intel who was the first to put dual-core on the desktops. Intel's dual-core strategy debuted with their Pentium 4 Extreme Edition line of PCs. AMD's first dual-core systems were in the server lines, which made a lot more sense. It's also where AMD has seen a huge growth in marketshare, and Intel has seen their numbers decline. This is because AMD is still offerring more computing power at a lower price point than Intel, and (if Sun and many others are to be believed) at a lower power consumption.
And for the record, there are plenty of businesses that are wanting dual core desktops and are willing to pay the (very slight) price premium for them. I just set our purchasing specs for the next year and all of the desktops will be dual core.
AMD reports peak values, Intel reports average values. Take an Athlon64 4400+ rated at 110W, and put it next to a P4 rated at 130W. Compare the total system power draw. There is no *way* the P4 uses only 20 more watts. 130W-rated P4s have been tested as drawing 170W+ during peak loads.
A deep unwavering belief is a sure sign you're missing something...
Generally you are correct, 64-bit code is generally slower as it requires more cache & bandwidth than 32-bit code. The only real advantage is in encryption, databases, etc which requires the large word widths where it can provide benefits (beyond more memory available).
Of course, we are specfically talking about x86 where the above isn't true.
In normal 32/64-bit modes (e.g. SPARC, POWER, etc) the modes are pretty much the same. For x86 it isn't, you get double the amount of registers (8 vs 16) which can provide LARGE performance benefits for a given task. Also, you can built your x86-64 binary and assume you've got SSE2, MMX, etc and any other instructions that have been added since the i386 came about. Basically, you can build a better binary.
Wikipedia is useful as always.