Computex 2003 - Day 4: XGI, Motherboards, and cheap Itaniums
by Evan Lieb & Andrew Ku on September 26, 2003 7:54 PM EST- Posted in
- IT Computing
More than anything else, Computex 2003 has been about fierce competition. The main competitive focuses at this year’s Computex were the launches of high-end desktop processors from AMD and Intel; AMD’s Athlon 64 FX and Intel’s Pentium 4 Extreme Edition. You’ve no doubt seen dozens of CPU reviews across the Internet comparing and contrasting these two high-end microprocessors in an attempt to help readers make informed decisions for their computing needs. Hardware review web sites eat up anything Intel and AMD nowadays, and so this type of wide coverage is not at all uncommon.
A little background on the current CPU wars takes us back to 1999, when AMD launched their K7 (Athlon) architecture that sparked this “arms race” of sorts. Since then, both AMD and Intel have had more or less equal shares of the performance lead, with Intel taking the lead at the high-end the last 10 months or so, and AMD securing the low and middle-end channels during that time. A couple days ago, we saw AMD take back that performance lead from Intel. Now, the only question that remains is whether AMD can supply enough CPUs to channels and OEMs to meet demand. Given the large die size and relatively low yields of current Athlon 64 iterations over at Fab 30 in Dresden, Germany, AMD will no doubt have their hands full. Various manufacturers tell us that AMD is confident that they can produce 400,000 Athlon 64 processors by the end of this year. While certainly a step in the right direction, this quantity is barely enough to feed high-end customers and enthusiasts, and no where near enough to supply the mainstream markets.
The competition in the desktop video card world is equally fierce, as ATI and NVIDIA have been biting and gnawing at each other for years now. This graphics competition really started to brew when ATI introduced the Radeon 8500 in the fall of 2001. After the GeForce4 came along just a few months later, however, NVIDIA pretty much held a dominant lead in every segment of the add-in video card desktop market up until September 2002, when ATI’s Radeon 9700 Pro become readily available. ATI was easily the high-end desktop GPU leader after that, and right up until GeForce FX 5900 Ultras became available in late June of this year. It was a dead heat between NVIDIA and ATI after the GeForce FX 5900 Ultra introduction, but that ended a few weeks ago when the world learned of NVIDIA’s serious DX9 shortcomings in titles such as Half Life 2. This seemed to signal the beginning of the end of NVIDIA’s steady market share lead over the last year. However, just a few days ago it was announced by Valve that Half Life 2 was going to miss the September 30th target. In a way, ATI faces the exact opposite scenario that AMD is in; ATI can supply enough DX9 (9600 and 9800) video cards to market, but may not see an upsurge in those sales (mostly in the retail channel) in the near future because of Half Life 2’s delayed introduction. Of course, there are other legitimate DX9 titles coming soon that will spurn sales of ATI’s DX9 cards, and there even may be a great deal of users that figure they can simply buy ATI cards now and wait for Half Life 2 (which we suggest you do if you’re buying now). Either way, competition is a great boon for any industry, and we certainly hope it never stops.
Anyway, read on as we discuss the latest happenings in our Day 4 coverage of Computex….
Facebook
Twitter
RSS
21 Comments
View All Comments
Maverick215 - Monday, September 29, 2003 - link
wonder if the XGI guys got any pointers from the Bitboys :)Anonymous User - Monday, September 29, 2003 - link
Don't wait for HL2, the interest rates are excellent. The economy has allowed all investors to become "flexible" in their search for graphics infinity. The competition in the desktop video card world is not equally fierce, HL2 has disallowed it. ATI will be the eventual standard of every process required, either it be Microsoft's next console or Doom3.Anonymous User - Monday, September 29, 2003 - link
Nah, spurn was a typo for sperm, seeing as the article was a load of w**k. ;)Anonymous User - Monday, September 29, 2003 - link
I'm excited about the ATI IGP9000 myself.I've sold quite a few nF2 IGP systems thanks to the full integration combined with adequate on-board graphics for most my clients needs.The IGP9000 will be a great solution for me to base builds on if the pricing is close enough to the nF2 G4MX boards.
BTW, here is a small typo "legitimate DX9 titles coming soon that will spurn sales of ATI’s DX9 cards" it should state "spur" not spurn :-)
Anonymous User - Sunday, September 28, 2003 - link
I'm afraid that the authors blew their credibility in the first couple of paragraphs."By the way, we’re not being sarcastic when we call these processors “low-priced”; the cost of entry for 64-bit processing has really never been this low for such a legitimate 64-bit architecture."
Say, what? Ever heard of Sun, and SPARC? Sun has been selling $1,000 entry-level SPARC *workstations* (not just bare processors) for a couple of years now.
And Opteron processors sell for as little as $229, last time I checked: just what about them shouldn't be considered 'legitimate'?
(And of course there's MIPS, if you want *really* low-cost 64-bit processing with an architecture far better established than Itanic's may *ever* be.)
"I2’s 32-bit performance is no where near as fast as Opteron series processors, but depending on the 64-bit application, I2 is much faster, and could be much faster in the future when more applications are developed specifically for IA-64."
Puh-leaze! Model 140 and 240 Opterons at 1/3 the price of the bottom-of-the-line $744 Deerfield are considerably more than 'noticeably cheaper', handily beat Deerfield in SPECint (880/933 to an estimated 750: split the difference between the 810 that a 1 GHz McKinley achieved with 3 MB of cache and the 674 that a 900 MHz McKinley achieved with 1.5 MB of cache), and aren't even *that* far behind in SPECfp (934/1012 vs. a similarly-estimated 1290, not that SPECfp is of all that much significance for most commercially-significant processing). Or, if you'd prefer to compare Deerfield against a similarly-priced Opteron, you see SPECint scores of 1095/1170 - i.e., *well* over 50% faster than Deerfield - and SPECfp of 1122/1219.
And that's just Opteron's 32-bit performance, compared against estimated Deerfield SPEC scores obtainable only with HP's HP-UX compiler using feedback-directed optimization. In the real world, using commonly-used compilers (and techniques) and taking advantage of its 64-bit code extensions, Opteron's lead will only increase.
Deerfield's lower power (at least relative to previous Itanics) will allow FP-intensive applications to start taking advantage of 1U rack-mount units to increase processing density, but that's about it. It's price isn't even *that* dramatically lower than the previous bottom-of-the-line Itanic that's been available for over a year at about $1300.
- bill
Anonymous User - Sunday, September 28, 2003 - link
"simply buy ATI cards now and wait for Half Life 2 (which we suggest you do if you’re buying now)."Hi, I don't understand this comment. Won't the prices of the video cards go down!?
I was planning on waiting as long as I can to get a video card before HL2 comes out, is this a bad idea?
-CalicoRabbit
Anonymous User - Sunday, September 28, 2003 - link
I have to question the faster in the future statement. There are only 20,000 chips out there last time I checked (and thats in 2yrs or so of selling them). Opterons will have already sold 80,000+ (intels recent goal with Itanium supposedly by end of year I think) in a few months. Why would you write software for 20,000 chips when an opteron has much more volume for you to sell to. Not to mention it usually takes 1 guy a week to optimize for opteron, while itanuim could take a whole TEAM ages to compile code for. Most have already figured this out. If you can't recoup the costs on software engineering (too few to sell it to) then why write for it?If you don't have a SPECIFIC application that is already optimized for itanium don't count on one being created. The ONLY chance you have of getting one, is if you can find an app where performance on other chips (64bit chips) are completely blown away by an itanium (which, I can't think of many). Opteron put another nail in itaniums coffin, and Intel will probably seal that coffin up as soon as they offically inform us that they're adding 64bit to the desktop (prescott). Buying Itanium/2 today is completely retarded unless you already KNOW you can buy an app that is what you need NOW. No 32bit killed this chip (well, opteron did...LOL). When I say no 32bit, I mean a Pentium100 performance doesn't count as 32bit. If its for all intents and purposes, USELESS, it doesn't count. They still have to prove its "faster" in anything yet. There isn't enough out there to say that if you ask me.
With the pricing and # of chips it takes to get that performance, you can buy a whole cluster of opterons to catch them, without the "what can I actually run on this expensive piece of crap" question. It would have been a great chip in 5 years if opteron didn't exist. Now its just dead. 20,000 chips sold in OVER 2.5 years (debut may 2001)? The industry obviously hates this chip. Most of those 20000 are in the hands of NON-customers. Testers etc. NOT users.
AgaBooga - Saturday, September 27, 2003 - link
#8, you have to remember that the Itanium does not have a "hardware" method of running 32bit apps like the Opteron does, and so it is very slow. In fact, I dought you can easily find some place using 32bit on an Itanium, it would be comparable to running it on a pentium 90mhz or so.Anonymous User - Saturday, September 27, 2003 - link
#13 Andrew Ku
I know that itanium2 are $5,000, and the new Itaniums are cheaper.
In my #8 post i just disagreed that Itanium2 "...is much faster, and could be much faster in the future..". I just asked for the benchmarks for the 1.0Ghz and 1.4Ghz, 1.5Mb L3 Itanium2. Then #9 said the diference in L3 cache did not make a diference in 64-bit aplications.
My #12 post is about the diference in L3 cache.
You just can't show a price for an Itanium2 (1.4GHz , 1.5Mb L3 cache) and then talk about performance of another totally diferent Itanium2 (1.5Ghz, 6Mb L3 cache) and say Itanium2 "...is much faster, and could be much faster in the future..". There are NO benchmarks for 1.4Ghz, 1.5Mb L3 cache at SPEC.org. Does Anyone know where can we find them? Or another type of benchmark?
Andrew Ku - Saturday, September 27, 2003 - link
#12 Keep in mind that previous Itanium2s ran just shy of 5,000 USD. The new Itanium2s are definately considered cheap in that respect.