Derek and I both worked on the Pentium 4 6xx article that went up today, so you know it's just got that extra bit of love that you all come here for :)
Overall I was relatively disappointed with the Pentium 4 600 series (and definitely in the new Extreme Edition processor). A larger, higher latency L2 cache is not the way to go - especially when looking at microprocessors from a performance per transistor added perspective. I'll be talking a lot about performance per transistor as a metric in my Cell article, which got put on hold while I worked on the 600 series article, but I will resume my focus on that today.
Looking at the amount of additional die area that the extra 1MB of cache consumes and compare that to the less than 10% of the die that AMD's on-die memory controller occupies and it's quickly obvious which offers the greater performance benefit per transistor.
It's very clear that, internally, Intel is facing a struggle between fighting for performance and fighting for dollars. Because while each revision or clock bump to Prescott does nothing to outclass or outperform AMD, Intel continues to do extremely well business wise. Convincing management to look at the AMD threat more seriously is difficult when what you're doing is making money, a lot more than AMD is making. I don't get the impression that Intel is going to be any more attractive even by the end of this year, although it does look like they will be first to dual core desktops, which is a sign of taking the AMD threat a little more seriously.
Overall I was relatively disappointed with the Pentium 4 600 series (and definitely in the new Extreme Edition processor). A larger, higher latency L2 cache is not the way to go - especially when looking at microprocessors from a performance per transistor added perspective. I'll be talking a lot about performance per transistor as a metric in my Cell article, which got put on hold while I worked on the 600 series article, but I will resume my focus on that today.
Looking at the amount of additional die area that the extra 1MB of cache consumes and compare that to the less than 10% of the die that AMD's on-die memory controller occupies and it's quickly obvious which offers the greater performance benefit per transistor.
It's very clear that, internally, Intel is facing a struggle between fighting for performance and fighting for dollars. Because while each revision or clock bump to Prescott does nothing to outclass or outperform AMD, Intel continues to do extremely well business wise. Convincing management to look at the AMD threat more seriously is difficult when what you're doing is making money, a lot more than AMD is making. I don't get the impression that Intel is going to be any more attractive even by the end of this year, although it does look like they will be first to dual core desktops, which is a sign of taking the AMD threat a little more seriously.
Facebook
Twitter
RSS
22 Comments
View All Comments
Viditor - Tuesday, March 1, 2005 - link
Andy - "They have such huge R&D resources I find it difficult to understand how significantly smaller chip companies can keep beating them"You have to have the perspective of time to understand that...
Jerry Sanders III (founder of AMD) put it best...
"Playing in the semiconductor industry is like playing Russian roulette with a twist: You put a gun to your head and pull the trigger, and four years later you find out if you blew your brains out."
You can have all the R&D possible, but it still takes many years to implement that into manufacturing...and by the time you do the whole landscape has changed once again!
Remember that Intel was designing Prescott before x86-64 had ever been heard of. Changing their design to incorporate 64 bit was a HUGE ask (much harder than designing it from scratch).
Also, prior to the release of the Athlon (which is still very recent by semiconductor standards), Intel had never had any real competition...
fuluku - Saturday, February 26, 2005 - link
http:\\www.highspeedsat.com\globalstar_satellite_phone_rental.htmThe_Necromancer - Thursday, February 24, 2005 - link
I have a question off topic, when will the non-micro-atx form factor mobos, of the ati bullhead aka xpress-200 come out, and how much will they cost???Chris - Thursday, February 24, 2005 - link
The problem is when certain tasks are linear and sequential in nature. For example you can't begin rendering a scene until the physics and AI have been processed, so there is very little opportunity to use concurrency. The traditional solution is to pipeline such that one core is working on the physics for the next frame while the previous frame is being drawn, but the corresponding AI for that next frame may depend on physics, so you cant run physics and AI concurrently, rather you have to finish one as quickly as possible, etc. You can only pipeline so far ahead before certain results must be completed in order and in finite time.In linear and sequential tasks which do not translate well to multiple concurrent tasks will see poor performance on multi-core chips whose clock speed is lower than a single chip :(
Games fit into this category of being very linear in nature since the composition of the next frame depends on everything from the previous frame. The exception is limited to a few cases of threading for the sake of threading (to brag that multithreading as a feature), asynchronous file loading, network operations, etc. It can be used, but due to the non-concurrent nature of a game engine, it only serves to convolute the code.
Stephen Brooks - Thursday, February 24, 2005 - link
There are a few GHz-enhancing (i.e. leakage-reducing) technologies around like tri-gate and metal gate. The last I saw these were an option for the 45nm process at the earliest, which is probably a 2008/9 thing (Intel say 2007 but they've been a year late with 90nm and it looks 65nm now too in terms of shipping).This is probably known to most people reading this, but the Prescott core would have rocked _in a universe where leakage didn't exist_ and would have hit 5GHz by the sounds of it. They seemed to plan the NetBurst idea back around 2000 when leakage was less of a problem.
So now Intel are on the defensive. Marketing and spin is everywhere, but I'd guess they only have to become _really_ scared once AMD's new fab is up and running. That ought to be pretty interesting.
JK - Thursday, February 24, 2005 - link
I'm more excited about Intel finally adopted AMD64 in a full line of their desktop processors. Sure, 6xx doesn't make huge gains where many would like, but the most critical thing about it seems to have been missed.And the upcoming dual core processors (the 8xx) will also be AMD64 capable.
Now bring on the 64-bit Doom 3 (or Quake 4!) with a 512MB video card. :P
Jim - Wednesday, February 23, 2005 - link
#12> Because all their money is going into Marketing rather than R&D. And judging by their success at selling crap just proves that the only thing that matters is marketing.f - Wednesday, February 23, 2005 - link
fThe_Necromancer - Wednesday, February 23, 2005 - link
Ahh, well Intel is going down hill, The death to Craig, the Death to Bill, THE Death to Nvidia, MuahahahahhahaAndy Bellenie - Tuesday, February 22, 2005 - link
...well, beating them in performance, anyway.