RDM,
Re: K7
Put me down for 767Mhz for end of year.
Also from fullon3d a few tidits for the AMD faithfull.
The FPU is 25% faster than PIII on a clock by clock basis.
cache is 1/2 speed.
Perhaps some call buying is in order on Monday morning methinks:
- Thursday, May 27th -
K7... Again
2:45pm MST - Krister "fyodor" Nielsen
Ok, that does it! Some much silly, ridiculous, preposterous and just plain stupid 'information' (if you can call it that) is floating around (and surfacing) concerning the K7. Let me first revisit The FiringSquad's K7 preview and the follow-up (which you can check out here). Their main point in the whole follow-up can be summed up as (paraphrased): 'The K7 won't run PII optimized software very well' (I'm talking floating point stuff here). This is just plain rubbish! Clearly, FS don't know what they are talking about. PIII FPU optimization mainly consists of pairing FADDs and FMULs, since the PIII FPU has two pipelines with one handling adding and the other handling multiplying. The K7 has three FPU pipelines, with one handling additing, one handling multiplying and the last one doing FSTORE/FLOAD operations. Thus, code that has been optimized for the PIII will also run very well on the K7. That's why they designed the K7 like that. AMD's engineers aren't stupid; they know that they have to make a CPU that works very, very well on existing code. And existing code is PII optimized.
There was also an comment in the FS follow-up from a reader, which FS believe explains some of the poor performance: That the third pipeline of the K7's FPU can't/won't be used at all without recompiling. This is just plain wrong. If the code contains instructions that the third pipeline handles, the third pipeline will be used. If the code doesn't contain any of these instructions, it won't be used (duh!).
Also, there seems to be a general sense of criticism (in general) that the K7 motherboard would require heatsinks (those white things I mentioned below), whereas Intel's does not. Well, let me just say that there are some pretty significant differences between Intel's and AMD's motherboards. A K7 motherboard has components that run at 200 and 400MHz. Nothing on an Intel motherboard runs at over 100MHz. All other things being equal (which they are not), these components would dissipate 2 and 4 times as much heat, respectively. There are some substantial architectual improvements in the K7 motherboard's over a PIII motherboard. Sure, they most certainly do run hotter, but they also offer some serious advantages (at least some of which Intel will address with their as-yet-unreleased 'Camino' chipset).
I would also like to point out (again), that the FS preview occured about two months ago. At that time (see Bob's comments below) only very few of the K7s out there (with OEMs) were 'real' K7s (whatever that means).
I know I'm grinding on now, but I really would like to reitterate that benchmarking is incredibly difficult. It seems really, really simple, but in truth it isn't. I could easily make some seemingly insignificant changes to the BIOS settings (like setting them to 'default'!), that would easily cost 30% in many benchmarks (like Quake2 or FPUMark). Add to that the whole (software) driver issue and you can see that benchmarking isn't easy at all. [Note: absolute benchmarking is incredibly difficult, relative benchmarking is much, much easier! I.e., it's pretty easy to see if a driver change makes your benchmark score higher or lower, but determing exactly what the (absolute) benchmark value is... well, that's not easy at all. Hell, there are companies who do nothing but benchmark! That's their sole reason for existing.]
Being amazing at Quake doesn't make you good at benchmarking (unfortunately :)). Chris over at The Upgrade Center might not be any good at Quake (he might be, I don't honestly know), but he has proven that he can produce some good, serious, credible benchmarks. Luckily for us, he got his hands on a final release sample of a K7. This is what he had to say:
Speed grade of 550mhz with 1/2 speed cache
Booted at 650, but was not stable with cache enabled
About 25% faster at Lightwave than same speed P3
FSB setting of 100-250mhz
2x-8x multipliers
For those that don't know, Lightwave is THE 3D animation program, and as such is very FPU intensive. It is, of course, PII optimized and not K7 optimized. 25% faster than a PIII is pretty damn quick, if you ask me. In fact, I'd say that classifies as 'wiping the floor with'.
Ok, I'll shut up now - this has got to be the longest post I've ever done :/
- Wednesday, May 26th -
IRC Chat
6:50pm MST - Krister "fyodor" Nielsen
Last night I joined in Ace's Hardware's anniversary chat. It was actually very good - at least the couple of hours I spent there. At one point an AMD fellow dropped by. He was, of course, bombarded with questions regarding the K7. Naturally he couldn't say much, but a few things of interest did surface:
hispeed bob, how hard is it for an individual to get access to pre-release products for testing??
Bob_AMD hispeed: One a select few OEMs get K7s. And among those only a very, very select few get *real* K7s.
Bob_AMD Well, actually by now there are many more real K7's out there. I'll let you figure out why.
Grave Bob_AMD: what's the status of the dresden plant?
Bob_AMD Grave: Its there, and we are ramping it up. :o) 0.18 is on schedule so far.
Arg! I have to go now. There were a couple of other things I wanted to point out. I'll do that tomorrow. Oh, and he confirmed that the weird white thing I was telling you about (on the motherboard) IS a heatsink, but he couldn't remember what it was cooling.
|
