65NM A64s have slower L2 cache..

Reports that the increased latency in the cache is so that in the future AMD can increase its L2 cache size doesn't hold water for me.
 
Doesn't AMD already use much higher latency L2 caches than Intel? Nevermind the huge bandwidth disparity. I'd lean towards the latency increase being for clock headroom, but considering the cache's previous limits already this is rather disappointing.
 
Cant they use the faster latencies for the chips now and only increase the latency when and if they move to larger cache?

They really should decrease the model numbers they use on the 65nm parts. The older 90nm 4800+ beats out the newer 65nm 5000+. That's pretty screwed up right there. Its certainly no way to win over the performance fans.
 
AMD is learning from Intel, decrease IPC for increased clockspeed - got Intel far didn't it? ;)

It does make me wish that AMD did not give up on the old Socket 939 platform so quickly.
 
I think it's just about "missimplementation" in a result of hurry-up product launch -- all in all, they had to show some working part on 65nm node to the world, at all.
From a technical point of view, it's not for the first time that AMD used more conservative SRAM layout design for the L2 cell array, for a premiere shrink step. They are just training for the big-bang (QC part).
 
Last edited by a moderator:
Read the discussion at aceshardware, according to Paul DeMone and few others, increasing latency maybe opened possibility to decrease power dissipation with 20%, speculation is that this way AMD wants to be "performance per wat" competative with Conroe based CPUs until K8L is ready.
Look at tests at Anandtech - Opterons loose on max performance, yet they win watt-race
Sacrificing 1-2% performance which is negligible and winning maybe 10+% in power consumption... good tradeoff
 
That may be their only recourse. They can't touch C2D right now. But, it has a similar feel to VIA's line of "power conscious" CPUs. Can't keep up on speed, design for low power. Oh, but they can keep the ratings the same. K! lol. Yuck.

It amazes me that my 11 month old Opteron 165 is basically still right around the top of the heap, at the 2.6 GHz I run it at. And 165 was introduced well over a year ago.

Considering the overclocking headroom of a Conroe (1 GHz oc with stock volts is common), AMD are in trouble. Can K8L do it for them? Kinda doubt it. I don't see it being the 30%-faster-than-K8 core that it needs to be to really get ahead.
 
Last edited by a moderator:
That may be their only recourse. They can't touch C2D right now. But, it has a similar feel to VIA's line of "power conscious" CPUs. Can't keep up on speed, design for low power. Oh, but they can keep the ratings the same. K! lol. Yuck.

It amazes me that my 11 month old Opteron 165 is basically still right around the top of the heap, at the 2.6 GHz I run it at. And 165 was introduced well over a year ago.

Considering the overclocking headroom of a Conroe (1 GHz oc with stock volts is common), AMD are in trouble. Can K8L do it for them? Kinda doubt it. I don't see it being the 30%-faster-than-K8 core that it needs to be to really get ahead.
consider: can they fight Conroe with old architecture on 90nm vs 65nm? No
Can they fight even on same process? Obviously no.
So, they are smart enough not to try the impossible.
As for how fast K8L can be - we'll see, saying that K8L can't be 30% faster than K8 is like saying that COnrow can't be 30% faster, whoch is wrong as we all know now :D
 
who cares about that 1%, really it's nothing like VIA which still hasn't caught up against pentium III and Athlon. (but a VIA processor is certainly what you need for office tasks, web, music playback etc., I hope they were more widespread in companies/school/universities/laptops)

about K8L, I don't doubt it'll match the conroe, afterall it brings OOO this, SIMD that equivalent to what's in conroe, plus a L3 cache.

about those "new" 65nm A64 CPUs.. are there single cores as well?

AMD still rules the single core market, where Intel is still selling his miserable pentium 4 and celeron. wanna how would fare a 65nm sempron 3000+ or athlon 3500+.
 
single core is dead for all users as far as i'm concerned. I dont think OEMs even offer them anymore in their setups, and with Pentium and Celeron Ds at $100 US or below (not the best gaming but they work superb for basic system function), why should anyone ever buy a single core again? $100-$200 can actually buy you quite a bit of performance these days processor wise. Its the buyers market right now.
 
Considering the overclocking headroom of a Conroe (1 GHz oc with stock volts is common), AMD are in trouble. Can K8L do it for them? Kinda doubt it. I don't see it being the 30%-faster-than-K8 core that it needs to be to really get ahead.

It's specualted the sustained FP/SSE performance for K8L will be somewhat higher than C2D and with an improved memory controller and cache hierarchy the integer IPC can only improve. I doubt performance wise there will be anything between them.
 
Q: What do you need dual core for?
A: Because I want things to go faster!
Q: Do you encode and burn DVDs while playing games? Or do other things all the time that makes your system unresponsive?
A: Eh...
Q: Btw, if you do, did you look at your memory usage? Because that's generally the bottleneck.
A: Well...
Q: Did you benchmark a single-core system running a 64-bit OS with lots of RAM against a dual-core system with less RAM?
A: Actually...
Q: Responsiveness is mostly in not needing to swap. Unless you want to do some folding at home while playing your game.
A: But...
Q: Then again, a Core 2 Duo also has the fastest single processor at the moment. More or less. So it's a good but expensive choice.
A: Thanks! I'll take one of those.
Q: Then again, it probably won't be noticeable faster, if at all, than filling out your RAM and reinstalling Windows.
A: ...

;)
 
Back
Top