AM2 any real performance increase

Phemo · 23 May 2006 at 11:38

harris1986 said:
anyone know how much cache the am2 chips have got?

Exactly the same as their Socket 939 counterparts. The X2 5000+ has 512k per core however and the FX-62 has the standard 1MB per core.

MrWhippy · 23 May 2006 at 11:46

If AMD don't have anything special up their sleeves to counter Conroe, maybe they'll put up a fight with some aggresive pricing strategies on what they do have.

we can but hope

Indy500 · 23 May 2006 at 11:49

http://enthusiast.hardocp.com/article.html?art=MTA2NSwxLCxoZW50aHVzaWFzdA==

This seems to show AM2 in a more positive light.

mcmad · 23 May 2006 at 12:11

People really need to get away from the idea that a smaller CD process will bring magical results, they simply wont, going from 90nm to 65nm will allow a higher clock speed & possibly more cache on board but thats all, its not a new architecture, its just allows a speed bump. Once a few people get AM2 under phase change that will give you an indication of whats to come from AMD.

The reason Conroe is looking unbeatable at the moment is nothing to do with the process CD its just that Intel are coming back big time with what looks like a cracking design (fingers crossed the real performance is as good as we are seeing so far).

funny thing is AMD will have to crank up the speed to compete.. nice role reversal coming up..

Minstadave · 23 May 2006 at 12:26

mcmad said:
People really need to get away from the idea that a smaller CD process will bring magical results, they simply wont, going from 90nm to 65nm will allow a higher clock speed & possibly more cache on board but thats all, its not a new architecture, its just allows a speed bump. Once a few people get AM2 under phase change that will give you an indication of whats to come from AMD.

The reason Conroe is looking unbeatable at the moment is nothing to do with the process CD its just that Intel are coming back big time with what looks like a cracking design (fingers crossed the real performance is as good as we are seeing so far).

funny thing is AMD will have to crank up the speed to compete.. nice role reversal coming up..

Yup thats exactly how I see it, AMD needs to do something new to keep up.

Corasik · 23 May 2006 at 12:47

AMD dont seem to be increaseing the clock rates that much with the 65nm process. I believe the FX62 is still on 90nm, and for the moment at least, the 65nm process will be used on the mainstream processors, to get the power requirements down to 65W, and 35W for the low power chips.

Whats more, the K8L that some people claim will eat up the Core2's is primarily a workstation processor, Its main improvements are supposed to be the FPU, and the SSE. But Conroe already improved SSE performance greatly, and raw FPU power is a benifit mostly to purely scientific applications. Conroe could well be the processor of choice for the average 'Desktop' computer, and it looks like Woodcrest (core2 xeon duo.. whatever they call it), on Bensley platform is going to be very quick too, and scales nicely up to 4 cores.

AMD may well hold onto the performance lead on 8way servers, only time will tell to see how badly the FSB hampers Woodcrest in 8way systems.

mcmad · 23 May 2006 at 12:52

Corasik said:
AMD dont seem to be increaseing the clock rates that much with the 65nm process. I believe the FX62 is still on 90nm, and for the moment at least, the 65nm process will be used on the mainstream processors, to get the power requirements down to 65W, and 35W for the low power chips.

they dont sell any 65nm products yet & wont for quite a few months, production tools are only now being installed.

Corasik · 23 May 2006 at 13:03

Hence the words 'will be', as in coming soon, not yet.

paddy001 · 23 May 2006 at 14:33

Im really hoping that AMD wint sit there and do nothing about conroe but as has been said before there doesnt appear to be anything amazing coming from thier roadmaps. Well nothing that appears to be able to rival conroe anyway.

Im seeing a massive role reversal coming with AMD taking the back seat and intel becoming the processor of choice in gaming rigs for probably about 2years+ untill AMD pull out the stops and bring out something that will put them in the lead again. However i really cant see this happening untill we get a completely new AMD architecture such as the rumored K10.

Paddy001

Indy500 · 23 May 2006 at 14:36

Would we have conroe if AMD hadnt produced K8 to beat down the Netburst P4\Xeon? Its in our interests for AMD to come up with a solution that rivals Conroe (competition, price drops, tech advances etc), although that may not happen for 6-12 months yet.

paddy001 · 23 May 2006 at 14:51

sr4470 said:
Would we have conroe if AMD hadnt produced K8 to beat down the Netburst P4\Xeon? Its in our interests for AMD to come up with a solution that rivals Conroe (competition, price drops, tech advances etc), although that may not happen for 6-12 months yet.

Couldnt agree more, I heard just recently that a couple of chinese companies were begining to make good processors. Competition in the market is always best for consumers look at the gfx performance leaps over the past 12 months.

NathanE · 23 May 2006 at 14:57

sr4470 said:
Would we have conroe if AMD hadnt produced K8 to beat down the Netburst P4\Xeon? Its in our interests for AMD to come up with a solution that rivals Conroe (competition, price drops, tech advances etc), although that may not happen for 6-12 months yet.

Yes wait for K10. It has Hyperthreading 2.0. Not 12 months away, more like 16-18 months. It's not on their roadmap anyway.

Indy500 · 23 May 2006 at 14:59

NathanE said:
Yes wait for K10. It has Hyperthreading 2.0. Not 12 months away, more like 16-18 months. It's not on their roadmap anyway.

Hmm...well I'm not in the market for a new power system right now, so that may come in time for my next upgrade.

Jay_t · 23 May 2006 at 16:29

Its certainly true that competition is good for the consumer. If it wasn't for AMD, an Intel CPU would probably be as much as a second hand car by now. Never the less, I will be jumping ships to the dark side when Conroe gets here.

Corasik · 23 May 2006 at 18:36

Whatever K10 may or may not have, it wont have a system called Hyperthreading. That design, and name is trademarked and knowing american industries probably has patents applied for.

If you mean what people call reverse hyperthreading, then someone should think of a better name, as a system to run unthreaded applications on multicpus really shouldnt be called hyperthreading anyway. Its also pretty pointless, as both AMD, and Intel have quad core cpus on the roadmaps by then, and the programmers will hopefully be writing multithreaded code anyway.

NathanE · 23 May 2006 at 19:45

Corasik said:
Whatever K10 may or may not have, it wont have a system called Hyperthreading. That design, and name is trademarked and knowing american industries probably has patents applied for.

Well yeah... you know what I mean.

Corasik said:
If you mean what people call reverse hyperthreading, then someone should think of a better name, as a system to run unthreaded applications on multicpus really shouldnt be called hyperthreading anyway.

"Hyperthreading" or symmetric multi-threading (SMT) just refers to any logic that helps the processor keep its pipeline(s) full.

Corasik said:
Its also pretty pointless, as both AMD, and Intel have quad core cpus on the roadmaps by then, and the programmers will hopefully be writing multithreaded code anyway.

Multi-core and multi-threaded software is indeed where future scalability will come from. However neither manufacturer have turned their backs (yet) on single threaded performance. A "reverse Hyperthreading" feature (or what I like to call HT v2.0) is simply a way for a thread of execution to utilise the pipelines of all (or a selection of) available cores on the chip when they would otherwise be idle or free pipeline space is available. Super-duper-scalar chips like the IBM POWER have had this sort of technology for donkey's years.

Minstadave · 23 May 2006 at 19:51

Reverse hyperthreading would be very interesting, currently nearly everything is single threaded, if you execute that single thread using multiple cores resources that would be very interesting indeed. A dual core processor running a single thread at twice its clock speed would be a beast.

NathanE · 23 May 2006 at 19:55

Minstadave said:
Reverse hyperthreading would be very interesting, currently nearly everything is single threaded, if you execute that single thread using multiple cores resources that would be very interesting indeed. A dual core processor running a single thread at twice its clock speed would be a beast.

It'd be nowhere near that. Only certain instructions can be parallelised. I'd say the gains would be closer to 15-25% under optimal conditions.

Think of branch prediction, the standard example. When a program has a bunch of nested IF statements each possible combination can be placed into the pipeline of alternating cores in the processor. The cores can each "read ahead" in the program by prefetching and executing the next few instructions - in the hopeful event that their branch turns out to be the correct one. Once the branch has been decided upon by the "original core" the thread can either jump across to the core which has performed the reading ahead/prefetching, or by the means of a shared cache the current core can continue execution using the work done by the other core.

This is just scratching the surface really.

Minstadave · 23 May 2006 at 20:05

NathanE said:
It'd be nowhere near that. Only certain instructions can be parallelised. I'd say the gains would be closer to 15-25% under optimal conditions.

I know you'd never get anything like double the performance, it was just wishful thinking

Even a small boost in performance is good though, most of the time you've got one core sat there doing not very much, may aswell put it to use

MrWhippy · 23 May 2006 at 22:26

Minstadave said:
...most of the time you've got one core sat there doing not very much, may aswell put it to use

you my friend, need to get folding