AMD AM2 To arrive early! + New intel cores

[Dunky]

What this is good for is legacy applications. Older, or at least current, games which aren't multithreaded will benefit. Lots of old, but still usable, custom business software will benefit. Loads of stuff that currently depend upon a single thread will benefit.

Will it though? Business software and older games are hardly taxing for current CPUs. Current games may have a slight benefit but by the time AMD release this then we'll have moved onto other games - potentially multi-threaded ones.

Even programs that are currently multithreaded will benefit; you can have a program that spawns 20 threads, but if 19 of them barely do anything and only one uses lots of CPU times, then you will still find that a multiple-core machine will not be used to its full potential. This will be because the operating system assigns a thread to a core; but if only one core does most of the work then the program will still benefit from split over two cores.

You have to remember that they're are other threads (mainly O/S threads) that play apart in this and the OS scheduler will handle the balance of the threads. Easiest example is the main thread on one core and O/S Threads, background tasks and the other 19 threads on a seperate core. If the OS has multiple cores available then it'll handle the loading of those cores so that half the cpu isn't twiddling it's thumbs (and the Windows / Linux schedulers do a pretty good job of this task).

The thing is, it is incredibly difficult to write code that is really well threaded. Sure, you can write code that spawns extra threads here and there, and that is a good thing. But synchronisation issues are very much at the forefront of any coder who is doing concurrent programming.

Yes it is difficult to write threaded code, but not impossible - it's something which comes from experience as a programmer. This is the way things are going so they better start looking at multi-threaded development because it does have great benefits and is becoming a vital skill. As you've said, concurrency between threads and sync'ing threads is an issue in programming but it's not an issue to which we don't have a solution. We know how to sync' threads in various languages - it's just utilising it to it's full potential which is the current problem.

The other problem is how many cores too take advantage of? Two is a safe bet, but what about in the next year or two when four cores becomes more popular? Adding extra threads does create overheads - this anti-HT idea would give the benefits of adding extra threads but without the overheads that comes with it.

Developers shouldn't write for a specific number of cores. In order for a program to be truely scalable, it should have as many threads as it takes without going overboard. If a program can utilise 100 threads effectively then it should be written to spawn 100 threads (Apache is something that comes to mind - I think Apache spawns a thread of each persistant connection to it). I remember when HT came out and quite a few programs 'updated' their software to use this when all they basically did was make it dual-threaded - this won't scale well on Quad proc/core systems. Multi-threaded programs are mostly about scalability and hence shouldn't be written for a specific number of cores / procs.

Adding more threads creates very little overheads unless they are poorly coded. In actual fact, in single proc systems it makes scheduling and multi-tasking a bit more efficient (but not by much) and in SMP enviroments it's used quite efficiently. Even if a thread doesn't do all that much, it still allows the scheduler to arrange threads as efficiently as possible and provide good load balance over all processors and cores.

 

[TooNice]

I've seen a few ES core benchmarks for the intels on certain forums as well but to be honest you can't even trust them, keep in mind that major players in the electronics industry realise the selling potential of hype, and actively post comments on forums to create hype around there products, by telling everyone how they have samples of the cores and how good they are and that when there released we should all blindly trust them and buy the product. I am not saying that the results are false its just best to wait until there avaliable to consumers and then wait for reviews and comments from the consumer products not from ES.

Remember TRUST NO ONE!

My take is that forum goers are educated users. They will look at benchies by reputed review site before taking out that credit card regardless of the hype. It would be very stupid for them to create hype and not deliver. Whether we like it or not, Intel is not stupid. So until I see reviews that say otherwise, I will accept the numbers available, and base my prediction on them, as many here have done: the Conroe has won this round. It is -not- what I would like to believe (just built an Opteron system), I think the odds are not in favour of AMD.
And if you think we know too little about Conroe to make that statement, then I guess no one should make any statements about the K8L, let alone the K10. There aren't any figures out at all
On a side note, I am also a bit skeptical about the reverse hyperthreading thing. I guess that it would make a difference, but perhaps more in line to what hyperthreading was for P4. But we'll see. This one is definitely too early to debate.

 

[bbreezeuk]

now intel have ditched netburst they should rock but im not convinced they will be bettter than amd

 

[Gashman]

why on earth are people saying conroe WILL demolish AMD, you realise how utterly stupid it is to say something WILL happen, lets look at the only real facts here, some reviews done on conroe (not available for while yet) on engineering samples, they scored great on bloody benchmarks, and now every single person on this forum is drooling, waiting to buy one, your all counting AMD out before seeing what they have up there sleave, which is frankly stupid since the guys at AMD are far from stupid, intel have the money, AMD have the brains, don't count the boys over at advanced micro devices out just yet

 

[Dunky]

now every single person on this forum is drooling, waiting to buy one, your all counting AMD out before seeing what they have up there sleave, which is frankly stupid since the guys at AMD are far from stupid, intel have the money, AMD have the brains, don't count the boys over at advanced micro devices out just yet

With good reason, we know what AMD have up their sleave...AM2 which doesn't exactly bring any great boost at all. Even K8L from what we know doesn't bring any major benefit to the take that Intel haven't already done with Conroe (maybe except the multiple FPU but that remains to be confirmed). We know to take benchmarks at this stage with a pinch of salt, but the fact that benchmarks are consistant between reviews means that this is something to look forward to.

I don't think people are saying Conroe will demolish it, but a 20% average increase in performance in not only benchmarks but games and music encoding (which aren't synthetic) does say that this round will be Intels and for the sake of competition that AMD bring something about sooner rather than later after AM2.

 

[Dolph]

If you actually look at the underlying architecture for the processors, it's easy to see why Conroe will own the K8 from a pretty great height.

The work per clock of the Conroe (always the strong point of the Athlon's over the P4 chip) is higher than even the K8, it is a genuine reversal of strategy, and Intel have done it even better than AMD did.

There's a good (and easy to understand) article about the architecture here at Arstechnica that explains why far better than I can.

 

[cypriobrit]

Nice link

 

[Dolph]

why on earth are people saying conroe WILL demolish AMD, you realise how utterly stupid it is to say something WILL happen, lets look at the only real facts here, some reviews done on conroe (not available for while yet) on engineering samples, they scored great on bloody benchmarks, and now every single person on this forum is drooling, waiting to buy one, your all counting AMD out before seeing what they have up there sleave, which is frankly stupid since the guys at AMD are far from stupid, intel have the money, AMD have the brains, don't count the boys over at advanced micro devices out just yet

Oh dear, I wouldn't say AMD have the brains over Intel, just because intel made a huge assumption and gamble with Netburst that didn't pan out in the end...

Perhaps a crash course in CPU architecture, (I'd suggest starting with the pentium pro and working up, including chips from both Intel and AMD) would allow you to see that jumping to either side is rather flawed. The big reason AMD have got the foothold they have was Intel's cockup with netburst (P4) that didn't scale as well as they thought it would, which in turn lead to very nasty fixes (such as increasing the pipeline length), that in turn meant they had to increase speed and therefore power consumption.... AMD got where they were by NOT being revolutionary when Intel tried to be, (The K7 followed on from the fundamental design ideas of the P6 core from intel, aka the Pentium Pro, Pentium II and Pentium III)

There are some very good architecture articles available here.

http://arstechnica.com/articles/paedia/cpu.ars

That would be a really good place to start.

I'm not an Intel or AMD fanboy, I've owned both, I buy whichever provides the best performance/stability/price ratio at the time I want to buy (I'm currently holding out for Conroe to be released before I make any decisions), but both sides have done good and bad things (anyone remember the AMD K6-2?)

Counting AMD out is not a good plan, but nor is assuming that they will have something to better conroe (or indeed any Intel chip) based on their previous performance, as their performance owed a lot to intel getting it wrong, rather than AMD getting it right....

 

[HangTime]

Even programs that are currently multithreaded will benefit; you can have a program that spawns 20 threads, but if 19 of them barely do anything and only one uses lots of CPU times, then you will still find that a multiple-core machine will not be used to its full potential. This will be because the operating system assigns a thread to a core; but if only one core does most of the work then the program will still benefit from split over two cores.

The thing is, it is incredibly difficult to write code that is really well threaded.

For me, this is the major problem with dual-core in the gaming environment (and consequently, the use of gaming benchmarks when assessing performance of new cpus). Developers have been trying to code effectively for dual core as far back as the 1990s, with Quake3 having SMP support (intended at the time for dual-cpu rather than dual-core, of course). However, the performance gains were only of the order of 10% because one thread was using far more cpu than all the others combined - simply deciding to put eg a 'sound thread', 'input thread' and 'ai thread' on to one core isn't going to suddenly give us a huge boost in performance if it turns out that they are only a small proportion of the total cpu overhead.

That's not to say I don't think that Dual Core is the way forward, but there is still a need for individual cores to deliver high levels of performance at least for the forseeable future. Things like the AMD X2 range are actually quite good because even in single core mode, an A64 at over 2ghz is still reasonably quick. But for me the danger will come if we end up with cpus consisting of say 4 cores running at relatively low clock speeds. In suitable applications designed to take advantage, we will of course be seeing stellar benchmark scores - but for single threaded apps we could end up in a situation were a 3 year old FX-57 or similar could actually be delivering superior performance.

 

[TooNice]

Oh dear, I wouldn't say AMD have the brains over Intel, just because intel made a huge assumption and gamble with Netburst that didn't pan out in the end...

Seconded. AMD won the Athlon vs P4 round (although they did trade punches, the Northwood was for a short time leading over the Barton). But it is not like AMD has always had the best tech/chip.

 

[Gashman]

all im trying to say is, following the trend, GPU/CPU wise newer architectures always seem to be better overall performers than the ones they replaces, and K8 is due to be replaced soon anyways, i would like to think another good intel vs. AMD battle, great for the customer mind

 

[Kamakazie!]

Who's to say that Intel have not got a good path of upgrade with the new Coroe setup? It might be foolish to assume that Intel have not got anything planned?

no one said it, im just basing it on what has happened recently with both manufacturers platforms.

the 975(is that right) or whatever socket conroe is going to be running on might well last a year or 2 but that would certainly be something different from intel. Im just gonna take a closer look at the roadmaps before i purchase, which wont be for a few months yet.
I heard rumours of K8L being on AM3 as well but then many more have said otherwise.
I've been led to believe that K8L will bring much more to the table than just FPU improvements and 65nm but we shall see.
It is certainly nice to see Intel competetive again. Even though them not being didn't seem to impact their huge market share as much as it should (partly due to AMD's lack of available chips)