Athlon -vs- Intel

-oRnein9-

I'm just thinking about upgrading at the mo, probably going to go for P4 3.2ghz, I was just wondering does the Athlon 64 give the P4 a run for its money. Which is better? I mostly play games, Cheers!

joePC

Yes the Althon 64 is a contender when put against the P4 but really only the FX1, go with the 3.2 P4 its a great processor and overclocks very well.

Do your research weither to go for the Northwood or Prescott core. The prescott runs at a higher temp.

Thanks joePC

Woden

bah it doesn't need to be an FX to beat a 3.2Ghz pentium, the 64 3200+ will do the job nicely in games

http://www.anandtech.com/cpu/showdoc.html?i=1956&p=19

however while the new amd processors however seen some nice improvements in things like encoding and stuff the pentium's still have them here i think, depends on what you what to do with the chip tbh. as joe says the pentium will overclock better then the athlon, this is not a result of the chip though but due to no working agp/pci lock on the current 64 chipsets. this should be rectified with the new socket platform for the 64-bit chips and the nforce 3 250 chipset. so i wouldn't upgrade to a athlon at the moment with the new socket change imminent i'd wait. while if you going for a pentium you might as well pick it up now

data

Mutant_Fruit

Personally a nice swanky AMD 64 3000+ is enough to match the P4 EE 3.4ghz in most applications.

The 3000+ has half the L2 cache of the 3200+, but that doesn't impair the performance that much at all! So from the benchmark above, you can easily see that the 3200+ and 3000+ are all easily able to take on the P4 EE 3.4ghz in most areas.

So with price tags of €242.22 for the AMD 64 3000+, versus a whopping €440.06 for the Intel Pentium 4 3.4 GHz, i think the choice is clear.

Of course, if you had €450 to spare, you could always get the AMD 64 3400+ which would really whoop ass.

EDIT: Check out this benchmark

Sir Random

Originally posted by Mutant_Fruit
Personally a nice swanky AMD 64 3000+ is enough to match the P4 EE 3.4ghz in most applications.

Lol, are you basing that just on the UT benchmark?
Check out the real benchmarks:
http://www.anandtech.com/cpu/showdoc.html?i=1956&p=17

-oRnein9-

I just reviewed my budget, I can only afford an Athlon64 3000+ or P4 3ghz, Which is better for gaming?

RobertFoster

wouldn't it be better to wait a while until the 64bit technology is developed a bit more and there's more support for it?

joePC

P4 3.0Ghz

Or

3000+64

I'd go for the P4, It is a better processor all round, even if it drops a few point in benchmarking you wont notice any difference plus the P4 is way better for intensive applications e.g. Encoding, PS rendering etc....

The P4 is better.

Thanks joePC

Mutant_Fruit

I noticed the way you chose to link the stuff that Intel has always been strong at... Yes, i know intel is good at content creation. It always has been....

But check out the other results...

DirectX 9 is useless as a comparison, so we'll go through the others one by one...

DX8: the 3000+ is on average 2-3 fps behind...

OpenGL: On average about 6fps behind teh P4EE.

DivX: Intel is ahead on this one by about 10fps as compared to the 3000+, but if you used XviD as your mpeg4 encoder, AMD would be ahead. its all depending on the optimizations...

Development Workstation Performance: AMD well ahead.

If you really think that performance increase is worth paying twice the price... then go for the EE. If you want to compare price bracket for price bracket, compare the EE to the AMD3400+, and it compares very favourably.

EDIT: AMD's are better for gaming, check any benchmark out. Intels are really only good when it comes to compiling and business apps.

Matt Simis

Originally posted by joePC
P4 3.0Ghz

Or

3000+64

I'd go for the P4, It is a better processor all round, even if it drops a few point in benchmarking you wont notice any difference plus the P4 is way better for intensive applications e.g. Encoding, PS rendering etc....

The P4 is better.

Thanks joePC

Grrr.. how bout some facts and figures to back that up Joe?

Ive had both, P4 is the baseline performance CPU, Athlon64 is what excels above. Go 64.


Matt

Mutant_Fruit

C'mon, check the benchmarks before you put up unsubstantiated comments like that. I backed up my comments.

The 3000+ will excel against a P4 3.0ghz. Previously i was comparing it to the top of the line 3.4ghz EE, and it was keeping up with it (most of the time) but not beating it.

The 3000+ will beat the P4 3.0ghz hands down in gaming.

EDIT: clicky linky

Adeptus Titanicus

Now now... Lets all take a deep breath...

As far as I can see there's sod all difference between the p4 3.0 and the Athlon64 3000+. Not so much as matters anyway. What's a few FPS between friends?

When it gets this close most of it is personal preference anyway.

Woden

the 3.0ghz p4 northwood will overclock better on air, you'll be doing well to get more then 2.2ghz out of the amd on air and it will start depending on your other components like how well you graphics card likes a higher agp frequency, if the 3.0 ghz p4 and the 3000+ where similarly priced i think i'd take the pentium and overclock the crap out of it

data

Mutant_Fruit

in quake3 the AMD64 3000+ has a 30fps gain over the P4 3.0ghz.

And over 20fps of an advantage in the Unreal Tournament 2003 flyby.

Why not go for the better beast?

Gerry

Originally posted by joePC
P4 3.0Ghz

Or

3000+64

I'd go for the P4, It is a better processor all round, even if it drops a few point in benchmarking you wont notice any difference plus the P4 is way better for intensive applications e.g. Encoding, PS rendering etc....

The P4 is better.

Thanks joePC

That post is a bit headwrecking, how can a p4 drop a few points in benchmarking, but still be better at intensive applications? I think what you mean by "benchmarks" are games only, but surprisingly enough, its a popular way of measuring the performance of a broad range of applications. What you say about encoding/rendering performance needs to be backed up, the p4 is only faster on apps that have been very well optimized for it ( like lightwave, cinerama ). Even within these applications, certain scenarios will result in the a64 whupping the p4, it just does not have well rounded performance. Theres very little difference between the a64 and the p4 on stuff like 3dmax, mp3 encoding. The generalisations you are making may have had more truth in them a year ago, but you should look at some performance figures now before making them again. Tomshardware is a good place to start.

Woden

Originally posted by Mutant_Fruit
in quake3 the AMD64 3000+ has a 30fps gain over the P4 3.0ghz.

And over 20fps of an advantage in the Unreal Tournament 2003 flyby.

Why not go for the better beast?

not sure if this was directed at me but sure i'll respond anyway , at this moment in time i just wouldn't recommend an amd 64 purely as i'd be thinking of how much i could overclock it, for a non-overclocker i'd think about switching back and going with the 64 proc again and if you are intersted in overclocking i'd still go with the recommendation of waithing for the socket 930 (is that the correct numbering) platform with then nf3 250 chipset and then overclock the 64 to my hearts content,

tbh i'm an amd fan, always up for what can be conceived as the underdog, i can't deny though that intel made some good stuff with there c chips leaving the xp far behind glad amd came back with the 64 bit though i think most of the benefits from that have come from the onboard memory controller its a tough call at the moment and i still recommend waiting

data

joePC

True the AMD 64 is better @ gaming, -Rnein9-
read this artical it should answer you questions.

AMD Vrs Intel

Thanks joePC

neokenzo

You can overclock the AMD more than 2.2Ghz. I've seen some ppl pushed it to even 2.4Ghz. Maybe it wont go as far as the the P4's but thats still something. The thing is getting the right mobo. Then you are able to really push it.

Sir Random

The AMDs are definitely better for gaming at stock, but the P4 3Ghz should easily o/clock to 3.4Ghz+, which will give it the edge.

If you plan to run a stock rig, get AMD. If you want to o/clock, get the P4.

Mutant_Fruit

One thing no-one has mentioned so far (including me) is that (unsurprisingly) the AMD is 64bit. Which means that it is future proof in a way the intels aren't.

When you run a 64bit OS, with proper 64bit drivers, you will once again get a speed boost. At the moment all that is being compared is the 32bit performances. Once you get a proper 64bit rig up, everything gets a boost.

I saw something before where Lame (an mp3 encoder) encoding times were compared between 64bit and 32 bit versions, and on the same chip the 64bit version was faster...

BloodBath

I know the prescott's are behind the northwood's and pretty much everything else in benchmarks but what's this about them having 64-bit extensions?

Like some of the early P4's having hyper threading. Is there any way of enabling it?


BloodBath

leeroybrown

64bit v.s 32bit:

People keep going on about 64bit being faster but it's only inproves very specific things:

- Ability to natively address larger memory space (server apps only benefit from this)
- Ability to use much larger numbers of threads (again server apps)
- Intensive floating point maths. (Lame falls into this category along with most scientific computing applications)

These points aside the 64bit binaries themselves tend to be ever so slightly slower (a few percent) due to a larger cache footprint. Solaris, one of the prime examples of a mature 64bit OS, by default is installed with almost all it's binaries as 32bit. The only things that come as 64bit are the kernel and specific libraries and binaries that are required for 64 bit applications. Some times there are two versions of specific libraies. Similarly, 64bit Linux is shipped with almost all of it's userland binaries in 32bit. Basically, 64bit only improves things that rely on it's strengths. Likewise an AMD64 build of Windows will be 32bit with whatever is needed to full 64bit compliance/abilities specially compiled. This is why Intel aren't planning 64bit extensions for desktop processors.

The performance of something like an Athlon64 is down to the design of the CPU and not that fact that it's '64bit'. The Athlon64 benefits from huge L1 cache (128Kb on the 3200), large L2 cache (1Mb on the 3200) and a good caching model (something the P4 could do with, although Hyperthreading goes some way to eleviating this). I'm a big fan of the Athlon64 and as someone who works with server equipment like the idea of it 64bit as I can see the advantages.

secret_squirrel

Im an amd fan but I'd lean towards the P4 3gz at this stage since afaik as long as you chose correctly your motherboards gonna have a bit longer life in it than an AMD based mobo thats gonna be out of date in the next few months (amd socket change that data mentioned). Although I think I read that the current AMD64 socket will have up to a 3700/3800+ coming out for it.

astrofool

u're a bit off there on your 64bit comments leeroybrown.

Intel is going to be bringing 64bit to desktop. 64bit compiled apps generally take up less space than 32bit compilations, and will use less instructions to do the same job.

In general, going 64bit gives about a 10-15% performance improvement, all other things being equal.

nadir

Originally posted by astrofool
u're a bit off there on your 64bit comments leeroybrown.

Intel is going to be bringing 64bit to desktop. 64bit compiled apps generally take up less space than 32bit compilations, and will use less instructions to do the same job.

In general, going 64bit gives about a 10-15% performance improvement, all other things being equal.

but this increases is only noticed on 32bit binaries becasue of the larger cache, and basically its not going to make much of a difference at all on a 32bit windows compile. You would be better off speding the cash on a fast 32bit processor.
So basically Leeroybrown is right with his 3 main points

- Ability to natively address larger memory space (server apps only benefit from this)
- Ability to use much larger numbers of threads (again server apps)
- Intensive floating point maths. (Lame falls into this category along with most scientific computing applications)

So if you are just going to be playing 32bit compiled games on a 32 bit OS , you are wasting your money buying a 64 bit chip, its pointless.
On the other hand as Leeroybrown points out, if you are going to to be running a tru 64 bit kernel - servers or what not, then 64 bit by all means.

OfflerCrocGod

Originally posted by astrofool

Intel is going to be bringing 64bit to desktop. 64bit compiled apps generally take up less space than 32bit compilations, and will use less instructions to do the same job.

In general, going 64bit gives about a 10-15% performance improvement, all other things being equal.

Ummm not quite right

Most 32-bit applications perform better compiled as a 32-bit binary, because more of the application binary fits in the computer's cache. When a 32-bit application is recompiled for 64 bits, the 64-bit binary is typically larger than its 32-bit binary. With a given cache size available on a system, performance may actually decline because of a greater number of cache misses when running the 64-bit binary.

From HPs site waffling about HP-UX and there is also this OSNews article (Size Factor).

Mutant_Fruit

yes, but the thing is stuff will be coming out designed for 64-bit from the ground up, not ports from32 to 64 bit. Thats where the speed boosts will come from.

For those of you using linux, you already have a working 64bit OS, and stuff can easily be optimizied for 64bit. For those of you using windows, you willb e waiting a long time for your proper 64-bit stuff.

All in all, the AMD chips are still a great 32bit solution.

astrofool

there's an article on anandtech that will explain it all better, www.anandtech.com search for it (it goes into machine code level, so you have to know a little about your 8086's and 6800 chips)

Col_Loki

So if you are just going to be playing 32bit compiled games on a 32 bit OS , you are wasting your money buying a 64 bit chip, its pointless.

But the thing is your not paying any extra for the 64bit chip (AMD64 & Fx's)and it performs extremely well in 32bit app's & games. Its an option that cant be bad to have dont you think, any possible performance gains for no extra money has to be good....... sure isint that why we overclock

Mutant_Fruit

It helps if you think of teh AMD 64's as excellant 32bit chips, well able to hold their own against similarly price-bracketed chips from intel, with the added bonus of having 64bit support, which will (in time) lead to greater performance boosts over intel's chips.

secret_squirrel

Originally posted by nadir
So if you are just going to be playing 32bit compiled games on a 32 bit OS , you are wasting your money buying a 64 bit chip, its pointless.

Not true - if your 64bit processor can run 32bit apps better than the equivalently priced 32bit chip (as is arguably true with the Athlon64 vs P4 ) Then it makes sense to get the Athlon64 since you get a faster chip with a degree of future proofing, and a possible performance gain down the road.

OfflerCrocGod

Originally posted by astrofool
there's an article on anandtech that will explain it all better, www.anandtech.com search for it (it goes into machine code level, so you have to know a little about your 8086's and 6800 chips)

A link would have been nice

int is by far the most frequent data type to be found (statically and sometimes dynamically) within C and C++ programs. 64-bit integers require twice as much space as 32-bit integers. Applications using 64-bit integers consume additional memory and CPU cycles transporting that memory throughout the system. Furthermore, the latency penalty of 64-bit integers can be enormous, especially to disk, where it can exceed 1,000,000 CPU cycles (3 nsec to 3 msec). The memory size penalty for unneeded 64-bit integers could therefore be very high for some applications.

Oh and you can also apply that to pointers 32 is now 64. Taken from the Open Group website. 64 bit binaries are bigger then 32 bit binaries. I would still go for the A64 though; the binaries will be bigger only if they are 64 bit and as has already been said Win64 will not be with us for a while.

dazberry

Originally posted by OfflerCrocGod
Taken from the Open Group website. 64 bit binaries are bigger then 32 bit binaries. I would still go for the A64 though; the binaries will be bigger only if they are 64 bit and as has already been said Win64 will not be with us for a while.

The article in general is dealing only with code portability from 32bit to 64bit. It doesn't deal with CPU specifics that might be quite beneficial in a 32bit environment. For example, the move from 16 to 32 bit (beyond the paging mode etc.) offered additional (and 32bit) registers, additional instructions etc. in 16 bit real mode.

So if 64bit features are available in 32bit mode I can't see this as being a hinderence. These days code is so optimised with longer (and more parallel) pipelines, and additional instructions since the advent of MMX etc. I can't see these features not being used if the 64bit platform becomes popular, without having to fully make the transition to a 64 bit memory model.

D.

OfflerCrocGod

Originally posted by dazberry
So if 64bit features are available in 32bit mode I can't see this as being a hinderence.

Umm I'm not calling 64 bit a hindrance I'm just saying the binaries will be bigger - not smaller as has been put forward in the above replies^^.

dazberry

Originally posted by OfflerCrocGod
Umm I'm not calling 64 bit a hindrance I'm just saying the binaries will be bigger - not smaller as has been put forward in the above replies^^.

No I got that one OCG I guess what I was trying to say is that if for example 64bit registers are available in 32bit mode (which might not be the case btw - I don't know), then a 32bit binary would not need to be any larger but could use some 64bit features - to its advantage.

mov eeax, value64
inc eeax

vs

mov eax, value32low
mov ebx, value32hi
inc eax
jnz end
inc ebx
:end

D.

Mutant_Fruit

that looks like a helluva reduction

Either way, its really up to yourself as to what you get. You can get an intel, and try to overclock the hell out of it, or you could get an A64, and try to overclock the hell out of that (you'll probably only get about 200mhz tops, but an A64 200mhz does more than an intel 200mhz.)

Then of course, with the A64 you always have the 64bit option waiting to be used, which can only help.

And as was already pointed out, the AMD's beat the Intel's in gaming performance, which is what you siad you'd be using it for the most.

astrofool

int's floats doubles and long's all have a max size, so will take up the same amount of space regardless of how many bits it is.

What it does mean is that at machine code level, adding large numbers doesn't need a carry, and uses less instructions as a result.

unkel

Originally posted by -oRnein9-
I'm just thinking about upgrading at the mo, probably going to go for P4 3.2ghz, I was just wondering does the Athlon 64 give the P4 a run for its money. Which is better? I mostly play games, Cheers!

ok now the simple answer

If you want to get your new pc now, are going to overclock it and / or want to run several intensive applications simultaneously (e.g. game, burn cd, heavy download or encoding), get the P4

In all other cases get the Athlon 64, esp if the main use for the pc is gaming

P.S. very nice all those benchmarks, but why do they never include max overclock intel vs max overclock amd? Always seems to be overclock one vs stock other :rolleyes:

OfflerCrocGod

Originally posted by dazberry
if for example 64bit registers are available in 32bit mode (which might not be the case btw - I don't know), then a 32bit binary would not need to be any larger but could use some 64bit features - to its advantage

It would afto be recompiled(for the machine code to be changed) - so no thats not going to happen; except possibly in microcode(but since that is internal to the CPU the binary would not be diff).

Gerry

Originally posted by unkel
P.S. very nice all those benchmarks, but why do they never include max overclock intel vs max overclock amd? Always seems to be overclock one vs stock other :rolleyes:

This is standard practise on tomshardware, it comes from a time where the p4 had a definite lead over the athlon. So they always had a clocked up athlon in the benchmarks, just to show what the athlon "could really do, if only amd gave it the chance". Sometimes they even stuck a vapochill on the athlon, hilarious, but expect nothing less from tom...

dazberry

Originally posted by OfflerCrocGod
It would afto be recompiled(for the machine code to be changed) - so no thats not going to happen; except possibly in microcode(but since that is internal to the CPU the binary would not be diff).

You're right it would require a recompile, although more so than ever (AFAIR) we're seeing optimized binarys towards platforms, or more likely optimized routines within those binaries - so the binary decides what its going to run based on the CPUID. So if the Athlon64 reaches a critical mass I think this will happen if it offers a sufficient performance benefit - on certain apps.

But tbh it's all a trade off. The longer pipelines are getting the more code has to be reorganised as to not stall the pipeline (AGIs), and the more important branch prediction becomes, and the requirement for out of order execution. Mostly at the compiler level, but I can't see it not being added if its worth it.

D.

OfflerCrocGod

Originally posted by dazberry
You're right it would require a recompile, although more so than ever (AFAIR) we're seeing optimized binarys towards platforms,

True OSS being a big creator of that kinda s/w - my Firefox is always optimised

Originally posted by dazberry
or more likely optimized routines within those binaries - so the binary decides what its going to run based on the CPUID.

Making the binary bigger (The checking code has to go somewhere - and if it fails it has to use 32-bit machine code).

dazberry

Originally posted by OfflerCrocGod
Making the binary bigger (The checking code has to go somewhere - and if it fails it has to use 32-bit machine code).

True, but nowhere to the extent of a full 64bit binary which is the distinction I'm trying to make - its what happened with the 386 even though most of your programs were realmode DOS. Say your app uses 64bit arithmatic, and you can for the sake of a couple of KB get a 10% speed increase on selected platforms, you're going to use it. If the CPUID code isn't already there (but if you're at this point - it probably is) its not much (< 100bytes), not nearly as much when CPUID wasn't common (~ 3-4KB).

So you have a 32bit binary, you're not stuck with all ints being 64bit and the associated overhead as described, 64bit aligned structures for speed etc. so its the best of both worlds until it becomes viable to move to full 64bit.

D.

astrofool

all int's are not 64bit, if they were, there would be no need for long, float and double

Dempsey

At stock speeds there is little difference between the Athlon64 and its Northwood equivalent. Athlon barely ahead. P4's (Northwood's) overclock better than the Athlon64

People buying Athlon64's now thinking that they are futureproof are wasting their time. The 64bit sockets are changing (not sure when but they are), BTX is coming in the summer. By the time 64 bit applications/OS's get optimised, you could buy a 64 bit cpu THEN for the money you pay now and have a better system that is future proof.

Thats my reading of it

dazberry

Originally posted by astrofool
all int's are not 64bit, if they were, there would be no need for long, float and double

Read the link Taken from the Open Group website that OfflerCrocGod posted .

D.