Comparing the technical merits of the Athlon and Pentium 4

JustHalf

Strangely enough, the GeForce 4 Ti4600 is CPU limited in Quake III at least up until 1024x768 (full detail), and I believe it's limited at 1280x1024 (full detail).

The reason for running the engine at the low resolutions is to guarantee that the graphics card isn't interfering with the result, so that what each chip contributes to the performance can be better seen.

Unfortunately, even at 640x480 this isn't always the case, as Phil can testify when playing Quake III with a GeForce 2 MX on a 1.6NW @ 2.5GHz (it wasn't a gaming machine, by the way). But you get the idea.

Gerry

It is to show the relative performance of motherboard/cpu combinations. Quake3 is very good at highlighting differences in bandwidth and latency.

As for benchmarks, well this topic wasn't supposed to be about benchmarks, it was supposed to be a general discussion about the merits of the 2 architectures.

This all came out of a discussion about what chip to get in a new machine, and it was put across that the athlon is far faster than the p4. About 2 months ago, they were roughly level overall, the athlon being much stronger in some things, the p4 in others. Now I think the p4 has pulled ahead, the top end p4 is faster overall than a top end athlon.

With regard to bus speeds, well the 533mhz p4 bus is an official speed, so I don't see the problem with it. AMD are strangling the athlon by not giving it an official 333mhz fsb, they have admitted that all their efforts are currently going towards the hammer.

Back to benchmarking, the athlon is optimized to run code compiled for the p6 architecture, it gains much less from optimizaton than the p4. Basically it does a great job of keeping itself fully occupied (ie all execution units executing instructions ) as much as possible. Besides, the palomino has sse, so even if there was sse code in quake3 (and I think most of the benefit actually comes from optimizations in nvidias q3 drivers), an intel compiler gives no advantage to the p4, until you start adding in sse2 code.

So, besides the few apps which are optimized for the p4, I don't think any other normal, commercial x86 apps favour one architecture over another.

BabyEater

At the minute the P4 has the advantage over the XP as a 2.5Ghz P4 outperforms an XP2200. There is very few benchmarks where the XP can touch the P4.
I think the P4 also has greater overclocking headroom as the new thoroghbreds while disipate less heat have a smaller die size so dissipate more heat per square mm. They should have a heat spreader as the P4.
The P4 is only at its best tho with the Rambus, DDR hits performance.
And i also like the P4's thermal protection as i could have done with some when i fried me XP.
Thats what i think anyway. When i bought me yoke tho it was the other way round where the XP had the advantage over the P4 it just keeps changing and as far as i can see this is Intels time for performance .

SickBoy

Originally posted by PPC
He's got a 1.8NW, 512Mb DDR and a TH7 II mobo and a GF2.
I just need to clear the details and get him to get Q3.

Dont mean to be picky here but the TH7 II uses RDRAM, just incase anyone is considering purchasing one.

cerebus

Originally posted by JustHalf
I'd like to see people compare the technical merits of the Athlon and Pentium 4 chips that AMD and Intel (respectively) have released.

I'd like the following grounds covered at least:
* Chip features and support

I don't know if this is the kind of thing you are looking for, but a guy a few cubes over used to work for Intel. We were discussing the relative performance of AMD and Intel uP cores recently, and he had some interesting things to say about the P4.

One of the things he mentioned was that part of the integer unit in a P4 runs at 2x the core frequency - so if you have a 2.5GHz core then part of the die is running at 5GHz. This gives a clock period of 200ps, which is pretty nifty! (Some devices I've worked on have used 100ps or more of clock uncertainty, which would blow that timing budget completely!)

I must see if he has anything else interesting to say about the different architectures used by AMD and Intel - his knowledge of AMD devices is pretty limited though. Also, he worked on the P3 so I don't know how up to date his info about the P4 is

On the fab side, I have heard that Intel should have no issues with moving to a low-k dielectric process (maybe they have already?). The use of a low-k dielectric gives some nice speed and power benefits. I'm not sure how AMD is fixed in this area, but I do know a lot of the merchant silicon foundries (TSMC, UMC, Chartered, etc.) have had trouble getting their low-k process up and running. This might be an issue for AMD in the future maybe?

PPC

Originally posted by SickBoy

Dont mean to be picky here but the TH7 II uses RDRAM, just incase anyone is considering purchasing one.

Yeah sorry.
He was getting a board with DDR but changed it.
Whoops:rolleyes:

Gerry

Intels process is a bit ahead of amd's at the minute. AMD were due to use SOI (Silicon on insulator) technology in the thoroughbred and barton, that slipped to only being in the barton, now it won't be in the barton at all.

The 0.13 micron amd process is not showing near the improvements in heat and headroom that the intel one did, this is because amd used a lot of tricks from 0.13 micron process in their 0.18 micron process, as far back as the tbird, and more extensively in the palomino. I think amd are having major problems getting their 0.13 micron process running smoothly, but I'm sure they will sort this out.

The thing about the p4's integer units, is that despite running at 2x the frequency, the p4 still has pretty poor integer performance.

Heres an excellent article about the p4 architecture:

http://arstechnica.com/cpu/01q2/p4andg4e/p4andg4e-1.html

Matt Simis

Well, as seen on the HW tweaking Board, I have changed from an AthlonXP 1912mhz (XP2500) to a P4 1.8 running at 2.2.

Im using a BD7 motherboard with PC2700 ram.
While 3DMark may have fallen out of favour, for what its worth, the P4 2.2 did outperform the supposedly faster AthlonXP by about 3%. The P4 CPU was cheaper than the Athlon XP (it was a AXP 1900 originally).

I think there are two ways of comparing the platforms (platforms, not CPUs, a CPU is nothing without a good platform).

(A) Absolute performance at any cost.

or

(B) Price Performance


I think most can (sometimes grudgingly) agree that the P4 wins in category A. A P4 2.53 with 1066MHz RDRAM is untouchable, and you can of course overclock it fairly easily.

Category B is much closer. For the lowest end, its the "P4" Celeron 1.7GHz versus the Duron\low end AXP. I would tend to favour AMD in this area. When you get to the mid ground, its a bit of a toss up. AMD if you dont overclock, Intel if you do (P4 1.6NW running at 2.4GHz is unlikely to be beaten, by say an AXP 1900 or 2000 even)

For the high end price performance argument, I feel AMD is sorely lacking a competitive "high end" part, it tops out at the 2200 (just this week), whereas Intel has plenty to offer all the way upto 2.53. The AMD Tbred is a disappointment to me, but I imagine\hope that Hammer will rectify this situation.


As a by the way, I have changed CPU to a P4 2.26 (533FSB) this week, and without using my chilled water cooler, just the Intel Heatsink&Thermal Pad it can run 3DMark with 100% stability at 2.85GHz. 3GHz+ is the aim with Chilled cooling and Voltage mods to come later.



Matt

Gerry

eh nice one matt. Post up some benchmarks there

Inspector Gadget

At the moment, it's clear from the way things are going that the high-end Intel parts are faster and will continue to be faster, until AMD get their fab processes sorted out.

However, all-conquering CPU performance isn't everything - I just did a quick price comparison, and according to www.komplett.ie the fastest Intel part (the 2.53GHz P4), at E900, is just shy of three times the price of AMD's current fastest part, the XP2200 at E306.

While that's a slightly skewed example, as the price of parts increases exponentially compared to how close they are to the "top of the heap", so to speak, it still speaks volumes.

Now, if we try to pick a "sweet spot" as regards bang-for-buck for the Athlon XP, as a semi-reasonable measure of the CPUs most likely to sell in volume (at the moment), it appears to me that the chip with the biggest price differentials either side of it is the Athlon XP2000+ at E204.

With the P4, the 2.2GHz and 2.26GHz parts represent solid value compared to their faster counterparts (based on the assumption that the OEM pricing trend will follow the retail pricing trend), but at E350 are substantially more expensive (70% more expensive) than the Athlon XP parts listed above.

Throw a
benchmark into the mix and the suggestion is that both running DDR RAM, there isn't much in it. Interesting...

Gadget

Gerry

Agreed, but most people here will be able to overclock the p4, and you don't need to buy a 2.2 or 2.26 to do that, a 1.6 or 1.8 will get you 2.4 - 2.6 ghz. If you will never be overclocking, amd is better value, otherwise its not.

koneko

I'm not sure what the fluck AMD are playing at, but their CPUs are getting worse and worse at overclocking as time goes on.

The TBird were great, the XPs were kinda bad, and the Thoroughbeds are meant to be even worse.

Yet the Intel CPUs are getting great OC'ing out of them. Sorry, did I say great? I meant brilliant.

If you want to get a CPU at the moment and get the highest speed possible with it, you're going to be very disappointed with yer Athlon XP.

Inspector Gadget

True, but because it's purely FSB-based, the overclocking will be dependent on your motherboard being able to change the FSB/PCI speed dividers (which most of the current P4 boards do, I understand) and your RAM and expansion cards being able to handle higher interface speeds (I've heard that NICs are particularly tricky).

So, to be honest, a proportion of the P4's "overclockability" is down to the fact that motherboard designers working with Intel CPU hardware have had to deal with permanently multiplier-locked CPUs than the AMD boys (as the older Socket A Athlons and Durons were easy to "fix" in this regard). That, and the fact that Via chipsets have a well-earned bad reputation, but that because of they're everywhere in AMDland that AMD themselves aren't doing chipsets any more, unlike Intel...

Having said that, Intel's silicon has always been "better" somehow than AMDs - the AMD stuff tends to be released quite close to its physical limits (heat, speed etc.) all the time (continuous research in semiconductor physics permitting) whereas Intel's generally seems to have greater "legs" (so to speak) - maybe they're at an advantage considering they've been the ones dictating how the x86 instruction set will change every time the processor's bus width increases (8 -> 16, 16->32, etc.) and therefore have a "head start". With talk of poor initial results from the prototype Hammers abounding, it'll be interesting to see what AMD can do to pull the rabbit out of the hat this time.

Gadget

Gerry

You get a 33% overclock without putting anything out of spec on any of the current p4 boards, and with the i845 chipset, you can keep the pci at sensible speeds at crazy fsb's.

The sis 645 chipset doesn't appear to have a 1/5 pci divider, but the machine still ran fine at 40mhz pci with a realtek nic ( which are normally dodgy, but perhaps this was more tolerant of fsb because it was integrated, I dunno).

Inspector Gadget

...I can't say that I've done a whole pile of heavy-duty FSB overclocking - I've never really had to, thankfully, but on a couple of occasions that I've tried, things have started "going west" (so to speak) when the PCI bus is running at as comparatively "low" as 36MHz - I think it's quite likely that Realtek RTL-based NICs that are to blame, too, for what it's worth.

To be frank, though, right now the Intel rigs look much more appealing if (like many of us here) you want to crank it up a couple of notches...

Gadget

Matt Simis

What type of benches do you want?

From memory (over the weekend) I got the following Sisoft Sandra Scores (which P4s score poorly on):

5800 CPU Arith
3600 (approx) SSE2 FPU
11,400 Multimedia Bench.
2620MB/s (for both results) Memory.

In Commanche 4 benchmark, at 1024x768, all effects on Max, I got 58FPS. Slow memory is holding everything back at the moment.

I will be getting new RAM soon, but for the moment Im using cheapy Komplett PC2700 TwinMos ram, at the slowest settings in the bios. The system is running at 3000MHz, 176MHz FSB (704MHz 4x pumped!), 176MHz RAM (DDR 352).

The PCI speed is run in spec at 33MHz, as the bios features 3 lock speed settings: 33MHz, 37.5MHz and 44MHz. It also offers standard dividers (3x, 4x, 5x) too. CPU voltage is 1.75V.


What was funny was that I had to adjust the software program Motherboard Monitor's "CPU Scale" to display Values upto 3100MHz, instead of the standard setting of only upto 2000Mhz.



Matt

Gerry

Extremely nice. If you could try quake3 benchmarks that would be great, run at lowest res, and lowest detail. Then go into game options, switch simple items on, and everything else off.

extract four.dm_66 from \baseq3\pak6.pak and put in \baseq3\demos as four.dm_67

then at q3 console:

/timedemo 1
/demo four

I got 310 fps on a p4 1.6A @ 2.48 ghz ( 155 fsb ). I had to run at 640 x 480 instead of 512 x 384 though, because the monitor won't run at 512.

I think the best ddr around is samsung pc2700, ( any module that uses samsung pc2700 chips should be good ). In reviews, it has been run at 200/400 at fastest settings, and 220/440 at slower settings. I tried to dig up a review on google, but I can't find it.

Matt Simis

Hi Gerry,


Ill benchmark it tonight. Why extract the file from the PAK file and then run the benchmark? Does the extraction process "ingame" really effect performance?

For all you AMD fans, Ill be getting a Tbred when Komplett get stock, and the system right beside mine will then have a Tbred and same gfx card, GF4 Ti 4200. Should make a good comparison.

Im very interested in how the TBred fairs.


Matt

JustHalf

You have to extract it because otherwise Quake III can't find it

SickBoy

Got 357 FPS @ 640x480x16 with the demo Gerry recomended. System is;
P4 1.7 willy not overclocked
GeForce 3 64Meg
1GB PC800 RDRAM
ABit TH7II
Command line was d:\quake3\quake3.exe +set s_initsound 0

Gerry

right well then everyone else should run with s_initsound 0 as well, because it does make a big difference.

PPC

It keeps bombing me out to the main screen.
Also is the demo ment to show up in the demos section?

Gerry

did you follow all those instructions?

Is it version 1.31 of quake3? This shows in bottom right of console.

PPC

Yip i tried that and its the newest version. Its 1.31 i think.
I had problems running Q3:Fortress so i had to change the hunk size to 256. I haven't tried benching after that.
That also fixed my problem on Q3:Sidral.
Must try it when i get home.
Spent ages just trying to get it to run some of them mods.
You can timedemo on HL aswell.
should we try that aswell maybe?

JustHalf

You have to open up the .pak file with WinZip (or similar program) in order to extract the file.

Gerry

HL timedemo is not really up to much, it doesn't scale well and is not widely used.

PPC

Yeah i extracted it but it still didn't work.
I'll try the hunk size allocation on it.

CodeMonkey

Originally posted by Gerry
You get a 33% overclock without putting anything out of spec on any of the current p4 boards, and with the i845 chipset, you can keep the pci at sensible speeds at crazy fsb's.

The sis 645 chipset doesn't appear to have a 1/5 pci divider, but the machine still ran fine at 40mhz pci with a realtek nic ( which are normally dodgy, but perhaps this was more tolerant of fsb because it was integrated, I dunno).

Oh, you might be able to help me on this then. I am assuming you are referring to the MSI 645E MAX2-L mobo here. Is the integrated LAN 10mbit or 100mbit? It doesn't say in any of the reviews I've read and didn't see any info on it on the MSI site. I am looking into getting that mobo with a p4-1.8gz northwood when they have the mobo in stock again.

JustHalf

I believe he was using an Abit board. But that onboard LAN will be 100Mbit.

TacT

the benchmarks in sandra are good apart from the hdd test, if you run the sandra hdd analysis then run it again, it gets slower every time it runs because information is getting thrown around the place. Although if you set a fixed pagefile size then use something like norton speed disk it will set the pagefile towards the middle of the disk making your access times faster and will give you an idea of what the performance is like. The only way to get accurate hd benchmarks from sandra is to defrag before testing, which is a waste of time indeed.

I usually use hd-tach myself although I know it's becoming slightly outdated atm...

I prefer the intel chips for stability and maximum compatability.
I like the athlon for it's cheap price and the performance it delivers for the price.

I've always favoured intel and they've left amd in the dust with their latest 2.53Ghz with Quad pumped fsb, not to mention the POSSIBLITY of using rambus for insane memory bandwidth (although with all the lawsuit games they are playing atm it looks like rambus is off down the toilet) and it's extremely unstable.

I praise amd for what they've done, imo they've given intel the kick in the butt that's made them come out with their new lines of cpu's I mean lets face it, they wouldn't be so rushed to speed up if no one else was there to play catch now would they

I have two systems, one is an athlon k7-750
The other is a pIII 1.0Ghz
The other I won't mention because it a p133mhz

They're both great performers and do all I need them to do the only reason I would choose amd over an intel now was if I couldn't in any way on earth get an extra 100 euro to spend on the quality that intel has always insisted upon. With my K7 overclocked to 108fsb and ram timings tweaked down I can get damn close the same performance in terms of memory bandwidth of a pIII 1Ghz on 100fsb. It surpasses the K7 at 1.2Ghz with cl3 ram in Sandra.