Does anyone know how many IPC a Pentium M has?

Crescent13

Diamond Member
Jan 12, 2005
4,793
1
0
I want to compare it to an AMD 64. This is what i'm thinking

AMD 64 = 9 ipc
Pentium M = x ipc

AMD 64 clock speed times 9 divided by "x" equals the equivilant speed of a pentium M.

I would like to find out what x is.
 

aka1nas

Diamond Member
Aug 30, 2001
4,335
1
0
It isn't quite that simple. It depends on what kind of instructions the chip is running. AFAIK, I think the A64 still beats it in FPU-intensive apps, while the Pentium M beats the A64 clock-for-clock in other things.
 

Lonyo

Lifer
Aug 10, 2002
21,938
6
81
It's pretty much the same as an AMD64.
ie: 8.75~9.25 in real terms. Depends on the application, but they are pretty much exactly the same in terms of clock for clock performance.
 

BitByBit

Senior member
Jan 2, 2005
474
2
81
The Athlon's maximum IPC is 3, since it has 3 integer units and 3 floating-point units, with 3 decoders to feed them. However, maintaining this maximum is not always possible, especially when instructions have to be fetched from the L2 or even worse, the memory.
Dothan has 2 integer units, but thanks to its very large, very low latency L2, along with it's superior branch prediction abilites, it acheives an average IPC much closer to its theoretical maximum.

So basically, the Athlon's theoretical maximum is higher than that of Dothan, but Dothan is performing closer to its maximum more often than the Athlon.
Also, Dothan's technologies, such as Micro-ops fusion for instance, serve to raise its IPC.
 

BitByBit

Senior member
Jan 2, 2005
474
2
81
How can it be 12 when Dothan can only decode a little over an average of 2 instructions per clock?
 

Sentential

Senior member
Feb 28, 2005
677
0
0
Originally posted by: BitByBit
How can it be 12 when Dothan can only decode a little over an average of 2 instructions per clock?

Dothan wasnt designed to do such. If you are talking games and single applications 12 starts to look pretty accurate. Along with 3D benches they also put off an accurate SuperPI time to prove their speed.

I guess Intel just didnt design them for encoding (hence the lack of SSE3) and other workstation suits to keep their P4s top dog. Also note the lack of RAM bandwith on Dothan. Very few guys, even on Sonoma cores, are getting the FSB much past DDR400
 

BitByBit

Senior member
Jan 2, 2005
474
2
81
I'm not talking about encoding performance here.
An instruction must be decoded before it can be executed by the execution units.
Dothan, like K7/K8, has 3 decoders meaning it can decode a maximum of 3 instructions per clock, though in reality this is closer to 2.
 

Lonyo

Lifer
Aug 10, 2002
21,938
6
81
Originally posted by: BitByBit
I'm not talking about encoding performance here.
An instruction must be decoded before it can be executed by the execution units.
Dothan, like K7/K8, has 3 decoders meaning it can decode a maximum of 3 instructions per clock, though in reality this is closer to 2.

We're doing comparatives here, really.
P4 does 6 things, AMD64 does 9 (or 1.5 times as many) and Dothan does say 12 things (in gaming situations).
Obviously there aren't really real numbers which can be specific, but just comparisonas can be made, ie: AMD64's do 1.5 times as much work as P4's per clock.
 

carlosd

Senior member
Aug 3, 2004
782
0
0
Originally posted by: Sentential
I *believe* it is 12. My friend has a Dothan setup and his at mid 2.6 can cripple a FX55 in the 3.4ghz range.

Picture proof is here:

http://home.comcast.net/~gautamb/cpucomparo.jpg

OK , meaningless synthetic benchmarks. In real APPS dothan won't cripple an FX55 at 3400 GHZ , specially if we talk about FPU intensive apps, the dothan performance will be laughable.
 

Sentential

Senior member
Feb 28, 2005
677
0
0
Originally posted by: carlosd
Originally posted by: Sentential
I *believe* it is 12. My friend has a Dothan setup and his at mid 2.6 can cripple a FX55 in the 3.4ghz range.

Picture proof is here:

http://home.comcast.net/~gautamb/cpucomparo.jpg

OK , meaningless synthetic benchmarks. In real APPS dothan won't cripple an FX55 at 3400 GHZ , specially if we talk about FPU intensive apps, the dothan performance will be laughable.

Name one "real app"... btw is this "real" enough for you?

http://home.comcast.net/~gautamb/2745pi.JPG
 

Lonyo

Lifer
Aug 10, 2002
21,938
6
81
Originally posted by: Sentential
Originally posted by: carlosd
Originally posted by: Sentential
I *believe* it is 12. My friend has a Dothan setup and his at mid 2.6 can cripple a FX55 in the 3.4ghz range.

Picture proof is here:

http://home.comcast.net/~gautamb/cpucomparo.jpg

OK , meaningless synthetic benchmarks. In real APPS dothan won't cripple an FX55 at 3400 GHZ , specially if we talk about FPU intensive apps, the dothan performance will be laughable.

Name one "real app"... btw is this "real" enough for you?

http://home.comcast.net/~gautamb/2745pi.JPG

How is that a real app? That's a synthetic benchmark, unless you need Pi to 1m places?
Real app is something like gaming (a real game, not a benchmark such as 3D Mark), encoding, RARing etc.
 

bunnyfubbles

Lifer
Sep 3, 2001
12,248
3
0
Originally posted by: Sentential
Originally posted by: carlosd
Originally posted by: Sentential
I *believe* it is 12. My friend has a Dothan setup and his at mid 2.6 can cripple a FX55 in the 3.4ghz range.

Picture proof is here:

http://home.comcast.net/~gautamb/cpucomparo.jpg

OK , meaningless synthetic benchmarks. In real APPS dothan won't cripple an FX55 at 3400 GHZ , specially if we talk about FPU intensive apps, the dothan performance will be laughable.

Name one "real app"... btw is this "real" enough for you?

http://home.comcast.net/~gautamb/2745pi.JPG

No, do you calculate pi for a living?

Give us benches of P-M playing real games, encoding real videos and music, archiving real files, hopefully you get the picture.

Until intel fixes the FPU performance on the Pentium-M it will only be a neat toy in the destop world of CPUs, it simply cannot be a complete processor and until then it'll only be the best processor for laptops.
 

Gautam B

Junior Member
May 16, 2005
13
0
0
Hi folks, those results are mine. I believe that the Dothan has a pipeline stage length of either 14 or 15, compared to the Athlon64's 12. It's raw IPC, as a consequence, is actually lower than the Athlon64. It's performance in raw integer calculation, as tested by the Sandra Arithmetic test and CPU Renderbench suffer as a result. In the Sandra test, it slots between an AthlonXP and Athlon64 at the same clock speeds.

However, not only is IPC an extremely hard entity to gauge, it isn't even effective for what you're looking for. In the real-world applications, the Dothan does lay some smack on the Athlon64. In 3D benchmarks, it's way ahead. Feels a little faster in actual gaming as well. I say this as being the former owner of a 2.7GHz A64. The Aquamark3 result, I don't know what to make of. No, I don't personally believe that the 2.67 gig Dothan is faster than a 3.4 gig FX55...but it was a fun run anyways. I got increases of different magnitudes in all 3D benchmarks over the A64, and this just happened to be the largest one. To grossly generalize, I would say that the Dothan is about equivalent to an A64 San Diego/FX clocked about 200-300MHz higher in 3D applications.

Hope this clears things up. I won't be able to run anymore tests for a week or two, but when I get the chance to, I'll give anything you guys might find interesting a whirl.
 

Sentential

Senior member
Feb 28, 2005
677
0
0
And AM3 and Futuremark arent games? How would running a timedemo on Doom3 or the like be different from what I already hosted?

I can understand the encoding peformance of Dothan. AFAIK thats primarly because of their lack of RAM bandwith.

Either way I do agree. For the moment its nothing more than a toy. A good one however
 

Gamingphreek

Lifer
Mar 31, 2003
11,679
0
81
No one knows how many IPC (Instructions per clock) the DOthan is. We do not know how many stages the pipeline is either. Intel will not disclose this information. However as said it is estimated to be slightly less efficient than AMD's.

-Kevin
 

Sentential

Senior member
Feb 28, 2005
677
0
0
Originally posted by: Gamingphreek
No one knows how many IPC (Instructions per clock) the DOthan is. We do not know how many stages the pipeline is either. Intel will not disclose this information. However as said it is estimated to be slightly less efficient than AMD's.

-Kevin

Probably because they lack a memory controller. As for other benches the best I could find was one on Banias which is an older 1Meg Dothan.

http://www.anandtech.com/mobile/showdoc.aspx?i=1800&p=13

Still tho, even with 1MB of cache its about as fast as a 2.6ghz northie at 1.6ghz
 

bunnyfubbles

Lifer
Sep 3, 2001
12,248
3
0
Originally posted by: SententialAnd AM3 and Futuremark arent games? How would running a timedemo on Doom3 or the like be different from what I already hosted?
Because you can't play those "games". We learn from history, its why we even bother with it, and history will tell us that there are many instances where synthetic benchmarks are good only to check whether or not your particular setup is behaving properly - ie, to check whether or not your 3GHz P4 with 1GB of PC3200 and GeForce 6800 NU are performing properly. Other than that there could be a synthetic bench where the P4 system performs faster than a 3.4GHz Athlon 64 FX with an ATI Radeon X850XTPE. Certainly you wouldn't assume the P4 system would be faster in every case. The worst case scenario for the AMD system is that it is slower in a synthetic benchmark you can't even play and enjoy like you can with a real video game.

I can understand the encoding peformance of Dothan. AFAIK thats primarly because of their lack of RAM bandwith.
A CPU needs to be starved of bandwidth to suffer from it (and the Dothan is far from as bandwidth hungry as the P4), the Dothan wouldn't see a magical increase in performance from increased bandwidth, it'd just be generally faster. Its the fact that intel didn't outright design the processor to be good all around, it performs very well for how you'd use it in a laptop. Like I said, intel needs to fix the inefficiencies for it to be a true desktop competitor, however once this is done, it will most likely no longer be as perfect for laptops as it would be for desktops, trading some power and heat for performance.

Either way I do agree. For the moment its nothing more than a toy. A good one however
While I sure as hell would love to have one for a gaming rig, it simply isn't cost effective. Of course then there's the fact you can't have one in an SLI rig and right then and there you automatically lose that one decent advantage. Dual core will only further the gap until we see a dual core PM that we can use on the desktop...


The PM has been nice for laptops, it COULD be good for desktops, if anyone is dropping the ball, it is Intel. Don't praise them for creating a product that is worthless to anyone except laptop users.

Although I seriously doubt Apple wants intel for the P4 line. I actually watched the Keynote and got the impression Apple is far more interested in future Intel stuff involving the PM. Heck, 3 of their 4 major computer product would obviously (painfully obvious) benefit from the P-M more so than the P4/PD - the iMac, Mac Mini, and iBook/PowerBook. Only the Power Mac would be a suitable application for the high power and heat P4/Xeon/PD
 

Gautam B

Junior Member
May 16, 2005
13
0
0
Reading some of this stuff just makes me want to kill Intel's management.

As we mentioned before, Timna was to feature an integrated memory controller, but at the time it was on the drawing board Intel still had delusions of grandeur about RDRAM being the future of all memory technology by 2001.
They had a potential A64 killer about 3 years...why didn't it see the light of day? Because of RDRAM...[rolls eyes]

You are correct about the Dothan not benefitting from greater memory bandwidth. It interestingly reacts very similarly to the A64 as memory speed is increased (read, not very much at all).

As far as the IPC goes, what I find most interesting in this article is the discussion of micro-ops fusion. Reducing the usage of the pipeline serves as a pretty good explanation of why the Dothan/Banias is as efficient as it is.

I bought it as a toy, but it has been serving me very well as a daily rig, at least as well as the A64, if not more. Rocky to get set up, but past that, really a wonderful setup.

Indeed its Intel that's dropping the ball. Once they get this architecture into a proper form that the end-user can user (if they ever will), I strongly believe that not only will they give A64 a run for its money, but will very explicitly gain an upper hand.

 

PlatinumGold

Lifer
Aug 11, 2000
23,168
0
71
Originally posted by: Gautam B
Reading some of this stuff just makes me want to kill Intel's management.

As we mentioned before, Timna was to feature an integrated memory controller, but at the time it was on the drawing board Intel still had delusions of grandeur about RDRAM being the future of all memory technology by 2001.
They had a potential A64 killer about 3 years...why didn't it see the light of day? Because of RDRAM...[rolls eyes]

You are correct about the Dothan not benefitting from greater memory bandwidth. It interestingly reacts very similarly to the A64 as memory speed is increased (read, not very much at all).

As far as the IPC goes, what I find most interesting in this article is the discussion of micro-ops fusion. Reducing the usage of the pipeline serves as a pretty good explanation of why the Dothan/Banias is as efficient as it is.

I bought it as a toy, but it has been serving me very well as a daily rig, at least as well as the A64, if not more. Rocky to get set up, but past that, really a wonderful setup.

Indeed its Intel that's dropping the ball. Once they get this architecture into a proper form that the end-user can user (if they ever will), I strongly believe that not only will they give A64 a run for its money, but will very explicitly gain an upper hand.

in the 32 bit arena, we still have no clue how their 64 bit iteration will perform, whereas we already know how the a64 performs in 64 bit.

again, i posted this somewhere else, but intel dropped the ball going from the p3 to the p4 platform. their panic that amd got to 1 ghz before them has hurt all of us.
 

Fox5

Diamond Member
Jan 31, 2005
5,957
7
81
Originally posted by: Crescent13
I want to compare it to an AMD 64. This is what i'm thinking

AMD 64 = 9 ipc
Pentium M = x ipc

AMD 64 clock speed times 9 divided by "x" equals the equivilant speed of a pentium M.

I would like to find out what x is.

AMD does not achieve anywhere near 9 ipc, but I believe the P-M would be 3 ipc theoretically(it's the same as the pentium 3 isn't it?). However, I'd say the P-M probably has by far the highest efficiency of any current microprocessor.
Athlon 64 and P4 are probably at around 10%, P-M around 30%, and intel plans to beef up the P-m's execution units to be on par with athlons.

BTW, Dothan may not suffer from a lack of bandwidth, but I bet it loves lower latencies. Faster memory/fsb means the latency is lower.
 

miketheidiot

Lifer
Sep 3, 2004
11,060
1
0
i believe p-m has the same number of decode units as a64. I think it may be lacking an FPU unit or two (k7/k8 have 3 fpu and 3 integer if i recall) While a64 benefits from having lots of low latency memory bandwidth and dedicated HT link to the chipset, P-m benefits from a better branch predictor and a large very low latency l2 cache. I expect the next generation for each will improve on each processors weaknesses.

EDIT: i can't tpye ;)
 

zephyrprime

Diamond Member
Feb 18, 2001
7,512
2
81
Numbers like 9 and 10 are way too high. 2 or three is probably the real world answer. 3 may be the theoretical dispatch limit. It might be possible to have more temporarily if there are some instructions in the pipe that have a latency of greater than 1 cycle but I could be wrong about that. Even if I were right, the temporary increase in IPC would not be indicative of performance because an instruction that has high latency is by definition slow.
 

IamTHEsnake

Senior member
Feb 4, 2004
334
0
0
If the a64 and the dothan were matched in clock speed the dothan would win, in games of course. I have no proof handy, but I have seen this on many occasions on different webistes.
So at the very least we know that pentium M IPCs > Athlon 64 IPCs
Note: I don't know how many either of the processors has exactly.