Has anyone seen benchmarks of A64 w/1MB cache vs. Pentium Ms?

[Mike01]

I've looked at a lot of online benchmarks and the concensus seems to be that Pentium Ms slightly outperform A64s at the same clockspeed in gaming and some rendering applications because of their 2MB of L2 cache compared to the older A64's 512KB cache.

Some of the newer A64s, however, have 1MB of L2 cache (such as the new mobile A64 3700+), and I was wondering if anyone had benchmarked this CPU in comparison to Pentium Ms.

 

[uOpt]

The bigger cache doesn't do much on the A64.
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2382&p=7

 

[Mike01]

I don't get it...

These benchmarks that you've linked to show me about what I expected to see (before)...that while the Pentium M is a great clock-for-clock performer, it can't keep up with most of the higher clocked A64s. However...

There's a slew of benchmarks (such as Game PC) that show the Pentium Ms significantly outperforming A64s.

It's like living in two alternate universes.

What's going on, and if someone is lying, why?

Also, these are all desktop benchmarks, in which the Pentium M is paired with a desktop chipset that optimizes its performance...are there any laptop benchmarks?

I have SysMark's Performance Test, which lets you benchmark your PC and then upload it to an online database to compare with other users.

I've downloaded some Pentium M 2.0GHz benchmark and both my old 2.2GHz A64 3400 and my current 2.4GHz 3700 spanked them by a large margin in all CPU benchmarks.

I'm confused.

 

[Mike01]

Just downloaded some recent PassMark baselines (uploaded by users) of 2.13GHz Pentium M CPU laptops and ran the CPU and memory tests:

Toshiba Tecra S2 2.13GHz Pentium M CPU Mark: 354.4
Sony VGN-A39GP 2.13GHz Pentium M CPU Mark: 448.4
Gateway 7426GX A64 2.4GHz CPU Mark: 487.1

It appears that the Toshiba had the CPU clocked dynamically, while the Sony and my Gateway ran the tests at full clock...there is some small performance loss when running a dynamic clock, but not enough to explain the difference. Maybe a bad bridge?


Toshiba Tecra S2 2.13GHz Pentium M Memory Mark: 393.1
Sony VGN-A39GP 2.13GHz Pentium M Memory Mark: 396.1
Gateway 7426GX A64 2.4GHz Memory Mark: 448.6

The Gateway uses PC2700 single channel memory. The test measure read cached and uncacheed speed, write speed, allocate small block speed.

Also, the Sony has an X600, and the Toshiba has a GeForce go 6600 (according to information filled out by the owners of the benchmarked machines). While the Passmarks graphics benchmarks are no where near as exhaustive as Aquamark or similar tests (it just rotates 3D spheres on screen and measures fps), it is nonetheless a good test of raw performance:

Toshiba Tecra S2 2.13GHz Pentium M 3D Graphics Mark: 552.1
Sony VGN-A39GP 2.13GHz Pentium M 3D Graphics Mark: 731.7.1
Gateway 7426GX A64 2.4GHz 3D Graphics Mark stock clock: 665.9
Gateway 7426GX A64 2.4GHz 3D Graphics Mark 415/215: 763.2

Again, these tests measure pure GPU processing power under simple instructions, not complex DX9 tasks. I'd always thought the 6600 go was much more powerull than the 9600/9700, but I guess I was wrong...unless it too can be overclocked like crazy. Also, I'm running Omega drivers, so that might account for some small performance gain.

 

[fbrdphreak]

A64 w/1GB cache

:Q:Q:Q:Q:Q:Q:Q

First, I think the GamePC review you're thinking of (the one w/Turion vs P-M) is a bit screwed up. There shouldn't be that much of a performance advantage w/the P-M and certainly not over such a wide range of areas; gaming for one. While it was interesting, it reeks of BS.

Second, overall the A64's are gonna be stronger general chips. They are pretty much just as efficient as the P-M's in terms of IPC, but of course they are clocked higher. A64's have the integrated memory controller, which helped in efficiency over the regular DDR P-M platform but that should be a moot point w/DDR2 & 533MHz FSB (assuming RAM speed is matched w/FSB).

Basically the A64 will in general be faster at most things, including games, but that is mainly due to its higher clock speed. Clock for clock, the two chips are VERY competitive and have strengths in different areas.

I would rely on AT's benchmarks for performance, others are going to be less reliable.

 

[railer]

Here's my i9300 p m 1.86:


cpu : 371
mem: 322 (2x512 dual channel...why so low?)
3d: 949

mem at 346 after reboot....still low....
cpu up to 395
gpu 1002

 

[Mike01]

fbrdphreak,

How did you get that 1GB quote? It was a typo I fixed a few minutes after making the post! You have strange powers!

Do you have any idea how or why these sites fake their benchmarks? Or are they just doing something wrong?

railer,

Memory speed, in my experience with overclocking, seems to depend more on CPU clock speed than memory clock speed. So if you're running your Pentium M with a dynamic clock, the memory performance would suffer.

I don't know if Pentium Ms can be run at full clock, but if they can, try doing so and run the tests again.

For exampe, my memory mark is 446 at full clock, 443 at dynamic clock and 151 at 33% clock. 151!

That GPU score is sick!

 

[uOpt]

Originally posted by: Mike01
I don't get it...

These benchmarks that you've linked to show me about what I expected to see (before)...that while the Pentium M is a great clock-for-clock performer, it can't keep up with most of the higher clocked A64s. However...

There's a slew of benchmarks (such as Game PC) that show the Pentium Ms significantly outperforming A64s.

It's like living in two alternate universes.

What's going on, and if someone is lying, why?

Mobile AMD64 only have 333 MHz FSB and no dual channel, that is one difference. But it probably isn't a big one as the benchmarks show dual-channel to be pretty much irrelevant.

The Pentium-M benchmarks on desktops do have faster RAM, but since they have a constant FSB that should do close to nothing.

For my quick-n-dirty testing (Linux development stuff where the AMD64 is not that overwhelming, C compilation, video encoding) the Pentium-M with 1 MB cache is about 25% faster per clockspeed than an AMD64 with 512 KB cache. But this is stretching it, I only have a 1.3 GHz Pentium-M and don't know what really happens with the faster Pentium-Ms.

 

[fbrdphreak]

Originally posted by: Mike01
fbrdphreak,

How did you get that 1GB quote? It was a typo I fixed a few minutes after making the post! You have strange powers!

Do you have any idea how or why these sites fake their benchmarks? Or are they just doing something wrong?

Its still in the title of the post edit your original post and fix the title too

In regards to the benchies, I'm not entirely sure. It just seems odd that the benchmarks are THAT skewed when in the past benchies fo P-M vs A64 (Turion is basically a low power A64) have shown a different result than what those people got.

We hope to have a Turion notebook in the labs soon, so maybe we'll have a definitive answer for your then.

 

[Mike01]

the Pentium-M with 1 MB cache is about 25% faster per clockspeed than an AMD64 with 512 KB cache.

This is what I'm talking about...the benchmarks on Tom's Hardware Guide and Anand Tech don't support that.

To be fair, you did say quick and dirty.

 

[fbrdphreak]

I would say that is mostly accurate, the 25% faster part, but you've got to reality (A) that A64's are clocked higher in general and (B) are much cheaper. So while clock for clock the P-M may be faster, the A64 can scale higher obviously. Pick your poison

 

[Mike01]

If P-Ms were 25% faster clock for clock, then a 2.13GHz P-M would be faster than a 2.6GHz A64, and it's not. Not according to Anandtech benchmarks, which show it to below most of the 2.4GHz A64s in all benchmarks. Ditto with Tom's Hardware Guide.

Maybe 10% faster, which puts it just over 2.3, or a hair below the A64 2.4, which is pretty much where it scores.

At least in desktop benchmarks...I have yet to see a Pentium M laptop score anywhere near a higher clocked A64 in any benchmark.

The PassMark benchmarks submitted in this topic are but one example.

 

[fbrdphreak]

10% sounds about right to me

I think if you were to take two laptops with identical video cards & HDD's, one Sonoma & one A64 you would have a decent match there. Most of the time when you see a Centrino laptop benchied against an A64 laptop, you are comparing a basic portable machine against a powerful DTR. Drop in 7200RPM drives, 1GB RAM (dual channel for the Sonoma laptop), and top-clocked processors, you'll see some close matches

 

[Mike01]

Perhaps. The PassMark benchmarks do suggest that the 2.13GHz P-M (CPU Mark 448) is almost exactly the same speed as an A64 3400 (2.2GHz, CPU Mark 447), and not that far behind the 3700 (2.4GHz, CPU Mark 487).

However, the price/performance issues is clear. A 2.13GHz P-M adds $300 (over a 2GHz) to the already pricy XPS 2 and doesn't even seem to be available on the 9300 (though a 2GHz adds $400!!). Meanwhile, the A64 3700+ is more powerful (though maybe not by that much) and costs a lot less.

And...I'd still really like to know where GamePC's phony benchmarks come from. Are they a Pentium M reseller or something?

 

[fbrdphreak]

You are right, the A64 is definitely the best bang for the buck; but for battery life there is only one choice

I'm not sure what to tell you about GamePC's results. But after looking over them again, there are definite issues:
AT's P-M Gaming Benchies
GamePC's P-M vs Turion vs A64 Gaming Benchies

In AT's benchies, a 2GHz P-M (400MHz FSB) scores almost the exact same in Halo as an A64 3200+ Socket 939. In GamePC's benchies, a 1.86GHz P-M scores WAY above a 2GHz A64.
This is also interesting. GamePC claims that they are using the P-M 770 (2.13GHz, 533MHz FSB) and other 533Mhz FSB chips. Excep they're using the DFI 855 chipsets, which only officially support 400MHz FSB. I'm sure you could OC or mod the board to run @ 533MHz FSB, but it is obviously not a Sonoma platform!

Regardless, their benchies are full of holes. Performance of Turion vs P-M shouldn't be judged until a scientific and fair benchmark can be done. It does seem that this place is primarily a reseller with an "In the Labs" section for reviews/articles. A reseller writing product reviews....hmm..........

 

[Mike01]

That does explain it all. Thank you. Lying bastards!


You know that the only reason I am obsessed with the power of my 3700 is because I have battery life envy.

 

[uOpt]

Originally posted by: Mike01

the Pentium-M with 1 MB cache is about 25% faster per clockspeed than an AMD64 with 512 KB cache.

This is what I'm talking about...the benchmarks on Tom's Hardware Guide and Anand Tech don't support that.

To be fair, you did say quick and dirty.

Well, the other problem is that I only have a 1.3 GHz Pentium-M and interpolated from there.

That one will have plenty of memory bandwidth for the amount of hunger it can develop.

If you put a 2.0 Ghz Pentium-M on the same bandwidth you can't expect it to scale like the clockspeed.

BTW< anybody knows what the max Pentium-M is that I can put into a Thinkpad R-40?

 

[fbrdphreak]

Originally posted by: MartinCracauer
BTW< anybody knows what the max Pentium-M is that I can put into a Thinkpad R-40?

My guess would be either a 1.5GHz Banias or MAYBE a Dothan (up to 2.1GHz) with a BIOS update. Head over to http://forum.thinkpads.com and post in the R series forum there

 

[RobsTV]

Here are some fresh tests of desktop comparison:

At stock speeds, the Pentium M 755 did very well in our benchmarks, outrunning some of the lower-end Athlon 64s and Pentium 4s in the majority of the tests. The real power behind the Pentium M comes when overclocked though; and the Asus CT-479 enabled us to go 200 MHz faster than the the 855GME platforms had for a total speed of 2.6 GHz. At that level we saw the P-M out performing the Athlon 64 FX-55 processor in gaming and the Pentium 4 in some media tests.

http://www.pcper.com/article.php?aid=133&type=expert&pid=16

The simple overclocking setup for laptops that many are now doing is running the 400MHz Dothan in a 533MHz laptop.
End results overclock a 1.8GHz to 2.4Ghz in a laptop such as Dell i9300..

 

[Mike01]

Wouldn't a fair benchmark of an overclocked Pentium M be to compare it to overclocked A64s and Pentium 4s??