> Dang. Don't they sell Pentium-M setups in desktops? heh.
>
> Jimmy

No, But I've read of an adapter board out there that lets you plug one into a desktop...P4 form factor as I recall...have to re-look that up.

Pentium-M is really Intel's "...oh yea...we shoulda done that on Pentium III" chip. Better instruction queues, fewer misrpredictions, better memory access.

The various AMD chip cores really differ quite a bit...
K7.....AppleBred, Morgan
K7-XP..Palomino, Thouroughbred, Barton, Thorton
K8...Clawhammer, Sledgehammer, Newcastle, Winchester

So while it may call itself an "Athlon XP", or "Athlon 64" the internal engine may be quite different. So to be fair...a comparison would have to use the same motherboard+RAM setup and compare different cores of the same CPU family (Athlon, Athlon XP, Athlon 64).

only for Info

my P4 3.2 with HT does a Seti WU in

2:35 for two WUS

that means one Wu in 77 Minutes for one WU, beat this

....

Greetings from Germany NRW
Ulli S@h Berkeley's Staff Friends Club m7 ©

> Yeah, but the benchmark doesn't say anything about how much credit you'll get.
> As I understand it, it's just an estimate of how long it'll take to finish a
> WU. Besides, the benchmark differs, depending on which version of BOINC
> you're using. v4.19 yields higher numbers than 4.25 (or later), due to a
> "bug". (Actually a design flaw caused the compiler to optimize the benchmark
> loop too much...)

Actually, the benchmarks have everything to do with how much credit you'll request. This is why you get more credit running 4.19 due to the inflated benchmarks.

---

> > Don't they sell Pentium-M setups in desktops?
>
> Yes, more and more do just that. Also take a look <a> href="http://www.asus.com/products4.aspx?l1=3&l2=54&l3=0&model=467&modelmenu=1">here[/url].
>
> UPDATED: Here are a couple of links to Pentium-M desktop motherboards:
>
> AOpen
> i855GMEm-LFS
> <a> href="http://www.dfi.com.tw/Product/xx_product_spec_details_r_us.jsp?PRODUCT_ID=3350&CATEGORY_TYPE=MB&SITE=US">DFI
> 855GME-MGF[/url]
>
>
And here's a newer model supports fsb of 533 and dual channel RAM

AOpen i915GMm-HFS

> Actually, the benchmarks have everything to do with how much credit
> you'll request. This is why you get more credit running 4.19 due to the
> inflated benchmarks.

You can claim all you like- it doesn't mean you'll get it unless the other results & claimed credit used to grant credit match yours.

---

Grant
Darwin NT

Have a look at this article from a German PC magazine: http://www.pcwelt.de/news/hardware/109635/index.html
Use babelfish to translate, or just have a look at the diagrams on the subpages...

---

> only for Info
>
> my P4 3.2 with HT does a Seti WU in
>
> 2:35 for two WUS
>
> that means one Wu in 77 Minutes for one WU, beat this
>

Sir Ulli,

Have you tried to force your P4 to do 1 WU at a time? (In other words, only 1 seti process).

I am curious how long it would take to do 1 and only 1 WU... If it's also 77 minutes or there abouts, I'd really like to know...

Thanks for your input.

Jimmy

---

5400 seconds! Nice job.

I'd like to add to this, since I happen to have an excellent comparative situation here, I think:

Home: 1.3 ghz P4 Rambus 384mb desktop ( link to stats
Work: 1.3 ghz P4 M 256mb laptop ( link to stats )

Look at the scores! Double as fast on my laptop for both benchmarks, same processor speed, cache, and comparable swapfile space, and actually MORE ram available on my desktop. Plus isn't the rambus architecture supposed to be "double" the memory bus/bandwidth? (doesn't quite understand, except that now Rambus sucks compared to DDR and people R laff at me for my desktop).

Anyways, I'm not sure what CPU my home desktop is. I've been told maybe a Wilhamette? But this is certainly a surprising difference (even with several MS office progs open + 2 other progs running on my laptop).

---

> Look at the scores! Double as fast on my laptop for both benchmarks, same
> processor speed, cache, and comparable swapfile space, and actually MORE ram
> available on my desktop. Plus isn't the rambus architecture supposed to be
> "double" the memory bus/bandwidth? (doesn't quite understand, except that now
> Rambus sucks compared to DDR and people R laff at me for my desktop).

RAMBUS was supposed to be better, which is why there was a premium paid for the licencing. Real world experience showed that it really was not as fast and therefore the premium price paid was not worth it.

There are more factors involved that are harder to see and do not show up in the benchmark scores. One of the primary ones is the presence or abscence of dual channel memory. If you look down by the 3.2 GHz statistics you can see that there is almost a one hour difference in average processing times for my systems caused by the lack of dual channel memory (and possibly other features of a higher quality motherboard).

So, other system factors do play a large role in the actual performance of the systems where a quick glance seems to indicate parity. The actual perrformance of the systems is the only true "benchmark" that counts which is why I play wet blanket when people post the benchmark scores. We use these scores for a rough guess as to performance and to give an initial size for work capacity. Afterwards we use it for the claim for credit, mostly because it truly is the "best" way to derive that value.

If you look at my lecture notes and the Glossary entries for Benchmark you can have (I hope) a better understanding of this topic.

In the early days of the Beta we still had hopes we could measure system memory speed, but that too was unsuccessful and was later dropped.

Last comment on system speed. There is another common misconception that lumping a large number of PCs in a pile automatically will give you similar performance of a mainframe. Systems like BOINC are pointed to as "proof" that this is a trueism. Unfortunately this is another counter intuitive (man I have a string of these this week) concept where the use of a loosely-coupled network of computers to facilitate computation is used and compared to a tightly coupled processing system. Though many problems are solvable using loosely-coupled networks of machines, not all problems are amenable to that architecture.

In fact, some web sites that were initially established using several thousand PCs have been moved onto mainframes to achieve better scalability, control, performance, reliability and lower cost of operation. In many of my talks, including this one, I would point at the PC and point out that the design of the PC does not do server tasks well for the simple reason that the ability of the system to support significant I/O is hampered by the current design (also a feature of high end systems such as Mainframes is the bandwidth of the available I/O channels). What is so surprising is that it does do as well as it does even though the design is the "single user-centric" mind-set.

I will probably get howls about this, but I will say it anyway ... the current crop of Microsoft OS shows the single-user mind-set in such things as "focus-stealing" and its inability to share many of the core system functions.

> Anyways, I'm not sure what CPU my home desktop is. I've been told maybe a
> Wilhamette? But this is certainly a surprising difference (even with several
> MS office progs open + 2 other progs running on my laptop).

To discover this you can get any one of several CPU "sniffers" that will tell you all sorts of things. I usually use CPU-z, but there are others ... I think Intel has one on their site that you also can use ...

---

RAMBUS got killed from the inside.

Pity, really. Was a good idea.

---

> RAMBUS got killed from the inside.
>
> Pity, really. Was a good idea.

Well, yes, mostly because of cost ... at least according to all I read about it ...

The white box sellers did not like it cause it messed up their margins ... profit margins that is ...

---

um... ok... Anyways, here's all the info I could dig up about what type of memory is in my work laptop. 128pin DIMMs, as expected, 256mb but of what type I still don't know.

DUAL IN-LINE MEMORY MODULE, 128, 266, 16X64, 8K, 200, BURN 2

Like I said, desktop is a rambus. I suppose I should be motivated to google Rambus and find out more about it, but I'm just not %).

Paul... I guess I still find it amazing that my last 1.3ghz P4-M laptop WU came off in ~9000 seconds, whereas your 3.2ghz P4 desktop WUs seen to come off in ~10,000 seconds. The thread-starter jimmy seems to be on to something. What is it about the P4-M chip that is crunching faster than chips supposedly 2x better?

---

> um... ok... Anyways, here's all the info I could dig up about what type of
> memory is in my work laptop. 128pin DIMMs, as expected, 256mb but of what
> type I still don't know.
>
> DUAL IN-LINE MEMORY MODULE, 128, 266, 16X64, 8K, 200, BURN 2

Well, with two you have a chance that you may have dual channel memory... the only way to tell is to look up the MB and check the specifications. Also, if there is documentation on the system they may talk to this ... usually in and around the discussions about installing like sized modules.

> Paul... I guess I still find it amazing that my last 1.3ghz P4-M laptop WU
> came off in ~9000 seconds, whereas your 3.2ghz P4 desktop WUs seen to come off
> in ~10,000 seconds. The thread-starter jimmy seems to be on to something.
> What is it about the P4-M chip that is crunching faster than chips supposedly
> 2x better?

Yes, well, we are back to the processing time vs. throughput. My 3.2 GHz machine is doing two WU at a time. So, it takes me 10K seconds to do 2 Work units and you take 18K seconds to do 2 work units. So, yes, the CPU I have is apparently slower ... but in fact does more ...

It is the difference between talking to your mother-in-law and talking to you significant other... one seem FAR longer to endure ...

The difficulties come from the fact that people like to focus on one aspect to get a feel for something's quality. Like 0 to 60 in 3.2 seconds ... but how often do you want to get tickets? And how often do you do that accelleration anyway? So better metrics are MPG, insurance costs, comfort (hard to quantify, but when I had a Lexus LS-400 I rarely had complaints) ...

The internals of the processors make miles of difference. If you scroll down the page you can see the the 2.0 GHz G5 chip does a SETI@Home Work Unit also in about 9K seconds... and yet the clock speed is only 2.0 GHz ...

---

> The difficulties come from the fact that people like to focus on one aspect to
> get a feel for something's quality. Like 0 to 60 in 3.2 seconds ... but how
> often do you want to get tickets? And how often do you do that accelleration
> anyway? So better metrics are MPG, insurance costs, comfort (hard to quantify,
> but when I had a Lexus LS-400 I rarely had complaints) ...

I started this thread off thinking that the Pentium-M chips beat the competition in all aspects:

1. Faster
2. Lower power consumption
3. Least expensive

Over the course of the discussion I have found several setups that run faster. So, I was hoping for the price$$$, the Pentium-M may still beat it out. Turns out, Pentium-M chips are the most expensive of the bunch.

I also pondered WHY the pentium-M chips were faster (I was guessing cuz of the 1MB cache, but I was WRONG). It turns out it's due more to it's CPU architecture than anything else. The Pentium-M's don't have SSE2/3DNow and a whole bunch of other features that enhance gaming. With that extra space, they have an extra ALU & FPU and better prediction circuitry which enhances general computing liking working Excel spreadsheets (and in our case seti).

I found out that communities such as tomshardware and anandtech were uninterested in the Pentium-Ms, due to their lackluster gaming performance.

I wondered why MY dual cored P4 was slower for seti (also running off a laptop). I think I figured that one out. It is due to lack of cache. But it turns out most P4 cpus do sell with more cache than mine has.

Kind of reminds me of the time when our company switched from Dell desktops to HP desktops. What a mistake. A Dell Pentium2 450Mhz, ran faster than any of the HP P3 800Mhz machines!!! (Come to think of it my P4 laptop is an HP)

It's been interesting, however, I think I have all the mysteries solved.

Jimmy

---

I seem to recall that PowerPCs had a better price/perfomance/power dissipation/size ratio than any x86-based machine, but I think you're want to keep to the Intel/AMD platform, right?

---

> I seem to recall that PowerPCs had a better price/perfomance/power
> dissipation/size ratio than any x86-based machine, but I think you're want to
> keep to the Intel/AMD platform, right?
>


My 2 cents: I just got a new laptop. It has a 1.5GHz Celeron M. Its times are 3 hr and below. It'll probably get a little faster when I put in a decent amount of memory. 256MB ram with 32M SMA is way too little for WinXP. Hitting the virtual memory way to much.

---

> I also pondered WHY the pentium-M chips were faster (I was guessing cuz of the
> 1MB cache, but I was WRONG). It turns out it's due more to it's CPU
> architecture than anything else. The Pentium-M's don't have SSE2/3DNow and a
> whole bunch of other features that enhance gaming. With that extra space,
> they have an extra ALU & FPU and better prediction circuitry which
> enhances general computing liking working Excel spreadsheets (and in our case
> seti).

I taught at several universities (night school type) here in Sacramento which has a big Intel presence. Several of my students were working for Intel and once they talked about a P4 version in the works that was based on a P-III architecture and it handily beat the P4's ... it never saw the light of day and all the test mock-ups were destroyed when that development branch was abandoned.

Architecture is a big thing. But it is very hard to quantify easily. AMD chips have advantages over the P4 for many things mostly because of differences in architecture. So, they do things faster. Yet with Hyper-Threading the P4s beat the AMDs in throughput (for the moment) becuase of their architecture which though slower in speed has greater capacity.

> I found out that communities such as tomshardware and anandtech were
> uninterested in the Pentium-Ms, due to their lackluster gaming performance.

Yes, the 0 to 60 crowd ... uninterested in many other aspects of performance ...

> I wondered why MY dual cored P4 was slower for seti (also running off a
> laptop). I think I figured that one out. It is due to lack of cache. But it
> turns out most P4 cpus do sell with more cache than mine has.

Yes, and in some cases at considerably higher cost.

> Kind of reminds me of the time when our company switched from Dell desktops to
> HP desktops. What a mistake. A Dell Pentium2 450Mhz, ran faster than any of
> the HP P3 800Mhz machines!!! (Come to think of it my P4 laptop is an HP)

The difference in thiings like chipset, memory channels, MB design, etc.

---

> I taught at several universities (night school type) here in Sacramento which
> has a big Intel presence. Several of my students were working for Intel and
> once they talked about a P4 version in the works that was based on a P-III
> architecture and it handily beat the P4's ... it never saw the light of day
> and all the test mock-ups were destroyed when that development branch was
> abandoned.

I majored in E.E. in college and took a few classes in CPU design. We studied the MIPS R8000 at the time. Always wanted to get into the industry but never got my foot in the door.

> > I found out that communities such as tomshardware and anandtech were
> > uninterested in the Pentium-Ms, due to their lackluster gaming
> performance.
>
> Yes, the 0 to 60 crowd ... uninterested in many other aspects of performance
> ...
>
> > I wondered why MY dual cored P4 was slower for seti (also running off a
> > laptop). I think I figured that one out. It is due to lack of cache.
> But it
> > turns out most P4 cpus do sell with more cache than mine has.
>
> Yes, and in some cases at considerably higher cost.

>
> The difference in thiings like chipset, memory channels, MB design, etc.
>
>

UPDATE: I have been basing my cache levels on what BOINC reports.

I listed cache levels for Duron/Ahtlon as I knew what they should be. As it turns out, all the cache levels that BOINC reports are wrong except for the Pentium-M processor.... duh.

Note to self: From now on, I shall use CPU-z and not trust what BOINC says, (I shouldn't even trust my own memory).

So, as it turns out my P4 EE, actually does have 2 MB Level 3 cache. 512KB Level 2 Cache, and 8KB level 1 cache. So, it ain't the lack of cache that's slowing my P4 down. It's probably the fact that it's in an HP laptop that's slowing it down. Duh. It even has DDR2 memory, don't know what the heck is wrong with it.

New Summary:

Pentium-M fast => superior architecture (only available at decent prices by buying laptops), (btw, it has DDR memory, not PC100 SDRAM as I had originally thought, ah the wonders of cpu-z). A few guys shared their performance of a Pentium-M 2.0 Ghz machine, and they scaled up nicely indicating that there isn't a lack of cache on the Pentium-M architecture. Sold for ~240-400, Rare.

my P4 was slow for the price paid => Did some tests which would indicate a lack of cache. But CPU-z indicates otherwise. Don't buy from HP, get a good mobo. Some guys here quoted more realistic numbers showing their P4-3.2s going as fast as my Pentium-M1.6 (AND doing two WUs at the same time, so really twice as fast. This makes more sense). Typical P4 3.2 chip goes for $240

Duron with cheap mobo => decent (for the price) but 1/4 as fast as the Pentium-M. $20, Rare.

Sempron with cheap mobo => decent (for the price) but 1/2 as fast as the Pentium-M. $72. Common.

Dang! Cache ain't king afterall. I'm real glad you guys tore apart my article. Made me look more closely at my data (and found quite a bit of original info to be wrong). I guess the magic bullet doesn't exist afterall. Oh well.

For absolute speed, it looks like the P4's have got it. You must get a good mobo for it though, or all that speed gets lost somewhere.

For absolute speed, with maximum efficiency, the Pentium-M's have got it. Mobo architecture doesn't seem to be an issue.

For the fastest speed given the price, Durons!!! An interesting point is that the P4s actually beat the Semprons on a price/performance ratio. Of course, after a few years of crunching, you might make up for that difference in your increased power bill ;-).

Jimmy

P.S. I am a big fan of AMD, but it looks like Intel wins out... Twice...

---

> I majored in E.E. in college and took a few classes in CPU design. We studied
> the MIPS R8000 at the time. Always wanted to get into the industry but never
> got my foot in the door.

I did electronics (Avionics) in the Navy but did software after I quit.

> So, as it turns out my P4 EE, actually does have 2 MB Level 3 cache. 512KB
> Level 2 Cache, and 8KB level 1 cache. So, it ain't the lack of cache that's
> slowing my P4 down. It's probably the fact that it's in an HP laptop that's
> slowing it down. Duh. It even has DDR2 memory, don't know what the heck is
> wrong with it.

I have been interested in the effect of the L3 cache as I don't yet have a processor with anything beyond a L2 yet.


> Dang! Cache ain't king afterall. I'm real glad you guys tore apart my
> article. Made me look more closely at my data (and found quite a bit of
> original info to be wrong). I guess the magic bullet doesn't exist afterall.
> Oh well.

Cache is important. And it is going to become more important with the next couple generations of processors. Mostly because with more cores and logical processors to feed, we will need more juice to continue to improve performance.

The good news is that more an more we are seeing multi-threading being included at all levels of the systems. Even MS is starting to really improve the adding of threads and I expect that when they finally do get Longhorn out of the starting gate we will see (Paul's opinion only guys) the first MS O/S that multi-tasks well ... though I may be wrong about that too ...

> For absolute speed, it looks like the P4's have got it. You must get a good
> mobo for it though, or all that speed gets lost somewhere.

Yes, good memory helps too... I don't always get the high speed memory like I should so that does hurt my performance some ...

> P.S. I am a big fan of AMD, but it looks like Intel wins out... Twice...

I am agnostic for the most part and I have had some good AMD systems and some that are truly horrible ... and I did get MB from good supplier but it did not help ... they need to get multi-threading going to really get even with Intel and then they will be able to compete on level playing field.

IBM also has some interesting stuff in the Power archituecture as shown, and I hope to see one of those chips in my PC before I die (note it is a chip carrier 4" on a side)...

Especially of interest is the fact that there is L3 cache (4 of them acutally, each with 36M !!!) ...

---

> For absolute speed, with maximum efficiency, the Pentium-M's have got it.

What about Xeon's?
They seem to do pretty well too - if you have the cash to spend. :)

---