Why did Sager let HP beat them to the punch? - Page 3

Clock for clock, the P-M is supposed to be 1.8 times as fast as a P4 (no HT). In particular, this marketing number published by Intel makes the P-M 1.7 appear "equivalent" to the 3.06 P4 (no HT, 533 FSB).

HT adds something like 1.7x speed in "typical processor-intensive environments". That means the 1.7 P-M = 3.06 P4 = 3.06/1.6 = 1.8 P4HT. So, yeah, clock-for-clock the P-M is faster. And the P-M consumes less power and produces less heat that an equivalently-clocked P4HT, since all P4HTs are desktop procs. That is, if you believe Intel's marketing.

But 1.8 GHz is a bit slower than 2.8 GHz. And who cares about clock-for-clock? You can get a P4HT at twice the clock speed of a P-M for about the same price.

If you want a Mobile HT-enabled chip, I think the Athlon 64 Mobile is the only chip on or close to the market with both those features. At least I think it's HT-enabled, the other A64s are.

-phubar

I'm not sure about a trouncing - but be assured that the Centrino is no slouch...

Benchmarks

Benchmarks

Benchmarks

Benchmarks

I would really reccomend reading the the two articles I've linked to pages of, as then you will learn that the Centrino is NOT a dog of a CPU...And teaming one up with a 9600PRO would fly off the shelves...

I dont think anyone was saying it was a dog - but reading those two pages you linked simply showed me that the 1.6 PM is about the same as a non HT 2.4 P4 - which is pretty mcuh what we were saying when the PM kiddies arrived and started spouting off that a 1.7 PM = 3.06 P4HT chip .... which is obviously bollocks

Hm... I seem to be spoiled by blender's efficiency. I'm sorry, but I've never even touched other 3D modelling applications because they are just so damned expensive. In any case, I'm sitting here, with a particular scene, around 12,000 verticies, rotating it in real time. This is on my AthlonXP 2000+ machine with a GeForce2MX graphics card, 32mb graphics memory. The other machine that I'm doing the same exact thing on is an AMD K6-2 400mhz with a generic S3 Trio card from who knows when, 4mb graphics memory. Granted, the K6-2 loaded the scene a lot slower, but there was no appreciable difference between the rotation speeds, which were without lag and in full frame.

So tell me, is it just because blender is so damned efficient that even huge scenes are not a problem? Man, if that's the case, I'm amazed why people are still using rhino and 3dmax because blender seems to be beating them handily...

If this is true, why does Intel still make NON-m chips? and in greater quantities?

Curious,
-myrkat

see

I thought that the Centrino processor is faster than the desktop, BUT, NOT one with HT.
Was there some reviews on this instead of heresay?


I think this is a good valid argument.

I let it go 5-10 times ... the Centrino is NOT a processor, right?. It is a P4-M processor nothing more, nothing less in my understanding. The Centrino comes from the bundled chipset, wireless options, etc etc etc.

I scanned the thread, and noticed no-one has yet to state the outrageous price.

almost 3k for what would be similar to a sub 2k sager system in specs?

Yeah sure...sign me up.

If you're going to argue semantics, the 'Centrino' processor is NOT the P4-M. The 'Centrino' is officially the 'Pentium M', abbreviated P-M. The 'P4-M' or 'P4M' refers to the Pentium 4 Mobile, which is, as the name implies, the mobile version of the P4, a completely different chip.


The P-M being equivalent to a P4 (no HT) with 1.8 times the clock speed is Intel's official line on Centrino performance. (And yes, that means the performance of the entire Centrino system, because Intel barely acknowledges that the P-M is available seperately.) Intel's come up with benchmarks to support this, apparently using tests that heavily benefit from extra L2 Cache because no one else seems to be able to quite reproduce that number. A factor of 1.8 seems to be the optimal, rather than typical, performance ratio - and it strikes me as suspicious that 1.7GHz (the fastest P-M) times 1.8 is 3.06GHz (until recently the fastest P4).


We have already discussed the price difference - go back and read the first 12 posts in the thread...

-phubar

In all things its a matter of compromises to get to where you want to go.

Centrino has a larger cache memory on the CPU PLUS the bundle of WAN and the Intel mainbord chipset. Its main goal is for mobile users who DO NOT have access to constant AC power.

Any of the Sager units: Performance hogs who needs lots of AC POWER. Unless you have constant AC power then you may have a problem with these units.

If you need the mobility and freedom that the Centrino offers then you have answered your own question. However there are better providers of this than HP/Compaq.

If you need raw power then look no further than Sager. Now all you have to do is find the model that is right for you.

Its that simple and that direct.

If you really want Centrino I would look at Dell or Toshiba before I would use HP/Compaq.

Good luck and enjoy whatever laptop you get.

and also from the dates of those.. most likely a 533 fsb...couldve been a 800 but i think it wouldve been notated as such.. so no ht and most likely 533mhz fsb... makes a difference

actually the Pentium-M is Banias.. centrino is intels marketing scheme for bundling it with their mini-pci wireless....

okay

thats correct...its NOT a processor. Centrino is the pentium m, intel wireless, and the chipset if im not mistaken. my bad.

So based on the fact that the 1MB cache is the reason these chips appear to do so well, If we plug in a P4EE with similar (or larger) caches we will get 1.8x performance out of our 3.2 GHz chips.

Somehow I don't think so... I smell fresh bovine excrement here. But if real world benchmarks do show the comparison to be true then I can't wait until I put a P4EE into my 8890.

Well, there's things like having a scene in the viewport with textures and lighting turned on. Not to mention theres doing stuff like subdivision surfaces or Maya's interactive realtime lighting renderer which probably push the GPU.

That aside, the reason people use things like Max, Maya, XSI, etc. is probably because things in their featureset and workflow that doesn't fit their needs in Blender. Go figure.

As for the whole PM silliness. Some people do value mobility (mobility being a composite of things like weight, size, and battery life) and are going to be willing to pay for it; otherwise Apple would never sell any PowerBooks or iBooks. And considering the question of the PM's performance, the whole PM 1.7 = P4 3.06 that phubar posted *was* Intel's own marketing hype. Which I think we all will agree is highly suspect. But, I don't see how its somehow innately wrong to pair a decent GPU with the PM as Devine_Madcat was implying; especially considering its the Mobility Radeon after all...

@ jamesshuang

No, you're not being spoiled by blender. Your 12,000 vertices will have little effect on most cards because 12,000 vertices is a relatively small number. I work in real-time 3d games and just as a sample, some of our scenes go into the 100,000+ poly range, not including avatars. Along with the poly count goes all the additional stuff like dummy objects, lights, cameras, vertex shading, textures and everything else appearing in the view windows. This is enough to upset a great many cards. If this were to be pre-rendered, the count would rise hugely as real time stuff is fairly angular and simplistic.

250.000+ poly's is not an uncommon number these days for a typical scene with a good bit of detail... provided you graphics card can handle it of course

I knew there was a reason I got my 8890-V with the 9600 Pro and 128Mb of graphics memory .

I meant to say that the "chip used in the 'Centrino' package is the 'Pentium-M'" - that's what I was trying to say by putting 'Centrino' in quotes... clearly I wasn't very clear, though. Bad me. Actually, the improved per-cycle performance of the P-M has more to do with having a more efficient architecture than the P-4. It's the same reason Athlon chips at much lower clock speeds perform the same as higher-clocked P-4s.

The cache size does aslo contribute to increased performance. But benchmarks of the 3.2 P4EE against the 3.2 P4HT (same chip without L3 cache) show only a 5-10% performance improvement in generic tests. However, as with any feature, you can build a test to specifically highlight the benefits of a larger on-board cache. So a test deliberately designed to perform optimally with at least 1 MB cache and choke with less than that will show a significantly larger improvement based on cache size. It's the non-universal improvement on these specific tests added to the more efficient architecture that let the P-M hit the 1.8x figure on Intel's marketing benchmarks.

There's already been extensive discussion of architectural changes that affect CPU clock-for-clock efficiency, do a search if that bugs you.

The bottom line is that the P4 architecture sacrifices massive amounts of efficiency in order to achieve stratospheric clock speeds. This lets Intel market their chips with a much bigger GHz number, which improves sales, since the average person equates clock speed with processing speed, even though a quick comparison of Athlons and Pentiums proves that not to be the case. Of course, an inefficient, high-clock chip also consumes a lot more power than a low-clock one, which is why the P-M is so much better for same-performance mobile applications.

Architectural inefficiency basically boils down to wasted cycles that go unused while working on a given task. HT seems to seriously cut down on the ineffiecieny of the P4 design by allowing a second thread to be processed in what would normally be wasted cycles in single-thread operation. This is why otherwise identical procs with HT are so much faster than non-HT in multithreaded environments, because HT allows the P4 to be used far more efficiently.

-phubar

A future Pentium replacement, I believe the CPU following Prescott will adopt a similar architecture to the Pentium M, i.e. (shorter pipeline, etc.).

The marketing issues of the current P4 are ironic. Intel designs a CPU with a long pipeline which, while ultimately ineffecient, can be run at very high clock speeds. AMD counters by marking their CPUs with a performance equivalent quasi clock speed. Thereby doing an end run around Intel. Then Intel, after supporting the "higher is better" fallacy, markets a much lower clock speed CPU (Pentium M) which trounces equivalent speed P4s. Talk about confusing the public. It took quite a while before the notebook manufacturers were willing to produce desktop replacement Centrinos. I think, because, everyone assumed that it was a low end CPU because of its low speed. Actually, before I had heard of the desktop notebooks using the 685PE chip and desktop CPUs, I was looking for a maxed out Centrino.

Re the performance diff due to the 1MB cache size, the Pentium M has a 1MB L2 cache; whereas, the P4XE-P4EE? has a 512K L2 cache and a 2MB L3 cache. The imminent Prescott also has a 1MB L2 cache. I believe that a 1MB L2 cache is probably as good as or possibly better than a 2MB L3 cache, so my 8890's 2.6 CPU will be replaced with a 3.2 Prescott, assuming that they can keep it below 90 Watts!

Im a bit late in quoting this, but since im still the first....

The P3 1.4 kicked the P4's butt, hard. Many a benchmark showed it upstaging 2.0s and 2.2s easily.
The reason the P3 wasnt given DDR, etc, was because the P4 was, and still is, the flagship chip. Do you REALLY think Intel wants people to know that thier $600 rip-off is being beaten by some other measly chips, that they happen to make? Hell no. it was the death of the Tualatin, and it will be the bain of the Banias...