Dual core Pentium 4 won't work on 925 & 915

[purrkur]

Another reason why not to buy Intel stuff!

[BCN]

well, at least they say it won´t, AMD does not say anything

[purrkur]

well, at least they say it won´t, AMD does not say anything

Well, actually companies building Opteron Motherboards have been saying that their current boards will be dual core compatable (like Tyan). You can also see more about how a BIOS upgrade will be the only thing needed over here and here.

But I understand what you are saying. AMD has not revealed much about their dual core workstation CPU's (they are launching their dual core server chips first while Intel is doing it the other way around). I am however betting on it to be the same as for the Opterons because in many cases they are using the same chipsets.

[Dave Rave]

what, they changed the socket again?
gee, fancy thinking of doing that
cheeky sods.
you'd think after ppga, fc-pga, fc-pga-2 , slot-1 and socket 423 they might give us a break.
oh well,.... lots of second hand parts to build seti crunchers with

[purrkur]

what, they changed the socket again?

Actually they haven't this time around. That is the funny part.

Man, I would feel like a dumb sucker if I had just bought myself a brand spanking new 915 or 925.

[Atropos]

Hi there,

I guess the reason why 925x will not support dual core Pentium is
that the Pentium D will not really be a "dual core" chip.

Intel just packs two Pentium Cores in one package. Each will have
it's own L1 and L2 cache. The Extreme Edition will have L3 cache also.
But the cores are "stand alone" each other. It's just like our BP6 setups
but not two sockets, only one.
The chipset must support SMP this way.

The "real" dual cores like IBM Power CPUs may share the L2 and L3 cache.
So if one of the cores needs loads of caching while the other don't the first
one can use the L2 of the other and vice-versa.

The extreme edition will have HT enabled for both cores meaning the OS
will "see" four (4) CPUs.
Hope the guys over at Redmond will make some of their software
stuff multi threaded. This could help Intel to sell their overpriced chips.

Bye,

Atropos

[purrkur]

Yeah, overpriced for sure

I must mention one thing though and that is that having dedicated cache per core is better than having shared cache, just like having dedicated memory per CPU is best in a multi-cpu machine. The reason for this is because the two cores don't know about each other and what they are doing. When you want to try to allocate resources in cache that two cores share that don't know of each other then you put overhead into trying to figure out how much you can use without hurting other processes etc.

As for overpriced and underperforming, Intel will probably try to sell these chips to gamers, claiming that performance will be so much better. The fact of the matter is that these chips will be slower under games because the dual core CPU's they make will be clocked lower than their single core chips. Since games don't use multithreading then you won't see any performance boost at all.

[Atropos]

Hi purrkur,

I'm not sure about the cache thing ...

Any time the power4 or power5 multicore beasts are compared to
"dedicated cache" - designs they win in the eyes of the tester.

As I don't have any experience myself (I am not now and maybe
will bever be able to spend 380 000$ for ONE multicore CPU) I just
can repeat what others think ....

Maybe it's the same thing when the P4 came to light:
Everyone was sure that the NetBurt architecture and myOps are the
future way to go. But reality showed that 1.4GHz Tualatin were way
faster tham 1.4GHz P4.

Sure - the P4 was designed to break GHz frontiers from the beginning.
But the next generation CPU should be bigger, better, faster, more
(choose the one you like ) than the actual...

Or am I wrong at this point?

G'night

Atropos

[purrkur]

Any time the power4 or power5 multicore beasts are compared to"dedicated cache" - designs they win in the eyes of the tester.

The least you can do when you quote a test is to share with us the location of that information so we can read it and judge by ourselves. I work with big iron hardware all the time and I do have the experience and education to have more than just an educated guess on this. Consider dual CPU machines. Why do the low end motherboards have shared memory (like the BP6) and the high end boards have dedicated memory per CPU? The same logic goes for cache which is just high speed memory. It is very easy. If Intel is going this way then I would say they have made the right choice if they want to squeeze performance out of their dual core CPU.

Maybe it's the same thing when the P4 came to light:
Everyone was sure that the NetBurt architecture and myOps are the
future way to go. But reality showed that 1.4GHz Tualatin were way
faster tham 1.4GHz P4.

Sorry, but no, it isn't the same thing. Yes you are right about the Tualatin and first P4 in terms of speed but the reasons for this are not even remotely related to dual core architecture setups. The reason for the initial P4 being slower than the Tualatin was because Intel introduced very large pipelines in the P4. Large pipelines make the processor "slower" because you need more cpu cycles in order to execute quite a number of things within the CPU. However, it also gave Intel the capability to clock the P4 to what it has today so the 1.4GHz was just the starting point for the P4 while it was the end of life specification for the P3. To be honest with you, I don't like the Intel design philosophy because it has nothing to do with performance, but many say that Intel marketing was involved (because they wanted to market the "Higher MHz means faster CPU concept) so the P4 is actually a poor design and not a good one. However, this has nothing to do with dual core setup, even if the dual cores will have the P4 core with all its faults like this one above.

Sure - the P4 was designed to break GHz frontiers from the beginning.But the next generation CPU should be bigger, better, faster, more (choose the one you like ) than the actual...

Or am I wrong at this point?

No you are not wrong. As I mentioned above, many of better knowledge than I have said that Intel designed the P4 the way they did because they wanted to create a design that can reach high clock speeds. This probably means that marketing was involved in how this design came about and I am sure that Intel was a bit ignorant about how fast their competitors would come out with good designs as well. Not to mention the fact that sometimes they seem to have no clue as to what they are doing. Consider the fact that the latest Prescott core was supposed to bring them over 4GHz. They backtracked on that one quite fast after launch because it couldn't reliably do it. How come they didn't know this before launching the Prescott CPU's (which by the way are the champions of heat dissipation).

I would still like to read the article you keep on mentioning.

[Atropos]

Sorry for not giving the source where I got the conclusion about
cache design for dual cores from.

It was in the German IT magazine called c't. I'm not sure if anyone cam or
would like to get German IT magazines over the channel or even over in
the US.
Sorry, the article is not accessible on the net. Just hardcopys on paper

I guess you got me wrong when I described the architecture of powerPC
multicores. Each core has it's own cache. But the cache can be accessed
by the other core and vice-versa. This way one core can directly access
calculation results, pointers, tables etc. that were done by the other one.
It's the same idea I guess AMD had when designing the opteron.
As you say each of them has it's own memory. (Not the same for most
Xeon setups.) But each CPU can access the other ones' RAM directly.

Hope everyone now gets the point I want to state.
(It's kind of hard for me to translate all the technical info I have so
that you can understand what I want to say. Please tell me if my English
is too poor. )

What I meant when I was talking about P4 design.
I am aware tha t the "long pipeline architecture" of the P4 was designed
for high MHz rates. The point was storing myops in cache and no "old"
x86 instructions seemed to be better choice than former designs.
When the P4 came to light it was not significant faster than latest P3.
That's why I think the design/speed combination was broken in a way
The P3 design cannot reach high GHz. So (at least that's what I think)
the first incarnation of P4 should have been P4 2GHz or better.

Hope I corrected myself in a way you can understand me, now

Bye,
Atropos

[purrkur]

Hi again Atropos,

I actually do know c't mag even though I don't speak a word of German. It is a well known paper that has uncovered many misteries having to do with both software and hardware. I would say that they are better at what they do than most IT mags out there. They do have some stuff in english on their website too.

Don't worry, your English is just fine although your previous post was a bit fuzzy.

If Intel is trying to make another poor implementation like you say they are trying to do with their dual core and also trying to sell it at a high price then I guess their arrogance will catch up with them sometime soon. I for one will not put a penny more into that company. For me it is AMD all the way, and even PowerPC if it would become a viable alternative at the same price. I use Linux so I am not too bothered by hardware choices.

[Atropos]

hi again,

it's nice to see that the mag. I rely on is known in other regions of this world.

If Intel is trying to make another poor implementation like you say they are trying to do with their dual core and also trying to sell it at a high price then I guess their arrogance will catch up with them sometime soon. I for one will not put a penny more into that company. For me it is AMD all the way, and even PowerPC if it would become a viable alternative at the same price. I use Linux so I am not too bothered by hardware choices.

I guess the inplementation of the Pentium D will not be that bad. My main
point is the architecture they will use. The two cores of a pentium will be as
if you plug two actual Pentium4 into a dual board. It's kind of having two
independent CPUs in one package.
The main flaw of Intels design seems to be that (in actual dual CPU setups
- like our BP6 setups) both CPUs must share the RAM and the path to RAM.
In AMD opteron setups each CPU has it's own RAM. I don't know yet what
AMDs dualcores will be like ... ;-?

As you say the only acceptable CPU by now is an AMD IMHO.

G'night

Atropos

[davd_bob]

I wonder how the Task Manager's Preformance graphic will display it.

[Dave Rave]

much the same as it does now for the P4 HT

[Atropos]

If you have the Pentium D Extreme Edition which will have HT
enabled for both cores will be kind of a quad CPU system for the OS.

Are there 4 graphes in taskmanager when you're running 4 real CPUs
in your box?

I guess that time will tell

Atropos

[purrkur]

I think that there will be four graphs in Task Manager. At least you have a bigger reason of having two graphs for dual core than there is for a single core plus HT! So I put my money on four graphs.

[InactiveX]

If you have the Pentium D Extreme Edition which will have HT
enabled for both cores will be kind of a quad CPU system for the OS.

Are there 4 graphes in taskmanager when you're running 4 real CPUs
in your box?

Found this in THG:



http://www17.tomshardware.com/cpu/20040 ... nt-01.html

I presume the dual-core HT computer would show up the same.

[Atropos]

thank you for the link ...
nice information about SMP with HT CPUs.

[phaedrus]

Yeah, overpriced for sure

Definitely. And definitely overpriced for the terrible design. Just to throw a word in on the bad design tactics of Intel, Hannibal over at Ars Technica (yes, I quote them a lot... they're a fun source for tech news) gave up on defending Intel a while back. Just check out the first page of this article. Apparently, he's gotten confirmation from insiders that the netburst crap was a marketing decision and not a performance decision. He's a rather reputable hardware reviewer (he particularly focuses on CPUs and CPU architectures).

As for overpriced and underperforming, Intel will probably try to sell these chips to gamers, claiming that performance will be so much better. The fact of the matter is that these chips will be slower under games because the dual core CPU's they make will be clocked lower than their single core chips. Since games don't use multithreading then you won't see any performance boost at all.

It will and it won't be better. UT2k3 is mildly multithreaded--the sound is done in a second thread. It apparently gives a little performance boost for dual processor machines. Doom3 is heavily multithreaded, and you get a significant boost from dual procs (to the point where I was just almost playable on my BP6). Games are going in the multithreaded direction. Everything is going that way, because the only way we'll get more out of our CPUs is by going to multicore type designs (either the Opteron, the Pentium D or the Cell processors). This is barring some breakthrough coming out of left field.

Jeff