Intel Pentium 4 670 & Pentium D 820

[Tim S]

http://www.bit-tech.net/hardware/200..._pd_820/1.html

[Hamish]

im curious as to whether you forced cpu affinity during the multi-tasking benches?
i've noticed that XP seems to be really crappy at dividing processes between the 2 cores

sandra and BF2 are really good examples of this:
Sandra cpu benchmark is multi-threaded - it uses 100% of both cores
Sandra Mem bandwith benchmark is single-thread but SMP aware - uses 100% of 1 core
BF2 is horribly coded pos - it uses 100% of one core then 100% of the other, it alternates between cores

i've noticed this a lot, so when im going to be doing some heavy multi-tasking i usually force te affinity of the 2 apps to different cores, this seemed to make a big difference (it also helps with stability of certain apps, bf2 for instance cant stay connected to a server for more than about 2mins if you dont force the affinity to 1 core or the other )

[Tim S]

I did, where possible

[Hamish]

okey dokey
did you notice any of what i was talking about with either chip? im wondering if its an XP problem (most likely tbh) or an AMD thing

also, could you stick the numbers on the multi-tasking graph for the same chip but without the multi-tasking bit, so we can see how many fps you drop by runing XYZ in the background?

[Tim S]

I don't have the X2 4800+ anymore, unfortunately

[Firehed]

About what I expected for results :P

Now Newegg is making my head spin (though that could be because it's 6am and I'm just getting ready to go to bed...) all of the 945p, 945g (one is ATX and one is BTX I think) and 955XE (or whatever letters) chipsets support the dual core LGA procs, right? Either I'm confused or newegg needs to update some compatibility notes.

The 820 seems like a good deal all things said and done. Not for gaming, but for everyday use anyways. I love my A64 for gaming, but I still miss how smooth everything felt with hyperthreading. And dual core pwns hyperthreading. *ponders setting up a flash countdown to me having enough 'spare' cash for an X2*

[TMM]

My ass prescott has a better performance:watt ratio then northwood

[Tim S]

Originally Posted by TMM

My ass prescott has a better performance:watt ratio then northwood

[mclean007]

Originally Posted by TMM

My ass prescott has a better performance:watt ratio then northwood

You have a prescott in your ass?? You should be on "Ripley's Believe It Or Not"

[RotoSequence]

I noticed a mistake on the first page

Originally Posted by bigz

It is based on the Smithfield core, which has the a similar feature set to the Prescott core and comes clocked at 2.8GHz with a 1MB L2 cache for each core - in fact, you could consider the Smithfield core to be two Prescott 1MB cores stuck together on the same die.

Technically its two seperate cores installed on the same packaging; there actually is physical seperation between the two dies. This is pretty significant, as if they were on the same die, communications between the two CPUs wouldnt have to occur over the pins.

In response to your question about anyone gaming at low resolutsion with a good card, I hate to say it, but there is one person I know who insists on playing Counterstrike at no more than 1024x768 on his 6800 GT.

Im surprised no one as of yet has taken an 820 and compared it with a 520 or 620 processor... thus far theres been no test to show exactly HOW much performance difference there is between two cores and hyperthreading at the same clock speeds.

Overall a pretty nice review bigz; you do seem a tad rusty though compared to previous reviews. By the way, do you think you used the word "workhorse" enough on the last page?

[Tim S]

Originally Posted by RotoSequence

I noticed a mistake on the first page



Technically its two seperate cores installed on the same packaging; there actually is physical seperation between the two dies. This is pretty significant, as if they were on the same die, communications between the two CPUs wouldnt have to occur over the pins.

I take it you haven't seen this picture then?

Presler is what I would call 'two separate cores installed on the same packaging'.

Presler will improve yield because you can pick any two cores and install them on the same packaging. On the other hand, Smithfield requires two good cores to be next to each other on the wafer as they are physically attached.

[RotoSequence]

You would be correct; I had not seen that picture My mistake, sorry

Do you have any inkling as to why it would require two good cores next to each other? they dont communicate with each other over the die, and might as well be completely seperate processors

[Bindibadgi]

Space on the package - look at the space required between the dual core 65nm's and you'll see why.

[RotoSequence]

I dont see whats stopping them from putting a millimeter or so's distance between two dies though... are you catching something else visually that I'm not?

[Tim S]

It is the way that the architecture works - Intel have opted for a shared bus because of the limitations of not having the memory controller on-die. AMD share the memory controller and the HyperTransport bus and the cores are connected before that, Intel's current dual core technology does not allow communication on-die because it still needs to talk to the northbridge (AFAIK).

[Tim S]

Originally Posted by RotoSequence

I dont see whats stopping them from putting a millimeter or so's distance between two dies though... are you catching something else visually that I'm not?

1mm between the two dies would create a 'hot spot' on the core, IMHO.

[RotoSequence]

Originally Posted by bigz

1mm between the two dies would create a 'hot spot' on the core, IMHO.

I honestly doubt it would do much; you can always fill any air gap with the heatspreader, and if anything, might actually improve dissipation as the heat would be over a wider area.

Originally Posted by mclean007

You have a prescott in your ass?? You should be on "Ripley's Believe It Or Not"

Another time someone at bit has cheered me up with a funny quote! Thanks

[Bindibadgi]

Originally Posted by RotoSequence

I dont see whats stopping them from putting a millimeter or so's distance between two dies though... are you catching something else visually that I'm not?

They way Intel seal their cores to the substrate requires a few millimetres around the edge of each die. Also, most Intel LGA heatsinks have a small, central copper core so two large cores would have areas outside of the contact area.

[Da Dego]

A note to Hamish,
It's XP that controls thread assignments. It's entirely buggered and I'd love someone to write a competent third-party piece of software to direct the cores. I've been bitching about this since the advent of HT. Perhaps they'll fix it in longhorn? I won't hold my breath, however...it's why I haven't really seen dual core as all THAT great a thing...I get sick of babysitting my HT, much less an entire damn second core! It will improve DRASTICALLY when MS pulls its thumb out and actually starts to develop a bit smarter version of thread prediction/assignment. Not the easiest thing in the world to do, but if they just bothered I'm sure it would show up in no time.

As for the article, great job as usual, biggles, but I agree with the comment about being a little rusty on the pure non-graphics hardware end...may wish to stick to writing with bindi on this topic. You're tackling the whole article the same way you would a GPU, and it's just not quite the same thing. I wish I could explain what it is about it, but the whole thing just feels like I read about a new gfx card as opposed to a CPU line.

Also, info on O/C would be useful...this chip is a lower budget chip in the dual core market...would be nice to see how high it successfully cranks...I'm also concerned about the heat output when you OC dual cores on the same chip. Maybe cover in another article along with the A64 X2? It'd give you a reason to get the chip back, at the very least.