IDF Preview

[Mephistopheles]

Interesting stuff from IDF...

* - <i>Grantsdale will indeed feature azalia and DDR2-533 dual channel.</i> Hopefully, Azalia will be much needed breathing room to the generic '97 crap that can be found onboard around Intel platforms. (doesn't mean it'll be on par with nForce2's excellent onboard sound... but hope never dies...) Also, DDR2-533 is, from my point of view, a strong indication of 1066Mhz FSB; P4s take a 15-20% performance hit from asynchronous operation (FSB-memory, that is). It would certainly increase P4's current IPC, which is something they need.

* - <i>Dothan Information</i>: Hopefully, we'll be seeing something dothan-related. And Centrino-related. But I'm not with high hopes on this one. I mean, at this point, the only true news would be that it got launched... and that is unlikely to take place just yet... maybe within a few weeks. (it would, however, give Intel one hell of a thunder)

* - <i>CT Technology Information</i>: What is Intel's response to A64 technology? How quickly will it be on the market?... This last one is of particular interest...

Will Intel's implementation suck? Or will it be any good?... Maybe Intel will try and market their notion of "Intel 64 - better than other 64-bit implementations" or whatever?.... (and maybe there'll be some truth to those claims?...)

I don't know about you, but I'm eager to see if Intel will try to show some stuff to fight off AMD... AMD has been giving them a hard time for a while now. If they react correspondingly, everyone will benefit...

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i><P ID="edit"><FONT SIZE=-1><EM>Edited by Mephistopheles on 02/16/04 09:31 PM.</EM></FONT></P>

 

[gobeavers]

I sure hope Intel announces some stuff that will make my A64 computer a harder decision (or do i?).

Anxiously awaiting yamhill....is it here yet?

 

[Kanavit]

CT= Clackamas technology. Intel's x86-64 cpu compatible

<A HREF="http://forums.amd.com/index.php?showtopic=8162" target="_new">http://forums.amd.com/index.php?showtopic=8162</A>

<b><font color=blue>Geniune Intel processor</b></font color=blue>

 

[gobeavers]

yes, I know. I guess I should have said it better. I would want an affordable proc, and one that would be coming out in the next 3 months. I don't want to wait forever. I need a new computer now.

Anxiously awaiting yamhill....is it here yet?

 

[Kanavit]

Yamhill will be in LGA 775 socket form with Grantsdale chipset(i915). which is still months away.

<b><font color=blue>Geniune Intel processor</b></font color=blue>

 

[juin]

Mars that not really far

Just to show dad

 

[P4Man]

> P4s take a 15-20% performance hit from asynchronous
>operation (FSB-memory, that is)

? Sounds like a serious exagerration. I doubt real world performance would be affected more than ~5% from running memory asynch.

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[Mephistopheles]

That said, we were also forced to recognize that memory performance in asynchronous mode sank by around 15 percent against synchronous operation. It just shows that you can't rely just on the processor clock rate for getting the most out of your system.

Try this <A HREF="http://www.tomshardware.com/cpu/20031230/5ghz-13.html" target="_new">link</A> for more information...

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>

 

[P4Man]

that link is so poor on specifics, and written by two of the most incompetent hardware reviewers I know, so allow me to use another link:

As we can see, the purely synthetic benchmark Sandra shows a performance drop up to 15% with the 2/3 multiplier used, and 5-6% with the 4/5 multiplier. The PCMark benchmark (also synthetic) is less strict: 10% and 5%, respectively. But in real-world applications (Quake3) the performance drop is less than 6% and 3%, respectively.

using a 2/3 multiplier, so using DDR266 instead of DDR400. yet only a 15% drop in synthetic tests? And only 6% in real world apps, with a lot less bandwith ?

Sure sounds like running memory in synch or asynch isnt *nearly* as important as having the highest possible bandwith, and your 15% claims just for the asynch are obsoletely unfounded. Even 33% slower memory AND a asynch operation don't incur such a penalty

I may have misread the article though, feel free to read it here:
<A HREF="http://www.digital-daily.com/cpu/intel-northwood-d/index05.htm" target="_new">http://www.digital-daily.com/cpu/intel-northwood-d/index05.htm</A>

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[P4Man]

>Mars that not really far

LOOOOOOOOOL !! Juin, that is a classic !!!
its only 56 million kilometers

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[Mephistopheles]

It's not that easy to benchmark. In this article, for instance, they operate a FSB-800Mhz P4 with DDR266, DDR333 and DDR400. But how much penalty is actually consequence of asynchronous operation and how much is from lower bandwidth? However, I have to agree that 15% was a high shot. Maybe 5% would be in order.

Also, if you OC, you can keep memory running asynchronously, but you wouldn't have a clean result either, because the processor would have different characteristics (higher clock). So giving a clear penalty number for asynch. operation would have to be done much more carefully, in my opinion.

In either case, Intel has <i>never</i> endorsed asynchronous operation at all. It would be a break from tradition if they did support DDR2-533 and wouldn't include the possibility of upping FSB to 1066Mhz.

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i><P ID="edit"><FONT SIZE=-1><EM>Edited by Mephistopheles on 02/17/04 03:44 PM.</EM></FONT></P>

 

[P4Man]

I'm not claiming that intel won't release a 1066 bus (isnt it even on their roadmaps ?), nor that it won't help performance. but not because of synchronous operation as such, but just because of the higher bandwith, lower latency.

>In this article, for instance, they operate a FSB-800Mhz P4
>with DDR266, DDR333 and DDR400. But how much penalty is
>actually consequence of asynchronous operation and how much
>is from lower bandwidth?

The point is, that both combined hardly give a huge difference. ~3% in real world apps. Even if 50% of that penalty is due to the asynch operation (which I very much doubt), you are looking at a 1.5% real world difference. The synthetic memory bandwith tests don't even show a penalty as large as you'd expect just from of the slower memory, so you could even conclude the asynch operation *boosts* performance (not likely, I agree, but still..)

> However, I have to agree that 15% was a high shot. Maybe
>5% would be in order.

On synthetic apps measuring memory performance, maybe. On typical benchmarks, not even close.

>Also, if you OC, you can keep memory running
>asynchronously, but you wouldn't have a clean result
>either, because the processor would have different
>characteristics (higher clock).

Someone could try running a P4 synch and asynch at both stock fsb, and stock fsb +1 Mhz. That should allow you to compensate for the slightly higher cpu/fsb speed. Or (not sure if its possible) test fsb stock (synch), fsb+1 (asynch) and fsb-1 (asynch). Average +1 and -1, compare it to +0 and
that should give you a pretty accurate number.. I won't be big though.

= The views stated herein are my personal views, and not necessarily the views of my wife. =