AMD TAKING ON CHIPSETS WITH HAMMER ?

[Guest]

Hmm. I dont think applications have to be rewritten at all to take advantage of Jackson. All you need is OS suppport (which would be very simple to implement).

If you are running just one thread, you wont see any improvement by adding Jackson technology. But if you run multiple threads / apps / multi threaded apps, Jackson will help. Its much like adding another cpu (well, 1/3 of a cpu). Doesnt help running one single threaded app, but it does help running multiple single threaded apps, or one (or more) multithreaded apps.

So, basically, the required rewrite for Jackson, is almost nill. Its the same optimization that you'd need to take advantage of SMP.

As for performance increases.. I seem to recall 30% was the maximum achieveable improvement, with 10% being 'normal'. Thats still a nice boost, if you consider how little die space Jackson requires. I think its a very clever way to get around cpu stalls, and I suspect it to help more as cpu's continue to ramp in speed faster than memory technology.

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

 

[mala]

PR-ratings are not based on anything at all really. They are just four numbers meant to give the buyer the impression that they are buying a "faster" processor than they are.

AMD says the PR-rating is the number of megahertz a thunderbird would need to have to get the performance of a particular Palomino. That's BS if you ask me.

PR-rating Real Clock XP/TB ratio

1000+ 1000 1.00
1100+ 1066 1.03
1200+ 1133 1.06
1300+ 1200 1.08
1400+ 1266 1.11
1500+ 1333 1.13
1600+ 1400 1.14
1700+ 1466 1.16
1800+ 1533 1.17
1900+ 1600 1.19
2000+ 1666 1.20
2100+ 1733 1.21
...
3000+ 2333 1.29

That suggests that an XP is 14% faster than a TB @ 1400MHz and 20% faster @ 1666MHz. If AMD stops here there might be some truth to it. Maybe XP scales better with frequency so that the increase in performance gets better at higher clockspeeds. But if they keep increasing the PR with 100+ every time they increase the real clock with 66MHz they have a problem. If a palomino really is 29% faster at 2333 MHz compared to a TB that says more about the "badness" of a TB than of the "goodness" of an XP. If we assume that XP scales perfectly that would mean TB only scales with 76% of the clock speed.

One thing you can be sure of is that a palomino will always complete a task at least as fast as a thunderbird. How much faster (if faster at all) is impossible to say because it varies from application to application. Another thing you can be sure of is that any desktop processor completes 0,1,2 or 3 instructions each cycle. No more (and no less =) ) If an application (or rather its inner loop) already is perfectly optimized for a processor there is no way you can beat it without clocking your chip higher or rewriting the software.

Another thing:
Someone asked for electrical specs for Itanium.
http://developer.intel.com/design/itanium/downloads/249634.htm

/Markus

 

[zengeos]

I kinda think Hammer performance steps will increase in 200 or 400 PR ratings points. So we may see something like Hammer 3400 then Hammer 3600, 3800, 4000, etc.

But I could be wrong! Just kinda seems that the 100 mhz speed steps aren't enough for most folks to warrant the prices, especially when the actual performance is marginally improved to the point it's not really noticeable.

Mark-

When all else fails, throw your computer out the window!!!

 

[Matisaro]

Thats why their pr ratings were independantly verified. The palomino does gain more per clock than the tbird, and I think the pr ratings are pretty accurate.

~Matisaro~
"The Cash Left In My Pocket,The BEST Benchmark"
~Tbird1.3@1.5~

 

[Guest]

Hah ! Hammer is just a half hearted attempt to get into the 64 bit market, and to stay in the CPU market before they become bankrupt.IA-64 is the way to go ! Northwood and future edtions of the PIV will outperform Hammer easy.

Bye Bye AMD ! *LOL*

 

[bum_jcrules]

Is that you "Bonehead_Inside" aka troll?

If you don't have something constructive to say get off the forum. Respect the forum.

BTW: that Itanium datasheet is for the 733/800 MHz version. Anyone have or find anything more recent?

<b>All for one and one for all...and 3 for 5! - Curly - The Three Stooges</b> :lol:

 

[mr_gobbledegook]

Don't critise what hasn't been released yet, Hammer has a lot of potential.

<font color=purple>~* K6-2 @ 333MHz *~
I don't need a 'Gigahertz' chip to surf the web just yet ;-)</font color=purple>

 

[mr_gobbledegook]

Just read Anandtechs interview with VIA which highlighted some aspects on Hammer and Chipsets. <A HREF="http://www.anandtech.com/chipsets/showdoc.html?i=1565" target="_new">CLICK.</A>As you know VIA is supporting AMD with a Hammer chipset.

The crucial diffrence with Hammer is that the Northbridge chip (usually part of the motherboard chipset controlling memory access) is to be integrated with the CPU instead. As we all know this will undoubtably increase performance, but in the interest of politics how will this affect AMD as a CPU manufacturer and more importantly the chipset business ?

AMD have always stated that they do not wish be involved in the chipset business but with Hammer it seems they have already taken the first step into this arena. How long before future versions of Hammer include a south bridge chip as well ? Surely AMD have to tread very carefully if they do not want to step on VIA's toes.

First impressions from VIA are fairly positive and dont seem to be bothered with the Hammer approach. In my opinion AMD have taken away some of the power that VIA possesses in producing chipset platform solutions as it will be much harder for them to enhance performance with future motherboards with just the Southbridge avaliable to them.

Basically AMD want maximum control over the performance of Hammer so that chipset manufacturers cannot comprimise its overall performance. But what will this cost AMD in the long run ?...

1. More complex overhead costs of improving the Hammer design in the long run i.e to remain competitive they will have to revise the CPU design as well as the Memory controller. Meaning it will take AMD longer to develop future core enhancements therefore slower delvery time to end production.

2. We could see a backlash from other chipset manufacturers as they will lose some business.

3. Thirdly as AMD are dealing with memory access and types they will have to cope with new versions of DDR memory as when they are released.

DRAM speed enhancements occur more times within a given CPU generation than CPU core enhancements. Therefore how will AMD adapt the Hammer to cope with this ? Will diffrent versions of Hammer be released for diffrent speeds/versions of memory ?
AMD is probably at its biggest turning point in its history of the company and will have to think very carefully about their strategy. It seems quite a lot for AMD to take on and I will be closley following this story to find out how AMD deal with these issues (they could get stuck in the mud if they are careless).

Finally...
With the use of Hyper-transport and integrated Northbridge where do the bottlenecks in the Hammer platform lie ? will it be in the Southbrige and Graphics Card ? From reports the Hammer platorm will feature AGP x8 but no cards use this..most people say that AGP x4 hasn't even been fully exploited.

<font color=purple>~* K6-2 @ 333MHz *~
I don't need a 'Gigahertz' chip to surf the web just yet ;-)</font color=purple>