AMD Dual Core Info and Performance

[OneEng]

Well.... and estimate anyway

From here: http://www.heise.de/newsticker/meldung/51845

The article is in German, but according to babblefish translation, it appears that the (N-5 and N-3) represent the processor is running at speed grades lower than (N) but are dual core 2 socket systems vs. the (N) example being single core 2 socket systems.

This is right down impressive results. These graphs show that performance will MUCH more than double with the new processors! In order to explain this, we have to look here: http://www.eet.com/semi/news/showArt...cleId=49400814

The key points I found were that the hardware pre-fetch has been improved (drastically I would say), and 10 new SSE3 instructions have been added (not sure what this does for K8 performance).

So someone explain to me again how it is that Xeon is going to maintain any semblance of performance parity with Opteron?

I am also curious about future desktop cores. Will the single core version of K8 be receiving the core enhancements on 90nm also? I am wondering how much of the improvement is due to the core enhancements and how much is due to the dual core.

It could be that the enhancments to the core alone will allow AMD to bump the model numbers at the same clock speed by ~500 (~15%). This would allow AMD to release an A64 4500 and 4700 at only 2.6Ghz (using 512Kb L2 and 1Mb L2 respectively).

Then again, it may only offset the increase in FSB that P4 is going to get along with the increase in L2 that P4 is also slated to get (2Mb is the rumor).

Reguardless of what model numbers AMD chooses to put on them, it looks like we have a good possibility of seeing better performing desktop processors from AMD next year

In the server/workstation market, Opteron is looking to be pretty unbeatable! In fact, I think that big tin is going to take a thrashing here pretty quick! Those things smoke!

[ThreeOnTheTree]

From what I gather from "5 speed grades lower" is that it will be 1.6ghz. I estimated that it would be released at 1.8ghz, so I'm not too far off if I'm reading it right. I think eventually 2.0ghz cores will be possible as well. 2.2 might be the maximum on 90nm, but I guess time will tell.

[uart]

So the blue graph (100%) is a current dual opteron system

The red graph (approx 125% to 140%) is a the same system with each opteron replaced with one of these new dual core thingies running 5 speed grades below that of the original.

The green graph (approx 135% to 160%) is as above but this time with the new dual core cpu's at only 3 speed grades below the original opterons.

It my above understanding correct? Hey that's a pretty neat cpu upgrade for the two way system.

[OneEng]

It my above understanding correct? Hey that's a pretty neat cpu upgrade for the two way system.

Well, yes and no. The rumors I have been hearing is that Oracle and other software companies will charge by the core, not by the die.

Oracle charges 24K per core. That means in order to upgrade my dual opteron e325 with dual core processors, I am looking at 48K in JUST liscense fees for Oracle

I am wondering how M$ will handle this as well.

[OneEng]

CNet has an article here: http://news.com.com/AMDs+dual-core+p...?tag=sas.email

The 30 to 55% figure seems quite small compared to the charts in the other article. Still, I would emagine that the dual core Opterons will be quite fast.

[Luds]

Oracle charges 24K per core. That means in order to upgrade my dual opteron e325 with dual core processors, I am looking at 48K in JUST liscense fees for Oracle

That will change IMO. The performance boost of adding a second core does not translate into a double performance gain, hence it seems disproportional to charge double the cost for licensing. In all actuallity, one is being penalized for an architecture improvement. Remember the hyperthreading fiasco? MS quickly changed their licensing on that.

I've read some interesting stuff on the subject but off the top of my head can't recall where.

Here they talk about the model being obsolete and new methods of charging for software will have to be implemented.