Initial Dual Core Benchmarks 04/05/2005

http://www.overclockers.com/tips00756/

I'm loving it

After reading the anandtech article, I really welcome dual cores. I'm glad "real world" performance is finally taking a front seat to GHz queens.

Something else intererstin to add.
http://www.tomshardware.com/business/20050330/index.html

Sjaak I'm loving the Avatar

hmm... if it looks like there is real-world benefit... i might be in on it.... imagine F@H on one of those.

speed bump said:

Something else intererstin to add.
http://www.tomshardware.com/business/20050330/index.html

Sjaak I'm loving the Avatar

Click to expand...

I dislike the THG approach to things. Still, there's some nice info there.

THG did a decent review of it all at least.

But truely I can't wait til Dual Cores hit. Finally got the Intel scores, now I want to see AMD scores on this matter. Use to own a Dual CPU system, and I can see the benifits from having 2 CPU's. I personally can't wait til these are out for a little while.

I'm waiting til the 65nm or til AMD releases the new socket possibly.

Oh speaking of Dual Core, check this out http://www.pro-g.co.uk/news/nid/580/

Velocity Micro has announced a new range of PCs that will use Intel's dual-core technology. The new PCs will use the Intel Pentium 4 Extreme Edition 840 processor and will ship with speeds up to 4.0 GHz. The new dual-cores feature Hyper Threading which allows for faster performance in software that is built to take advantage of threaded technology.

Click to expand...

Dual 4 Ghz? how the FAK do they want to cool that?

Sjaak said:

Dual 4 Ghz? how the FAK do they want to cool that?

Click to expand...

They said water cooled. THG has a thing on the PEE CPU saying that @ 4.0Ghz they wouldn't consider it stable on air, and recommened water cooling for 3.8Ghz and up. But they didn't say if they increased voltage or not, they just pumped up the multiplyer on the CPU to make the jump to 4Ghz.

deathman20 said:

They said water cooled. THG has a thing on the PEE CPU saying that @ 4.0Ghz they wouldn't consider it stable on air, and recommened water cooling for 3.8Ghz and up. But they didn't say if they increased voltage or not, they just pumped up the multiplyer on the CPU to make the jump to 4Ghz.

Click to expand...

Hmz, maybe 90nm is finally showing off some advantages over good ol' 180nm.

I'm extremely happy to see this concept finally hitting the consumer markets. Ever since I purchased my first 486dx2 I have been drawing up schematics of cubic processor design 3x3x3 processors of which the core processor handles thread assignments..

Essentially we would be able to delegate to 26 processor cores within one processor, and in theory run them at a mere 100 mhz. And as you know, 100 mhz technology is now stock in your basic handheld and prodocues less than 10 watts of heat, making a cubic processor feasible these days.

Today we write 2005, when will we see the cubic processor concept hit the consumer market?

Cheers, Flix

26 processors

and a 3d processor cannot yet be made with the current lithography technologies, it allows for only one layer of transistors. Also, the yields of such a complicated (and large) CPU would be terrible, making it expensive.

Sjaak said:

26 processors

and a 3d processor cannot yet be made with the current lithography technologies, it allows for only one layer of transistors. Also, the yields of such a complicated (and large) CPU would be terrible, making it expensive.

Click to expand...

You could stack dies and interconnect them.

But isn't a true 3D Die something they are still working at making?? Instead of having everything laying flat on the die, making everything standing up in the die? I swear I read something about this.

Gnufsh said:

You could stack dies and interconnect them.

Click to expand...

Some time ago i've read something about stacked (multiple-layers) CPU's, and the conclusion was that with conventional etching technology it would be amost impossible.

Future Multi core cpus have many benefits:
For instance with several different cores you could run a fast core or two and some slower cores. The slower cores can take care of doing things like the background virus scan and the overheads from perpetuals such as USB devices (im on a 800mhz single cpu system my mouse uses noticeable cpu resources if I move it about with task manager open) Or mabey processing driver calls i read some place that dx9 etc uses some times upto 50% of the cpu resorces to provide the gpu with what it requests and a key reson why x box does so well is the effient implimentation of this.

The fast core could be used to game on and because like the article says there could be the possibility for multiple levels of cpu speed on the full carrier the heat output could be more reasonable and performance in comparison with single threaded single cpu situations.

Due to limitations in how much multithreading is practical, fast cpus are still necessary in the package. Implimenting two speed grades cores on one cpu would enable good performance in all situations.
For instance a render farm some times uses thousands of cpus to render all the frames in a movie. The high level of parallelism here is suited for many many slower cpus because they can work on there process and all return there work to the central pc. However in a real-time render such as a game the next item to process is almost random, does the character stop? Jump? Die? Here it is not possible to render the next step simultaneously as it is not planed. However physics, A.I. and the players interaction with the game(input and output ie render) could be prime candidates for threads in this case.

Allowing the cpu to do the actual cpu resource allocation is an interesting, in fact possibly good, idea but in a multi core multi cpu situation harder to implement and its the server farms that will benefit most from these where cpu density is vital.
This problem of resource allocation could be treated in a similar way to cache coherency on smp systems. AMD implemented a 'snoop bus' on the old k7 mp chipset, this dedicated bus idea can instead of being used just to insure cache coherency could be used to inform the other cpu how busy its partner is. When one cpu is trying to allocate a thread and the other cpu has the resources to process it the thread could be transferred to the second cpu using high speed bus similar but much faster than hypertansport.

The cpu would require processes executing on it to have a id that would dictate to the allocater the priority of the task and an estimate of the complexity of the task. all these things would complicate the implementation of hardware and code. A complexity estimate would decide how fast the core that processes it would need to be, the game being complex the mouse not so.

Looking closer to today a dual core will bring the benefits of smp to the masses! Hyperthreading's a start but a true dual is grate to use.
I have a smp system, one things sure is that you never want to go back unless you are a compleatly obsesivegamner and the diffrence between 70 and 75 fps are crucial.

whats smp and why have a dual http://www.ocforums.com/showthread.php?t=78677

I went from a single to dual back to a single But my single was a huge upgrade to what my dual was hehe

Yes I confess I have considered building a new pure gaming rig but id rather buy a new dually with some more memory bandwidth. The old k7 is a bit lacking in this department.