News AMD announces triple-core CPUs

http://www.bit-tech.net/news/2007/09/18/amd_announces_triple-core_cpus/1

Can't afford a new quad-core processor but your current dual-core just isn't enough for you? Well AMD is covering all of your bases with the introduction of triple-core Phenom processors.

 

... by the time i'm ready to upgrade my C2D we'll have octacores. waste of R&D time, they should have put it into their quadcore development to get it out earlier.

 

Doubtful there was a whole lot of R&D in this one. Celerons were just Pentium parts with half of the L2 cache disabled and that's more then like the same scenario that this is: a faulty core that's been disabled. That'll help AMD save from binning a bunch of parts and give a little bit of a boost to their product lines with an additional option. These triple-core processors might just run slightly better then their quad-core brothers as they'll have all the same resources to share but one core less in the competition.

I guess we'll see come next year though.

 

This didn't require any R&D. As far as I can see it is just a quad-core with a single core that has been disabled, probably because it was defective. Intel can't do a similar thing due to the way their core-design works. For AMD this is a cheap way of filling a "gap" in the market and using defective quad-cores, without having to sell them as dual-core and incurring a bigger loss.

 

i dont see a huge market for 3 core processors however it will mean that they will be able to sell some of the slightly fault quads

does this suggest that the quads aren't binning very well?

 

ROFL

so if the article is true, maybe we can do a "pencil trick" on triplecore to unlock the 4th one? LOL

 

Its not even like quad core processors are that expensive these days, look at the q6600 for example, get them for £170ish

 

I think this is a brilliant move. There has been no R&D as some have suggested - AMD will simply test the four cores of an X4 CPU and, if one is defective but the chip is otherwise useable, why not make it an X3 and sell it for a price between the same clockspeed X2 and X4 parts?

Intel hasn't done this because their four core parts are two dual core dies in a single package, so they can independently test each die before packaging them together and only use fully functioning dual core dies to make quad core parts, with the broken dual core dies going to make budget single core parts.

Also, maybe three cores is a sweetspot for some - I regularly run a couple of tasks at once and it would be nice to have headroom for more, but I don't really need four cores. If it was only a little more expensive to get 3 rather than 2 cores, I'd probably be tempted.

 

I agree, this is a fantastic move. There isn't really anyone that looses out on this. It saves a lot of waste. It is basically extra income for "free". There will sure be a market for this. When people see triple core just slightly more expensive than dual core, then they might fork over the extra pounds to pay for the triple.
Plus they can then start to cut down on the dual core manufacturing more and more. They could even get to a point soon when they dont really need to make dual cores anymore and for the budget processor they can just sell off any defective quads they get, even disabling ones with 2 defective cores and selling them is possible.

 

I think that it is a pretty good idea, but it is very strange. And, considering that getting a quad core CPU and a motherboard that can support it at stock speed at least is relatively inexpensive. AMD are going struggle, as there processors at the minute still look, in my eyes, weaker than the competition at Intel.

 

This is great news and very clever on AMDs part. Salvage broken quads and fill in the market for upgraders who dont have enough for a quad. Even though I will be getting a Q6600 I am happy AMD is getting back in the game. As we all know competition is great for the consumer.

 

If it is a defctive core mod, then this is a win-win scenario for AMD and could help their position, If only the defunct core is disabled will this leave the full shared L3 (I think?) cache available to the remaining cores? could be interesting to see the performance effect this may have.

 

AMD's architecture doesn't really thrive on large cache sizes, it thrived more on the lower memory access latencies associated with the on-board memory controller.

 

it will depend on if perfomance scales like it logically should.....will these now have to be seperately programmed?

 

Whilst it may not cost them in R+D how many people are biting their nails worrying about C2D being too little and quad being too much? Tenner says you could count them on 1 hand

 

I should think it will depend on the developers.

They have to design there program around a choice of paths, eg:

1 thread (won't use extra cores at all)
2 threads (will use exactly 2 cores)
3 threads (will use exactly 3 cores)
4 threads (will use exactly 4 cores)
Multi-threaded (as many cores as are available)

If they choose to use any but the first or the last they also have to do the ones above it in the list or they remove possible users. A fair few companies have gone for dual core support so far and those won't benefit from 3 or 4 cores.

From a user perspective multi-threaded is the most attractive as it will scale with any number of cores now and in the future.

From a company perspective multi-threaded will take the most work to pull off (specific core implementations could be as simple as shunting a given part of the code onto a second core permanently if available, where as multi-threaded requires the program to decide how to do it on the fly) but once they've got a solution that works they should be quite able to re-use it in future.

The way I see it, the more core options we have and are represented in the market, the more likely the devs are to actually go multi-threaded, which will be good for us

 

"Houston we have a problem"!!!!

If this Tricore news came out a month or so after the release of Phenom it would be an easier sell but since the news is only a week after the launch of the 6 months late Barcelona, it must put shivers down the spins of all the major Server OEMs. Barcelona needs to be reliable and with this news following so closely after the launch of Barcelona reliability of the quad core Barcelona will be very much in doubt.

Seems to me that most here believe this is wonderful and do not realize that AMD is dumping the Barcelona reliability problem on the backs of the desktop enthusiast. No wonder Mr Richards quit AMD when he did, he did not want to be the mouth piece of a Barcelona project that has turned sour. I realize GPU companies have released cores that did not function correctly and released them as slower but still functional cores but AMD's announcement must set some kind of speed record for admitting that yields are not what they should have been, beings that AMD does not have a Phenom desktop chip released yet. You do not cut into your own sales of the high end quad core Phenoms pre release, most of the time a company will release the worse chips as slower products such as the 1.9 GHz Barcelona when compared to the 2.0 GHz Barcelona. You do not go out of your way to announce "hey guys, why purchase a quad core that you may have been waiting a year for when we have perfectly good Tricores for you." This kind of news usually comes from the competition and not from the company that has just spent billions upon R&D and fabs to manufacture a Quad core

 

How do you get from low yields to reliability problems ?

While it does demonstrate that they are having some yield problems, enough that they will have a decent number of Tri-Core chips to sell. I just don't see how you can presume the latter is also true from this.