Video cards soon to adopt mutlicore architecture.

[smoth]

http://news.cnet.com/8301-13924_3-9968992-64.html

tl;dr, nividia has realized that multiple small cores are better then one big core. They will likely be making this their new direction.

so if the newest nvidia cards are going to be multi-core with what they say are weaker cores then how will spring handle this? I suspect that multicore video cards are going to be a big possibility in the future and it only makes sense.

[jcnossen]

AFAICS that article talks only about AMD going multicore, nvidia will keep their one big single core design.
Anyway graphics rendering is already highly parellel, multiple cores will likely be handled by the driver or the hardware without applications knowing it.

But this might also just be marketing talk. AMD might be unable to deliverer a faster core design than nvidia, so they're trying to talk around that.

IMO in a few years it will all go towards stacks of chips:
http://www.eetimes.com/showArticle.jhtm ... =208402316

[smoth]

Impressive, that would be pretty cool jc!

[AF]

I think that this will more likely impact in that some graphics methodologies will start to get a lot faster. Some may become less efficient, but I think that these are probably already considered less efficient already, and that the reduction of speed due to architecture of these operations will be wiped out by the faster gpu so people wont notice.

[[Krogoth86]]

Well this is no big news as multi-cores are the way to go. NVIDIA also has announced (just as ATI) that the current generation (at least for NVIDIA as their new GPU was presented today) will be the last monolithic one...

In the future they'll be multi-core too and the classes will be like high-end has 8 cores, mid-range 6 and low-end 4 to 2 to just give you the idea. By applying techniques a bit different to today's SLI / CF (well we maybe will see a snippet of this for the upcoming ATI 4870 X2 which is rumored to no longer use a bridge chip and so won't suffer from the usual multi-gpu disadvantages) there'll be no problem with getting the performance into the applications...

The drivers should manage all this and so it doesn't seem like the engines will have to deal with this (at least nowhere near how it's for multi-core CPUs)...

[jK]

You have still the old architecture of GPUs in mind, GPUs are since a long time multi-cores.

Todays GPUs have all multiple shader units with each multiple render units, so a GPU has >100 cores!

And Nvidia even dropped the fixed-function pipeline in their >geforce8, it is now (all) done on the shader units (FFP was very limited in its expansion on more threads).
And in the future GPUs will even have many more cores (multiple thousands), because it is so easy to integrate: in contrast to cpus the gpu works per-fragment (there are several millions of pixels to render each frame!) and non pixel result depends on any other pixel result (-> their are totaly independent), so the programer doesn't need to _anything_ to make usage of those many cores.

The only issue is now if you integrate all those cores in one chip or if you put them in several ones, what is cheaper because mostly 30-70% of the produced chips are defect (sometimes only a few shader units don't work, so they deactivate those and sell them as low-end/consumer cards). But using multiple chips would also make the boards larger (I remember times when ppl laughed about the vodoo 5 6000 and now any high-end card is larger and even needs additional power support), also the produced heat and the used energy would nearly doubled.

So the future are stacks (in many different types) like jc said

But GPUs are already multi-cores and Nvidia's "threaded optimization" is a setting for the drivers running on the _CPU_.

[Caydr]

Joke post... right?

[imbaczek]

um, doesn't one of the recent radeons have like 320 stream processors?

[Archangel of Death]

Their top end cards have had that many since the X2k days iirc...