Second add in card of most value: PPU or SLI GPU?
- Thread starter rwolf
- Start date
I think you need to specify how far into the future we're talking. SLI will definitely have much wider scope in the near future - especially with ATI saying their version is much better and has much better compatibility than Nvidia's current approach. It will take quite a while before hardware accelerated physics is widespread. But of course we need to have the hardware out there at reasonable cost before developers start taking advantage of it. Having Epic behind you can't hurt though.
IgnorancePersonified
Regular
PPU for the reason that cpu performance has hit a wall that I serioulsy doubt will be over come any time soon.
well if its true and the dx over head is around 30% or greater utilization on current cpus then dual core cpus will be a huge boon. Put the dx threads onto one core and have the game it self use the other core , that should give at least 30% speed up , factor in all the other backround apps nad windows itself all running on a second cpu leaving a cpu free to do all game stuff. That would be a huge increase in speed . Then x86-64 being programed for and we should see some larger than expect gains in the next year or so .IgnorancePersonified said:
This poll is a bit strange because it makes some assumptions or forces the voter to make some and the answer they give won't necessarily make sense unless their assumptions, and therefore reasons, are laid out.
For example, does this poll assume that SLI will be needed for adequate "hardcore" graphics? Does it assume the PPU will make a such a difference for physics?
See how it's easy to tell yourself that SLI won't be needed at all and therefore PPU becomes the default?
Because a dedicated PPU will most likely far surpass any CPU, multicore or not, and SLI is not wanted, but maybe needed, the only sensible answer I can give is for PPU.
Nobody wants to be dependent on SLI. Nobody wants to be dependent on a discrete PPU. These are just things that may provide enough added value that they make sense for the hardcore/enthusiast crowd.
However, if the PPU has support from very few titles, or titles that don't interest me, and SLI becomes needed to experience games in all their glory upon release, then I will obviously want to come back here and change my vote. I'll hold you responsible.
For example, does this poll assume that SLI will be needed for adequate "hardcore" graphics? Does it assume the PPU will make a such a difference for physics?
See how it's easy to tell yourself that SLI won't be needed at all and therefore PPU becomes the default?
Because a dedicated PPU will most likely far surpass any CPU, multicore or not, and SLI is not wanted, but maybe needed, the only sensible answer I can give is for PPU.
Nobody wants to be dependent on SLI. Nobody wants to be dependent on a discrete PPU. These are just things that may provide enough added value that they make sense for the hardcore/enthusiast crowd.
However, if the PPU has support from very few titles, or titles that don't interest me, and SLI becomes needed to experience games in all their glory upon release, then I will obviously want to come back here and change my vote. I'll hold you responsible.
IgnorancePersonified
Regular
Bandwidth: - 1st iteration of Intel Dual core chips are going to have bus contention issues. A single core chip is already relativley starved. Hammer addresses this and would appear to have room to spare when you compare single channel to multi channel however;
Cost: Both AMD and Intel are not going to lose highend cpu revenue streams just to provide you or me with a dual core chip at 1/3 the price of a current high end cpu price. Sorry - I don't buy the marketing. What that translates into is an expensive cpu vs a relatively cheap ppu = PPU.
I would prefer to see intel forgo it's net burst architecture, bite the bullet and intergrate a ppu in the space used by 1meg of cache. I think that is inevitable. then we could have dual core cpu and dual core ppu in the next process shrink or a combination of.
Cost: Both AMD and Intel are not going to lose highend cpu revenue streams just to provide you or me with a dual core chip at 1/3 the price of a current high end cpu price. Sorry - I don't buy the marketing. What that translates into is an expensive cpu vs a relatively cheap ppu = PPU.
I would prefer to see intel forgo it's net burst architecture, bite the bullet and intergrate a ppu in the space used by 1meg of cache. I think that is inevitable. then we could have dual core cpu and dual core ppu in the next process shrink or a combination of.
Nah, all we'll have, once dual-core becomes ubiquitous (I would imagine most high-end systems will have at least dual-core within two years, and even many mainstream systems by that time), Intel and AMD will start moving towards a more Cell-like design, where you have a small number of high-IPC processors, and a large number of processors designed for parallel processing. These are what are going to really accelerate physics and other easily-parallelizable operations.
Blazkowicz
Legend
IgnorancePersonified said:
but high end CPUs were always about being 10 or 20% faster with 300% the price, and people buying them anyway. The cheapest dual core P4 will be at $240 if if recall correctly (2.8ghz with no HT)
In a few years I guess all the low end would be dual core, as AMD and Intel will have no other way of spending their transistor count (and you can expect a lot of "2x the power!" propaganda soon)
to answer the question : if it's for future gaming, then buy it in the future, not now
PPU sounds nice, unless it means you will be stucked to tech demos and a few games for two or three years..
About nvidia SLI : incompatibilites and problems with stuff like post-processing make me uncomfortable (I ran for years with voodoo2 SLI, then Voodoo5, SLI was basically "plug in then forget about it"; nowadays it's more complicated)
I don't care anyway, I can't afford even a single 6600GT or 9800pro. You can buy a 4, 8, 12 or 16 pipelines version of basically the same thing, so SLI is only useful for the very high end.
A single 512MB card may even be a bit more future proof than a pair of 256MB ones
(if you want better texture detail in a doom3 or UE3 powered game)PPU.
edit
removed and idea whose merit has come into question.
end edit:
I only say PPU in oh the next 5 years because I feel (and desire) physics will become the next big field to gameplay and maybe even the "wizbang" hook so to speak.
However with WGF2.0, advanced chips like R500 etc with multi-GPU possibilities and probabilities way out in the future...and that fact that we will reach percievably realistic physics before we reach "real" level graphics in 2D I also pick the GPU/VPU and more so in the long term.
I also feel if we should ever enter the 3D realm for "real"...err...holodeck real...it will be easier to achieve real physics than real surroundings so having more graphics processing power would seem to be more valuable.
...I also picked PPUs because Chalnoth hates them
edit
removed and idea whose merit has come into question.
end edit:
I only say PPU in oh the next 5 years because I feel (and desire) physics will become the next big field to gameplay and maybe even the "wizbang" hook so to speak.
However with WGF2.0, advanced chips like R500 etc with multi-GPU possibilities and probabilities way out in the future...and that fact that we will reach percievably realistic physics before we reach "real" level graphics in 2D I also pick the GPU/VPU and more so in the long term.
I also feel if we should ever enter the 3D realm for "real"...err...holodeck real...it will be easier to achieve real physics than real surroundings so having more graphics processing power would seem to be more valuable.
...I also picked PPUs because Chalnoth hates them
Somebody, for the love of all that is holy, show me how a PPU is useful without totally redesigning systems around it in exactly the same way as was done for 3D accelerators all those years ago.
Cause guess what? It ain't happening again. Not for another 5-10 years.
Cause guess what? It ain't happening again. Not for another 5-10 years.
IgnorancePersonified said:
Single-thread performance has hit a wall. Multi-thread performance still has a way to go, and then there's asymmetric multi-core options (eg. Cell).
IIRC the Ageia PPU is about 125M transistors isn't it? Intel/AMD could choose, should they wish, to replace one of the cores of the DC A64/P4 with such a unit. If it was useful for other purposes beyond physics (multimedia maybe), and the HPC-on-a-budget brigade would love it, then this model may not be as daft as it sounds.
IgnorancePersonified
Regular
Yep - that's why I said "or a combination of."
Chalnoth said:
Why? It's one thing to argue that they're wasted transistors for anyone who doesn't play games, but frankly a lot of the capabilities of the vast bulk of CPUs sold today are rarely if ever put to use.
I can't see a *pure* PPU ever ending up in a CPU, but a more generic vector-compute-style unit I can see (especially after Cell).
Similar threads
