More SLI
- Thread starter PatrickL
- Start date
chavvdarrr
Veteran
IF on slower CPU, results are lower, wouldn't it mean that drivers offload VPU's job to the CPU ? Then these results would not indicate real-world performance... am I right ?
Chalnoth said:
2-3 frames in the driver seems rather high. People are spending money on getting high-end broadband to lower their network latency to <10ms (ping times <20ms) for games. Having the graphic driver insert 30ms extra latency would be *bad*. As for double/triple buffering, people disable vsync for the same reason (that and the boost in framerate in the case of double buffering).
Cheers
Gubbi
That means you'd have to "invert" the vertex shader, which is somewhere between hard and impossible for anything more complex than a simple matrix multiplication. The plane doesn't neccesarily stay as a plane.Chalnoth said:
Yes, there are other places that contribute to the latency, but none of them get any faster by AFR.Chalnoth said:
You are right that "one quick, one slow", and higer memory consumption also can be problems with AFR.
And I agree with Gubbi that 2-3 frames in the driver seems high.
Richthofen
Newcomer
PatrickL said:
a small market which is driving the low end. Lead the high end and you will sell more low end and mainstream cards as long as they don't lack too much behind. SLI is a great and clever move by Nvidia that's for sure.
They will totally dominate benchmarks with that sollution. No one will care if it is 60 or 70 or 80%. Everything over 20 to 30% is looking like a true domination.
Further on it makes products looking more attractive.
The opportunitiy to add a second card later on for a cheaper price once the first one gets to slow is driving sales.
i am sorry but it is a non sense. Domination of ben,chmarks with 2 cards against one? You are again thinking people are dumb.
Look at the Hardocp/nvidia configuration. Can you give me the price of that combinaison ? And did you look that 2 6600 GT on a very expensive system are not able to be always better than one card ?
Look at the Hardocp/nvidia configuration. Can you give me the price of that combinaison ? And did you look that 2 6600 GT on a very expensive system are not able to be always better than one card ?
PatrickL said:
Sadly yes people are dumb. Care to guess how many X700XT's will get sold because someone's buddy tells them that ATI has the fastest card on the market (XT PE) ?
PatrickL said:
This has always amused me. People keep referring to the small market to which SLI will appeal as something negative when the market for other IHV's offerings is absolutely zero because they offer NOTHING!!
Last time I checked something > nothing PatrickL said:Because you were not reading the forum (Holidays?) when the whole thing started
Anyway i am done with theses arguments, only time will tell.
No argument. Just can't follow your train of thought. If you don't want or can't afford to use SLI how does that negate the fact that it is available for those who can? People today pay $100 extra for 5% increase in performance so I don't understand the bitching about paying $400 for an extra 50-60%.
PatrickL said:
Time will tell if Nvidia delivers. Not if SLI is a worthwhile solution. I've always stood by the concept.
Actually, they do. The other latencies in the system, provided you are GPU-bound, are waiting for the graphics card to do its thing. So, if more frames can be rendered in the same amount of time, then the other latencies in the system are reduced.Basic said:
Even if there's only one frame of latency in the driver (there has to be at least one for the CPU and GPU to operate independently), AFR will reduce total latency. Consider:
Let's imagine that total rendering time takes 30ms for one graphics card, and that it takes 15ms to do the CPU side of the calculation. Here, then, is the total delay:
Input -> 15ms for the CPU to prepare the frame.
30ms waiting in the driver until frame sent to GPU.
30ms for the graphics card to render the frame -> output
Total latency: 75ms
Now, imagine we put another card in SLI mode here. You still have to wait 15ms for the CPU to prepare the frame, and 30ms for the graphics card to do its rendering, but since a graphics card will be available to start rendering every 15ms (instead of every 30ms), the above then looks like:
Input -> 15ms for the CPU to prepare the frame.
15ms waiting in the driver until frame sent to a GPU.
30ms for the graphics card to render the frame -> output
...reducing total latency to 60ms. This isn't as small as if you had a graphics card that could render twice as fast, but it's still better latency than just one graphics card rendering at the same speed as each individual SLI card.
Now, keep in mind that when comparing these latencies to network latencies that with network latencies you have the additional factor of not having infinite bandwidth. Basically, if too many things change between subsequent frames, the server just won't be able to send all of that changed information to you quickly enough. This means that the "ping" to the server is not necessarily the total latency between when an object is updated server-side and when it is updated on your computer.
What I always liked about SLI is the fact that you can spend your big bucks for a top of the line card when it first comes out, and buy the second card when they are near EOL for cheap.
You buy your $300 card when it first comes out. Games barely stretch your card so you run at high res & quality.
1-2 years go by, and the next gen cards come out. They offer double the performance of your puny solution, and they cost $300. Your model card is in the discount bin for $100. The Next Big Game comes out and your card is way too slow. You pick up a second card, get nearly the same performance as the new cards, but you've spent $400 in two years instead of $600.
I knew people who were picking up thier 2nd V2s years after they first came out. Its a great way to increase the life span of a card.
You buy your $300 card when it first comes out. Games barely stretch your card so you run at high res & quality.
1-2 years go by, and the next gen cards come out. They offer double the performance of your puny solution, and they cost $300. Your model card is in the discount bin for $100. The Next Big Game comes out and your card is way too slow. You pick up a second card, get nearly the same performance as the new cards, but you've spent $400 in two years instead of $600.
I knew people who were picking up thier 2nd V2s years after they first came out. Its a great way to increase the life span of a card.
Of course, it's a bit more than that, as you'll also have to spend the money for the motherboard that has two slots. Granted, if you're planning on doing this upgrade over two years or so, you'll probably upgrade your motherboard at least once during that time and so will only need to spend an extra $20-$50 or so....
