Single card vs SFR. Its not that hard to understand is it? That extra card is giving a whooping 20-25%.
Last edited:
Did you not read the post above you?
"This graph highlights how frame times and frame rates aren’t the same — according to our first graph, D3D is faster than Mantle. Actually play the game, however, and Mantle is easily the superior solution."
If you split the frame into 50%/50%, 1st GPU has 4GB of VRAM and needs to render ~50% of the workload, the 2nd GPU has 4GB of VRAM and needs to render ~50% of the workload. That's why your can load much more than 4GB of VRAM into 1 frame and split the resources. With AFR this isn't possible since GPU 1 renders frame 1 and GPU renders frame 2, meaning that at any given frame, GPU 1's 4GB of VRAM is the limit for frame 1.
Since it's unlikely that a single frame can be split into 50% VRAM and 50% GPU load evenly for each GPU, that's why scaling is not going to be similar to AFR's. However, it means you can theoretically use both of the GPU's VRAM pools simultaneously.
Did you not read the post above you?
"This graph highlights how frame times and frame rates aren’t the same — according to our first graph, D3D is faster than Mantle. Actually play the game, however, and Mantle is easily the superior solution."
If you split the frame into 50%/50%, 1st GPU has 4GB of VRAM and needs to render ~50% of the workload, the 2nd GPU has 4GB of VRAM and needs to render ~50% of the workload. That's why your can load much more than 4GB of VRAM into 1 frame and split the resources. With AFR this isn't possible since GPU 1 renders frame 1 and GPU renders frame 2, meaning that at any given frame, GPU 1's 4GB of VRAM is the limit for frame 1.
Since it's unlikely that a single frame can be split into 50% VRAM and 50% GPU load evenly for each GPU, that's why scaling is not going to be similar to AFR's. However, it means you can theoretically use both of the GPU's VRAM pools simultaneously.
But you still need most those resources in the VRAM, as the same resources are needed on both halves of the screen. There may be some small gains, in the likes of frame buffers, but you still need the same assets available to both halves of a screen. Even if some isn't needed in a particular frame, it will still be loaded, as they need to be available for when the gamer changes his view.
Single card vs SFR. Its not that hard to understand is it? That extra card is giving a whooping 20-25%.
That's not what his post says. If in the top half of my screen there are 5 buildings to be rendered for GPU #1, and the bottom half of my screen has 10 cars to be rendered by GPU #2, why does GPU #1 need to render 10 cars and GPU #2 need to render 5 buildings? This is how AFR works, but not how SFR works. That's the whole point of SFR is you split the workload on the same frame, meaning you don't need to duplicate all of the assets. In AFR since GPU 1 renders the entire frame 1 and GPU 2 renders the entire frame 2, all of the assets in 1 frame need to be rendered by each GPU interchangeably.
Your posts tells us nothing about vastly improved frame times -- graphs were already posted with 290X vs. 295X2 that show a major improvmeent in actual smoothness. Further, your post tells us nothing about why the full 8GB of VRAM can't be utilized, or say 5GB, or 6GB in SFR. You seem to be focused on FPS, which misses the entire point of the discussion - frame times and aggregate VRAM utilization.
What is the benefit over a single card?
- Improved frame times that make gaming smoother despite lower FPS than is possible with AFR.
- Access to more VRAM than is available on 1 card via AFR.
Are you not reading the thread all?
You completely missed the point. Nobody in any significant numbers are going to buy dual cards or more for 20-25% gain.
For such a well informed member of this community I'm finding it extremely hard to understand how you don't understand the thread are only focusing on raw FPS gain....
Are you just not understanding the thread?
Lets say average FPS is 50 but frame timing is terrible to the point where the game is a stuttery mess. Then, lets say we get a 20% gain in FPS, so now we're at 60FPS, but frame timing is perfect due to the addition of the extra card and you also have an additional 4 GB of VRAM to work with.
Either way your post is just utterly ridiculous in the first place. People bought the GTX Titan for it's miniscule improvement for $1000.
People will go dual cards if there is a gain, simple as that.
Seems like you're purposely feigning ignorance for the sake of a ridiculous drawn out conversation over this when you are fully capable of understanding what this brings.
Unless there is a language barrier or something but I thought you had a good grasp of the English language.
For such a well informed member of this community I'm finding it extremely hard to understand how you don't understand the thread are only focusing on raw FPS gain....
Are you just not understanding the thread?
Lets say average FPS is 50 but frame timing is terrible to the point where the game is a stuttery mess. Then, lets say we get a 20% gain in FPS, so now we're at 60FPS, but frame timing is perfect due to the addition of the extra card and you also have an additional 4 GB of VRAM to work with.
Ah, cool. This seems like a direct response to tweets I made pointing out that there is some irony in the "4GB means 4GB" campaign because of the way both camps count memory in multi-GPU cards. Looking forward to seeing the tech, although I would be surprised if a significant portion didn't need to be mirrored either way.
Do multi-GPU cards need to mirror VRAM at all? I thought it was just because the RAM is on separate cards but on a single card shouldn't both GPUs be able to use the same VRAM (and thus the same contents)?
There is a cost/benefit case you need to remember. RS for the same reason told everyone to buy multicards over the flagship card.
Really? They haven't even been caught lying like NV did, so aren't you jumping the gun? Give them the benefit of the doubt
AFR has its advantage of raw performance, at a penalty to smoothness or frame latency variations and individual vram.
SFR has its advantage in smoothness and maximizing vram on both cards combined. It scales less so it has less raw performance.
I can see where SFR would shine and be the better solution.
I wonder how it would be received by most gamers. Sure it may run smoother, but you will also see significantly less FPS than AFR. This means the cost of that extra GPU may seem less impactful to them, and cause them to complain about the cost to performance they get. Psychologically, that FPS meter is really difficult for some people to overcome.
In traditional SFR memory must be mirrored so it doesn't have memory advantage.AFR has its advantage of raw performance, at a penalty to smoothness or frame latency variations and individual vram.
SFR has its advantage in smoothness and maximizing vram on both cards combined. It scales less so it has less raw performance.
I can see where SFR would shine and be the better solution.
Makes me think that G-sync/freesync is worthless to those gamers. I think large variance in FPS is worse than less max FPS. Just look at those min FPS, that's an improvement. Less variable perception, more time on immersion.
SFR is not the only solution. I think it's the best for strategy games, but not really suitable for other game types, because in some scenes the raw performance can be very limited.
While I didn't do much research in this area, I think pipeline splitting is a good option for some games. One GPU can do the first half of rendering and the other can finish it. Mantle supports GPU-to-GPU memory copy, so this is an interesting technique.
Building a job based multi-GPU pipeline is another option. It will allow to use two physical GPUs as one logical hardware. But synchronizing the queues is important, and it can be very hard.