Thursday, January 3rd 2013
AMD to Fix GCN Latency Issues with Driver Updates
Last month, an investigative report by The TechReport found out that despite being faster, AMD's Radeon HD 7950 graphics card isn't "smoother" than NVIDIA GeForce GTX 660 Ti, in that shows signs of higher frame-delivery latency, a theory proven by high-speed camera recordings. Over the holiday, AMD's David Baumann responded in discussions around the web talking about the issue, in which he put AMD's stand.
Apparently, AMD Catalyst drivers still have refinement left in working perfectly with GPUs based on the Graphics CoreNext (GCN) architecture. Baumann explained that GCN, and AMD's older Very Long Instruction Word (VLIW) architectures feature fundamentally different memory management, and drivers that make the most of it are still a work in progress. Baumann stated "Over the early part of the year you'll see a few driver updates help this across a variety of games."
He continued "Additionally, when we switched from the old VLIW architecture to the GCN core there was a significant updates to all parts of the driver was needed - although not really spoken about the entire memory management on GCN is different to prior GPU's and the initial software management for that was primarily driven by schedule and in the meantime we've been rewriting it again and we have discovered that the new version has also improved frame latency in a number of cases so we are accelerating the QA and implementation of that."
Source:
The TechReport
Apparently, AMD Catalyst drivers still have refinement left in working perfectly with GPUs based on the Graphics CoreNext (GCN) architecture. Baumann explained that GCN, and AMD's older Very Long Instruction Word (VLIW) architectures feature fundamentally different memory management, and drivers that make the most of it are still a work in progress. Baumann stated "Over the early part of the year you'll see a few driver updates help this across a variety of games."
He continued "Additionally, when we switched from the old VLIW architecture to the GCN core there was a significant updates to all parts of the driver was needed - although not really spoken about the entire memory management on GCN is different to prior GPU's and the initial software management for that was primarily driven by schedule and in the meantime we've been rewriting it again and we have discovered that the new version has also improved frame latency in a number of cases so we are accelerating the QA and implementation of that."
95 Comments on AMD to Fix GCN Latency Issues with Driver Updates
Does the CPU memory bandwidth have a major effect? I know it certainly affects Crossfire and SLI setups with regards AMD.
Nvidia definitely beat them to the punch on this one (as often seems to be the case with anything driver-related), but it's good to see them "fixing" it so quickly.
Manual playthroughs, even when averaged over several runs are not as consistent as scripted tests, as TR admits. Variations in frame rendering times like we see are exactly the kind of thing you'll get from this way of testing. Still, the result doesn't seem a million miles off.
Must be a pita to manually play the same bit of game over and over, lol.
However, I am entirely sold on frame time testing rather than FPS. I've always considered FPS a misleading metric, and I'm delighted that someone is actually measuring something more meaningful. I no longer need to bother with FPS-based reviews, which is ideal.
Edit:
In that test, it shows the 660 getting a higher FPS, I thought the 7950 had a higher FPS originally.it was actually originally less.This is a very interesting discrepancy:
Tech report:
TechPowerUp!:
How did they get a higher FPS?
I disagree that frame latency on its own is useless.
Say (plausibly enough) all I'm interested in is whether the game is stuttery to play
If there are no frames that take more than 30ms to render, I know that the game is not stuttery. I don't need any more information.
If, on the other hand, the minimum FPS is 30 but I have no latency metrics, I have no guarantee either way.
Also fame latency is great, but what if you have excellent frame latency but only 15 FPS?
I do agree however that frame latency is a great thing to add to FPS benchmarks.
On the other hand, say I render 1 frame in the first minute, and 23,999 frames in the second minute. That's 200FPS across a two minute benchmark. An extreme example, but just to illustrate my point.
Regarding different environments, in a game like Skyrim the load varies hugely from rolling plains to animation-intensive combat to lighting-intensive indoor areas.
PS: That site I think it tested the game at 2560x1600 or somth.
Frame latency, even by itself gives you everything you need to know. Aside from the obvious laggy frame, you can easily compute your FPS from the frame time.
For instance, if all frames are rendered within 16ms, then you know that there's no stutter, and you also know that the minimum frame rate is 60fps. If the frame time is 8ms, then you also know the minimum frame rate is 120fps.
Frame time just simply a much more comprehensive measure of performance than FPS. TR is doing something right.