Digital Foundry Article Technical Discussion Archive [2015]
- Thread starter Deleted member 11852
- Start date
-
- Tags
- digital foundry
- Status
Presumably this is on the Jaguar core, which physically supports 128-bit registers. It externally supports 256-bit SIMD, but cracks them internally into two separate ops.
The ISA exposes 8 or 16 registers, but each core has a pool of 72 rename registers with which to map it.
The cores are single-threaded, so the whole core context switches if it is dealing with full-scale threads. Naughty Dog's fibers allow the six OS-visible worker threads change what they are doing without incurring the cost associated with such frequent changes if they were applied to an OS thread.
Rikimaru
Veteran
It's for tools. There is no much sense using it on Jag, is not it?
AVX-256 does not seem like a benefit, but I do not recall AVX-128 coming up either. The Insomniac presentation I'm aware of focused on SSE and 128-bit registers.
Yeah but at least you get 3 operands with avx right? Still could be useful over SSE for 128-bit.
Yes, AVX128 is somewhat useful. It has VEX prefix for multiple operands and that slightly reduces the register pressure and unnecessary mov instructions. 256 bit wide AVX also slightly improves the instruction cache utilization (as one instruction does double the work). Jaguar (and Bulldozer/Piledriver/Steamroller) all need two decoder cycles to decode a 256 bit wide AVX instruction (fastpath double). 128 bit wide instructions need only one cycle. So there is no gains for instruction throughput.
AVX doesn't include many new instructions over SSE4.X. Mostly it just adds double wide float math (and that does not benefit Jaguar).
According to Agner's guide, one internal difference is that Jaguar has MOV elimination only for 128-bit registers. 256-bit instructions might experience a regression in terms of more instructions being tracked by the engine.
So after reading about the issue with black crush on some games in some DF articles and finding out that my HDTV is full range compatible -calibrated it for standard RGB (game mode) and full RGB (computer mode)- I got a couple of Far Cry 4 screengrabs on my Xbox One and here is the result:
This is how Far Cry 4 looks in Standard RGB mode:
http://1drv.ms/1yc9Y5C
And this is how it looks in Full RGB mode:
http://1drv.ms/1HZhhh4
Additionally, I found out while playing at a friend's house, that his pirated version of Far Cry 4 for the PC has a HUGE black crush issue.
And I mean it. The video settings, Contrast, Gamma, Brightness...were all set at default, so it wasn't a problem of the PC -other games looked perfect on it-.
The game ran fine otherwise, but all his games looked near perfect except Far Cry 4, which made shadowed zones look pitch dark even in daylight and the game was almost unplayable because of that, except in broad daylight.
It seems to be a security measure intentionally created by developers to make life harder for users who downloaded the pirate version. I wonder if there is a way DF staff could find out if it's true. But everything points out to this theory.
This is how Far Cry 4 looks in Standard RGB mode:
http://1drv.ms/1yc9Y5C
And this is how it looks in Full RGB mode:
http://1drv.ms/1HZhhh4
Additionally, I found out while playing at a friend's house, that his pirated version of Far Cry 4 for the PC has a HUGE black crush issue.
And I mean it. The video settings, Contrast, Gamma, Brightness...were all set at default, so it wasn't a problem of the PC -other games looked perfect on it-.The game ran fine otherwise, but all his games looked near perfect except Far Cry 4, which made shadowed zones look pitch dark even in daylight and the game was almost unplayable because of that, except in broad daylight.
It seems to be a security measure intentionally created by developers to make life harder for users who downloaded the pirate version. I wonder if there is a way DF staff could find out if it's true. But everything points out to this theory.
I'm not quite sure how a dev could implement something like this - I can see why you think it may be a copyright protection, in the old days the protection methods used to protect videos was a pulsating dark screen but that coule be overcome by a modified lead...I just can't see why this was done - I mean, if it were possible, why not just set the brightness so you couldn't see anything?
I am not sure a PNG screenshot is 100% representative of what you actually see on your TV, more so if you use a xbox console and want to compare levels of crushed blacks.
shredenvain
Regular
What file type do the consoles use for their framebuffer anyway?
Rurouni
Veteran
Btw, how many games are there on PS4 that have AF patched? Recently it was DmC. If the sample is big enough, maybe DF should make an article about it and try to measure whether adding AF impacted the FPS or not. Probably throw in some PC benchmark for comparison on various AF setting and also ask a dev (would be better if the dev is the one that ship a game without AF on PS4, but with AF on X1, and patch it later) about this whole PS4 AF thing.
Seeing that AF (patched) had no impact on performance on DMC and Dying Light, it would seem from my point of view, there really wasn't any issue... other than a handful of developers misunderstanding on how to implement it correctly.
Seeing how R* just added POM back to the console editions of GTA V (without any impact to the previous patch that improved PS4 performance), it would seem these missing effects/techniques are more developer error, than hardware specific (issue).
Article from who? Sony?
Anyhow, a thread was created on this issue and was closed after Sony and few developers responded at GAF (see comment below).
Iherre at GAF
djskribbles
Legend
Of course AF has an impact, but it appears to be very minimal (as I have always expected). Two games have now added a good amount of AF with no noticeable impact to performance (DmC and Dying Light).
I think the other thread was rightfully closed, because it has pretty much been proven that some developers simply didn't turn it on. Hopefully the other half a dozen or so games get fixed, too.
I think the other thread was rightfully closed, because it has pretty much been proven that some developers simply didn't turn it on. Hopefully the other half a dozen or so games get fixed, too.
Last edited:
It would appear the resource impact is either minimal or already consumed with the AF effect off, as adding it back on isn't causing any issues for the games. Although, that said, perhaps these game can patch it in because they have spare bandwidth, and other games will notice more of an impact leading to avoiding the patch?
I dont see much difference between the twoSo after reading about the issue with black crush on some games in some DF articles and finding out that my HDTV is full range compatible -calibrated it for standard RGB (game mode) and full RGB (computer mode)- I got a couple of Far Cry 4 screengrabs on my Xbox One and here is the result:
This is how Far Cry 4 looks in Standard RGB mode:
http://1drv.ms/1yc9Y5C
And this is how it looks in Full RGB mode:
http://1drv.ms/1HZhhh4
Additionally, I found out while playing at a friend's house, that his pirated version of Far Cry 4 for the PC has a HUGE black crush issue.
The game ran fine otherwise, but all his games looked near perfect except Far Cry 4, which made shadowed zones look pitch dark even in daylight and the game was almost unplayable because of that, except in broad daylight.
It seems to be a security measure intentionally created by developers to make life harder for users who downloaded the pirate version. I wonder if there is a way DF staff could find out if it's true. But everything points out to this theory.
yeah whats the difference?
- Status
Similar threads
