-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
Performance Analysis: Assassin's Creed Unity (Digital Foundry)
- Thread starter cb1115
- Start date
dr. apocalipsis
Banned
That hobby you have of being always late have a term for sure, I will check it. Later, so you can get it in time.
You dont need 32GB of DDR4 RAM and an octocore Intel CPU either to run this game over 30fps with better settings than console.
travisbickle
Member
so look up the differences between coding an OpenGL renderer and a DX renderer, now you have an idea of how code needs to differ between platforms, then that's not taking into account any differences in the rest of the api for say interfacing with the system's OS for networking file i/o, trophies / achievements. Setting up job systems and managing multiple threads, controller handling accessory handling... I can go on and on.
According to him optimization doesn't matter because it's the same code:
So code magically works across platforms "mostly the same or better" performance.
Simply put there are not enough face palm GIFs on the Internet to demonstrate the fail.
Crimson_Fate
Member
You need to compare multiple GPU's at a constant CPU speed to see an impact which is meaningful.
http://www.gamersnexus.net/game-bench/1692-assassins-creed-unity-gpu-benchmark
So, they work in this hardware for 4 years and just the last month to say 'oups,we aimed too high'. Very competent.
bronzodiriace
Member
in 24 frames like cinema we trust
Lol, makes sense.
More like "oops, we forgot that we make open world games that, by nature, don't run in controlled environments and are subject to a bunch of player induced variables that might tank the fps. We also forgot that we have been showing the game on top of the line pcs that run laps around what these consoles can achieve, on top of it being advertised on several controlled vertical slices of the game that in no shape or form will correspond to the end result, especially on consoles. tee heee, they will buy it anyways, screw it. Just make sure no one gets to review the game until it's out there and on people's hands and we're golden."
dr. apocalipsis
Banned
Tell me how would you manage to make a given architecture with such gflop and many gddr5 run worse a given code than same architecture with DDR3 and M$ logos all over it.
Oh, wait, moneyhats.
Some day Sony fanbase will stop to blame devs for everything that goes wrong. That day isn't close.
Still waiting for some dev to come here and tell us all than PS4's CPU performs better than Xbox one. Not that the console is better overall, that is a thing clear enough. Can you tell the difference?
JohnnyFootball
GerAlt-Right. Ciriously.
Aside from Splinter Cell, Ubisoft has traditionally had more marketing deals with Sony. Farcry 4 next week will have Sony marketting behind it ironically.
I remember how people commented that Farcry 3 ended up being what everyone hoped Assassins Creed 3 would be!
Maybe we will have the same thing happen for AC:Unity and Farcry 4.
Chobel
Member
Yep, now do you know the meaning of "There is nothing about the XBO that allows it to do anything over the PS4" and "In essentially every way, the PS4's hardware is more powerful than the XBO's."?
Just out of curiosity.
Simply Sarah
Member
Yeah, it's super clear they have never worked with cross platform or cross API code, so as you've said there's no point putting much value in what they're saying.
I wouldn't waste more time on that, mate. We have given those examples over and over again, and still isn't enough for some people.
You need to compare multiple GPU's at a constant CPU speed to see an impact which is meaningful.
http://www.gamersnexus.net/game-bench/1692-assassins-creed-unity-gpu-benchmark
Geez, I would hate to see 50% framerate dips on a platform. I guess even PC version is hit with extreme changes in framerate.
Haha, you're cute. Ignorant, but cute.
Have fun with your fantasy land, I guess!
Well why would an SQL query with sub selects be slower than a query that gives you the same result set but uses joins with indexed tables? The simple answer is that there is often more than a single way to get from a -> b and the fastest way in API 1 might not be the fastest way in API 2 even when running on the same hardware.
dr. apocalipsis
Banned
Thread name:
Virtual testing of PS4 and XBO GPUs prove PS4 has bigger grafix numbers :
I fully agree with that quote in that context.
Now keep talking about CPU. Show me where does he, or any other dev, say that Xbox's CPU is worse essentially in every way.
I would be more than happy with a steady 24 FPS in this game.
dr. apocalipsis
Banned
So Sony's fault for a lackluster API regarding it's ability as AC:U player.
Lazy Sony
Simply Sarah
Member
Is there even a point to all your posts beyond trying to shoehorn why badly coded game X runs slightly less badly on platform A compared to platform B into reason Y?
So Sony's fault for a lackluster API regarding it's ability as AC:U player.
Lazy Sony
Yeah, you're either utterly clueless or just completely trolling.
So Sony's fault for a lackluster API regarding it's ability as AC:U player.
Lazy Sony
Yes, it must be Sony's fault that one game (which runs like shit on both consoles) runs worse on theirs. Clearly it's an API issue and not..you know....Ubisoft releasing a shitty unfinished product.
Hope you're being sarcastic there, otherwise...
SapientWolf
Trucker Sexologist
I don't think it's the CPU. If it was, the PS4 could move up to 1080p since a higher resolution shouldn't have an affect in a CPU limited game. The relatively small difference in CPU capabilities alone doesn't explain the wide performance gap we saw in the DF video. A member of the AI team even stated that the AI isn't using a lot of processing power. It looks like most of them just go through canned animations, and the people in the background don't have a lot of detail or behavior. The only people that you interact with are the ones you are pushing out of the way.
Also, the PS4's low point was when you had a lot of shader effects on screen (https://www.youtube.com/watch?v=clQfCP3NFuc#t=45s). Similar story with the XB1, and I noticed that there was no volumetric lighting coming through the window in the XB1 scene. So it may be the case that the PS4 has higher graphical settings. There wasn't a large crowd on scene doing complex things. The scene with the large crowd at the end kept a steady 30fps.
The team wrote a white paper on GPGPU, so it's not like they ignored it. There's just a limited subset of algorithms that you can use it on (i.e. well suited to SIMD). I suspect that the problem is memory related, and 32GB of DDR4 and another 4GB of GDDR5 on the GPU provides a crap ton of memory bandwidth. That's something that devs seem to struggle with on the consoles (thus the lack of AF).
Muppet of a Man
Member
As a discrete piece of hardware, X1's CPU is clocked 150 mhz high than PS4's CPU and happens to be exactly the same CPU. It is known that, by itself, it has a minor advantage. There was a question about how the X1's CPU performed functionally, especially prior to the SDK updates and the relinquishing of the in-game Kinect CPU reserves. Matt and others said it was easier to get performance out of PS4's CPU in prior threads. This may have been due to the memory subsystem setup: the fact that PS4 actually leverages hUMA with its unified pool of high speed memory and volatile memory pointers for information tagged as needing to both be worked on by the CPU and the GPU. Not having to copy information back and forth between pools of CPU and GPU memory means less copying/memory overhead, less bandwidth usage, less latency, and more available CPU cycles and GPU cycles to do productive work with. This could be one possible explanation why in most scenarios, the PS4 CPU outperforms the X1 CPU, even though it is clocked higher. The utilization of the PS4 CPU (and by extension GPU) could just simply be more efficient in most cases.
Kazuma Kiryu
Member
This video is amazing.
Chobel
Member
Are you kidding me? Are even reading these comments I linked? "In essentially every way" means virtually and non virtually.
His other comment (which I also linked) was in "Sniper Elite 3 Digital Foundry Face-Off", do you see "virtual" in the title?
And FFS! Why are you ignoring that he said right in this thread that there is no excuse for this?
No one is arguing that PS4 CPU is more powerful, only you twisting what we're saying so it fits your narrative.
Muppet of a Man
Member
Lockjaw333
Member
You need to compare multiple GPU's at a constant CPU speed to see an impact which is meaningful.
http://www.gamersnexus.net/game-bench/1692-assassins-creed-unity-gpu-benchmark
Ahahaha that low benchmark is something else. Literally made me laugh out loud.
That makes sense. I would eventually like to see someone create some +-% of scale depending on the CPU and GPU used, based on some hardware baseline. I know that's easier said than done, but I bet some approximation of performance could be achieved.
vpance
Member
Lol
It really does deserve more love.
It's not an easy decision but there are limits that even Ubisoft should not be allowed to exceed
OMG that Daytona music was perfect.
Dude, common man I'm not trying to start a seperate debate but right now I'm looking at at a k-means clustering algorithm that has been optimized to run on a pretty average GPU and seeing hell of performance increase.
Right now, infront of me. Please do not tell me that AI type algorithms arnt suited to GPU use.
I dont get what you trying to show me, I'm saying AI is very well suited to GPU use???
Cizeta-Moroder
Banned
"We should act on the interest of the power gap..."
"I AM the power gap!"
"I AM the power gap!"
GPUs are very good at embarrasingly parallel algorithms, that have minimal branching (I.e. decision making). AI code is basically all decision making, it is the absolute worst sort of thing you could ask a GPU to do unless it is disruptively trivialised.
Incidentally, doing k-means on a GPU is probably a bad idea too and could be done more effectively on a CPU.
Not when you offset the compute intensive parts of the algorithm to the GPU. it's MUCH faster
It's more complicated yes, because you of speed that GPU and CPU communicate is considerably slower. But even that time cost of that is offset by the speed gains of using GPU compute.
May I ask why you think GPU Compute could be done more efficiently on CPU only?
GPU Compute is is also good for decision making algorithms, I'm heading out of work but will edit this or respond shortly.
TheAngryMonkey
Member
It doesn't matter how fast the ram is if the CPU is the bottleneck
Ashad Mamood
Banned
If the game was programmed to take advantage of each consoles abilities then there is no way the X1 version would be superior. This means the PS4 version isn't optimised and more than likely just ported over from the X1 version
The problem I have with doing k-means on a GPU is because k-means would require frequent (depending on the rate of convergence I suppose) reduction operations and global synchronisations (which can only be done by re-launching the kernel). Those reductions would also not generally operate on the data set uniformly, but rather one would need to be done for each cluster of items, which could be distributed all over the data set at random.
If you have an absolutely massive data set, then I imagine that the compute might overshadow the reductions, but I'm still not convinced a GPU is the best platform suited for this task which is more about making decisions and moving data about rather than raw compute.
I absolutely disagree that the GPU is good for decision making algorithms, it really does not make sense based on how the hardware actually works. Often you'll have a 16 or 32 item wide SIMD execution pipeline that runs through the data. This means that branching is extremely costly relative to the peak performance you could be getting, because each work item should be performing the same instruction at the same time.
Ashad Mamood
Banned
If Ubisoft actually fully utilized both X1 and PS4 there is no way in hell the X1 would perform better. So what are are Ubisoft doing with the extra 40% power of the PS4 GPU and the benefits of unified GDDR5 Ram? And if the CPU was bottlenecking this game so severely then why didn't they just lower the NPC account 5-10%???
The Maverick
Member
You're like those people that say if God was real, he would drop a million dollars out of the sky right then and there. And when it doesn't happen, you use that as proof that you were right all along.
So Sony's fault for a lackluster API regarding it's ability as AC:U player.
Lazy Sony
As I have said more than once now it is not to do with the quality of the API just that they are different and something that may be quick and efficient in one may not be in another so it requires tuning for each platform.
marianitten
Banned
TotalBiscuit about the PC version:
https://www.youtube.com/watch?v=SgpzT5V5Mgs
He usually do the "Let's not play" with terrible and unplayable games. Usually indie garbage pre-alpha.
Just consider, that he has an Amazing PC rig. If he is having problems with that....
And people blaming Sony/MS for this.
https://www.youtube.com/watch?v=SgpzT5V5Mgs
He usually do the "Let's not play" with terrible and unplayable games. Usually indie garbage pre-alpha.
Just consider, that he has an Amazing PC rig. If he is having problems with that....
And people blaming Sony/MS for this.
mugurumakensei
Member
he's a bit misinformed. ac: u has fxaa, msaa (very taxing), and txaa(even more taxing than msaa). As well, he seems to call anything less than 1080P/60FPS bad performance. as well, benchmarks show the game is cpu bottlenecked which is why he can never hit a locked 60fps. its beyond any current processor. in cutscenes though, bokeh dof tanks gpu performance explaining the tanking there.