consoles are all going x86 chip and AMD GPU
#1
Posted 21 February 2013 - 03:13 AM
#2
Posted 21 February 2013 - 04:58 AM
I wouldn't be surprised if the new XBox is also using a system-on-a-chip (what AMD calls an "APU"). The fact that AMD offers these all-in-one chips, and NVIDIA doesn't, might have been the deciding factor in the new consoles going with AMD (this is just a guess; I have no special knowledge of this). There are a lot of benefits to systems-on-a-chip - lower power requirements (hence also less heat generated), lower-latency communication between CPU and GPU, it's easier to use a unified memory architecture since the memory only has to talk to one chip, and it takes up less space in the device.
#3
Posted 21 February 2013 - 06:44 AM
#4
Posted 21 February 2013 - 07:26 AM
#5
Posted 21 February 2013 - 07:46 AM
at this point, there is no shortage of games for any of them
#6
Posted 21 February 2013 - 09:35 AM
You can buy a hell of a lot of consoles for that
Hell you could put together a console farm for that
#7
Posted 21 February 2013 - 09:45 AM
#8
Posted 21 February 2013 - 11:51 AM
Their tech specs isn't anything so shining. First of all AMD CPUs are a lot behind Intel's ones right now (not just in performance, but also in technology) - their 8-core systems have hard time beating Ivy Bridge i5's ... although I like their idea "Need more cores!", it doesn't work if they won't scale the rest of stuff accordingly (see Bulldozer epic fail with caches).
Using graphics memory instead of standard memory is not as clever idea as it seems to be. Okay it will work for graphics, but doing wild allocations/delocations on GDDR5 is a lot worse thing that doing it on DDR3. Running some fully managed code on this might be a bit of hell (especially at time when garbage collector comes in) - of course it probably won't be problem number 1 for games, but it can be problem in few years when they'll be running ... peak bandwidth also doesn't say that much.
1.84 TFlops gpu doesn't say a ****. And I mean it, peak TFlops rate are measured as "how much MADD operations can we make in one second" and multiply this number by 2. So it sounds nice, but unless they say what specific model is, it's information with value of 0.
If you don't know how to speed up application, go "roarrrrrr!", hit the compiler with the club and use -O3 :D
#9
Posted 21 February 2013 - 12:43 PM
#10
Posted 21 February 2013 - 12:59 PM
tyree said:
I read about AMD's fusion, or APU brand for desktop PCs a while back. The benchmarks were horrid, on par with integrated chips. I just did some recent checks on AMD's new "Trinity" brand and the benchmarks have improved, but are still lacking when rendering 1080p. I've always liked the idea of simplifying the architecture into a single chip, but if performance is important then sticking to a dedicated card is the way to go for now. Though I do imagine in a couple years, like what happened with sound cards, video cards will become obsolete. Now if they can make those chips small enough for phones... profit!
#11
Posted 21 February 2013 - 02:43 PM
#TheNut - I don't think it's great idea to pack 2 chips into 1 like this. There still is gpu -> and thats still horrible. Gpus are a big bunch of mess, don't get me wrong now - I use OpenCL and CUDA - but these languages are bloody mess compared to standard C, working with them feels like scratching with toe on left feed at your right ear (and I don't even mention debugging, where most useful is the printf function - because debugger is garbage compared to anything on standard CPU). So honestly I believe more into Intel Xeon Phi or such platform.
If you don't know how to speed up application, go "roarrrrrr!", hit the compiler with the club and use -O3 :D
#12
Posted 21 February 2013 - 05:23 PM
Vilem Otte, on 21 February 2013 - 11:51 AM, said:
I disagree. There is plenty you can get up to on a fixed-hardware platform with low-level APIs, that you simply can't do on a PC where you have to work with an abstraction of the hardware and a D3D/OpenGL driver that you don't control, that you don't know what it's doing and it doesn't know what you're doing. I could give a lot of examples of this on PS3. We're still getting the hang of the PS4 HW, but I'm sure in a couple years there will be a lot of examples on it too.
#13
Posted 21 February 2013 - 07:34 PM
On the other hand you actually "can" (in really REALLY strong brackets) do the same on PC, but you have to do it for each piece of hardware out there (or at least those that you're going to support). It isn't worth it for a game project, but F.e. some driver project with standardized interface (open source drivers for AMD are out there and quite useful). But this is more-likely because situation in GPU is very messy, F.e. in CPU part you can work with assembly because there are some solid standards supported by vendors ... so it's more likely problem of GPU vendors on PC.
If you don't know how to speed up application, go "roarrrrrr!", hit the compiler with the club and use -O3 :D
#14
Posted 21 February 2013 - 08:09 PM
Vilem Otte, on 21 February 2013 - 07:34 PM, said:
Only if you care about portability.
Vilem Otte, on 21 February 2013 - 07:34 PM, said:
Yes, I haven't looked at the open-source Linux drivers, but it would definitely be interesting to see what you could do by taking over some of the driver responsibilities in your own app. I suspect it's still not quite as much control as on a console though, since in a PC environment you have to allow for multitasking. On consoles, you still have to cooperate with the OS (it will take up some memory and compute resources, has a hook to render stuff over your frame, etc.) but it's a limited amount of interference.
#15
Posted 21 February 2013 - 09:55 PM
There are different kinds of multitasking implementation, there are different OS kernel implementations (and so different ways to communicate with drivers and thus devices), etc. - and basically this will be based on specific OS. You actually *could* give user full control over hardware, but implementing this in driver can also be bloody mess, and it will most likely kill a vision of having somehow "portable and usable driver" without rewriting it from scratch on another OS, and writing a little more abstract api for using it. Also the question is, would most users use the functionality? Because building such applications for few users out of large user base is definitely not worth it.
Basically it isn't caused by multitasking, it is caused by high complexity and generality of modern operating systems (and not just modern ... all general operating systems*). It is better to follow the way OpenGL actually does - build some kind of (not entirely high-level, neither entirely low-level) abstraction over drivers, and "force driver developers" to implement needed functionality. The result is more usable than giving user full access to device without any abstraction (also this gets really messy with more hardware, because you can end with lots of ugly ifdefs in code).
* I'm though not blaming OS developers, I and my friend once (few years ago) tried to roll my own operating system, ended with very simple Unix/Linux clone. Even though it just was able to run simple elf_i386 binaries and use some standard libraries (smaller variant of libc), it could use SSE/MMX instruction sets and of course mutlitasking (it ran several command lines switchable through F1-F12, because I haven't implemented any graphics interface
I wrote a simple graphics driver for it too, you could just switch text mode resolutions, clear screen, write character, change character settings (color) or move carriage. Note that in standard generic gfx drivers you give user like 1000 times more control that I gave him through my driver, even though OS gave me ability to do anything to hardware from driver - like changing text mode to grahics mode (but not from user application).
If you don't know how to speed up application, go "roarrrrrr!", hit the compiler with the club and use -O3 :D
#16
Posted 22 February 2013 - 10:12 AM
Taos started out on transputers, it could work out network topology in a couple of seconds and optomise itself on the fly.
It feature binary portability and you coded in a custom assembler like language called VP. (virtual processor)
Most of you will have used it at some time, without knowing about it. We wrote a java vm in vp that ran 140 times faster than Sun's so it was bought for lot's of devices.
But in the end we didn't have enough money to deal with the growing numbers of cpu's appearing and we couldn't get the big boys to realise that Taos was the way forward. Imagine a world when you never had to port anything, it just ran on any device.
That didn't fit with what the big boys wanted. They wanted differentiation between devices. They want product that you can't get on any other platform.
So we went bust.
Now as the market starts to merge, as TV's stop being simply a display and start to become interactive, as the difference between a computer and a domestic device becomes smaller and smaller people are starting to realise that we need something. So we are now doing Antix Game Player.
Antix Game Player gives you the binary portability that Android tried to provide and failed.
The bad thing about this is that you have to have a high level graphics API, and Opengles has become the standard. Note not Opengl. OpenGLes
The first two version of opengles were flawed, 1 was fixed function pipeline, 2 had shaders, but was missing things like render to texture.
Opengles 3 is looking very good, I should be getting my first es3 device in a few months. Going to be fun.
D3D is dieing. Microsoft have already started pulling support for it, MVP's have been told they are no longer required, which only leaves us with Opengl
When you write for a PC you have all sorts of problems. The test department at Pure had 24 PC's with different hardware configurations, and all of them had removable hard drives so we could swap OS versions as well. Even then we had issues with some end users machines.
The PC market is a nightmare, when was the last time you walked into a shop and saw a rack full of pc games? Around here you are lucky if there is a single shelf in the whole store.
Rack after rack of XBOX and PS3 games though. So many you struggle to find what you are looking for.
I would not be suprised at all if the big developers started dropping the PC as a viable platform.
#17
Posted 22 February 2013 - 12:05 PM
If you don't know how to speed up application, go "roarrrrrr!", hit the compiler with the club and use -O3 :D
#18
Posted 22 February 2013 - 01:50 PM
Quote
I think that was a stage games went through for a while, but it's been changing lately because of Steam. Mostly, they were tired of piracy which Steam has taken care of and most customers are happy because they can get reasonable prices and don't need it. Most games I care about or want come out for the PC, some are ported later, which I don't care about because they are too expensive when they first come out anyway.
#19
Posted 22 February 2013 - 01:52 PM
Vilem Otte, on 22 February 2013 - 12:05 PM, said:
Stainless said:
Stainless said:
#20
Posted 23 February 2013 - 09:27 AM
the nut your exactly right about sony using existing tech instead of creating new tech. that should make it easier in pretty much every aspect from cost to development
1 user(s) are reading this topic
0 members, 1 guests, 0 anonymous users












