CUDA, PhysX Are Doomed Says AMD's Roy Taylor

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.
Nvidia killed PhysX when they locked the drivers
"Oh no, you can't have PhysX with an AMD as a primary card"
"But I AM your customer. My second card IS Nvidia"
"No you CAN'T"
"OK then, I DON'T WANT IT".

On the other hand AMD DIDN'T locked TressFX and I thank them. I am an AMD customer (3 PCs, 2 Phenom II, 1 Athlon II) but this period of time happens to be with two Nvidia cards. I was with AMD cards for years (HD3850 CF/HD4830 CF/HD4850/HD4890/HD5670...) but I found an 560ti and an GT240 at half the price and couldn't resist. Well, AMD DIDN't punished me like Nvidia would. That's why I will continue being their customer always waiting the right time to replace my two Nvidia cards with AMD ones.
 
1. Senior PR manager of Nvidia, Bryan Del Rizzo, explained that multi-threading had already been available with CPU PhysX 2.x and that it had been up to the developer to make use of it. He also stated that automatic multithreading and SSE would be introduced with version 3 of the PhysX SDK. PhysX SDK 3.0 was released in May 2011 and represented a significant rewrite of the SDK, bringing improvements such as more efficient multithreading and a unified code base for all supported platforms (Windows 7, Windows Vista, Windows XP, Mac OS X, Linux (not GPU accelerated), Wii, PlayStation 3, Xbox 360.
https://en.wikipedia.org/wiki/PhysX

2. Consoles have to sell to more than just hardcore audience (meaning lots of sales from yrs 3-8) in order to attract devs. I don't think they will even sell in the first two years as mobile kills them. Ouya, Gamepop, shield, steambox, wikipad, gamestick, madcatz, even now even google & apple are entering consoles domain. Never mind the damage phones/tablets have already shown they're doing to wiiu/vita sales (50%off in Q1 for wiiu, OUCH). GDC 2013 survey of 2500 devs show nobody is planning games for next gen or wiiu/vita/3ds (more planning xbox360/ps3 games by far) and all are dwarfed by the 60% planning mobile games. No sales in millions=no devs making games (see vita/wwiu)=no profits for AMD=no magically optimized AMD games proliferating.

3. Umm...See #2.
A. Mobile is already multicore also. Ok, so everyone benefits...Shocker. However sales will dictate if any dev even bothers for consoles this time. They won't. GDC doesn't lie.

B. Most AAA games going multi-platform - Agree but they will be OpenGL/WebGL/HTML5/Javascript etc...So consoles won't matter here either. If you make a game for any of these it can run anywhere with minimal coding. Unreal 3 engine got ported to firefox in 4 days with WebGL and Javascript. Impressive. This is the future, not proprietary (no matter how close to PC) consoles. A game on consoles, even next gen, doesn't just run on a PC. You have to dev for ALL 3 when you make a game on consoles. Not the case when you dev for OpenGL/WebGL/HTML5/Javascript which basically can run on any device with a proper driver or browser.

It's comic this guy bashes NV (their competitor) for constantly making mobile consoles (umm, just shield right?) when they have all next gen consoles showing no love for PC :) It's comic that because of AMD showing so much freaking console love, drivers sucked for the last year, 30% of their engineers got laid off, bulldozer became bullsnozer, XP drivers still suck for multigpu (xp has 38% share still last I checked and are running dx9 paths not fixed by AMD's new driver, only dx10/11 got fixed), enduro still sucks, next gpu delayed etc etc...Consoles made everything they USED to do well suck. Thanks AMD. At the same time the effects of this on Intel is a cpu perf microdrip at $50 extra (raised i74770k etc $50) and an NV who was able to just put off Titan until needed which should have been a release card much earlier. The same can be said for maxwell delay (why release when AMD delays?). NV did NOTHING driver perf wise until AMD got their act finally together in Nov with never settle drivers. Then all of the sudden NV has driver improvements monthly for the next 5 months. If AMD had spent on drivers from the get go instead of consoles (and basically releasing hardware with beta drivers) NV owners would have had those perf drivers a year earlier.

Cuda is growing, not dying. The purchase of PGI last week for Cuda programming will further this cause in HPC etc. Nodoby has dropped Cuda, they are merely adding OpenCL if they can afford to or AMD pays them to (adobe). 65% of the discrete world runs on NV, so they all have cuda already and physx in large numbers in the field just by default. Cuda is gaining as it is taught everywhere because it is FUNDED, unlike opencl with basically just poor AMD backing it. Cuda meanwhile has had great funding for 7+yrs which is why its entrenched so deeply and used in nearly every PRO app you make money with (adobe, etc). IF it's not directly in the app, it's supported in a plugin. Only toms doesn't pit one against the other (cuda vs. opencl) because that would be UGLY and they couldn't keep loving opencl without looking like fools.
 
1. Senior PR manager of Nvidia, Bryan Del Rizzo, explained that multi-threading had already been available with CPU PhysX 2.x and that it had been up to the developer to make use of it. He also stated that automatic multithreading and SSE would be introduced with version 3 of the PhysX SDK. PhysX SDK 3.0 was released in May 2011 and represented a significant rewrite of the SDK, bringing improvements such as more efficient multithreading and a unified code base for all supported platforms (Windows 7, Windows Vista, Windows XP, Mac OS X, Linux (not GPU accelerated), Wii, PlayStation 3, Xbox 360.
Still]https://en.wikipedia.org/wiki/PhysX
Click to expand...
Still can't use it with an AMD GPU. You're just moving it off your nVidia GPU to the CPU.

Cuda is growing, not dying. The purchase of PGI last week for Cuda programming will further this cause in HPC etc. Nodoby has dropped Cuda, they are merely adding OpenCL if they can afford to or AMD pays them to (adobe). 65% of the discrete world runs on NV, so they all have cuda already and physx in large numbers in the field just by default. Cuda is gaining as it is taught everywhere because it is FUNDED, unlike opencl with basically just poor AMD backing it. Cuda meanwhile has had great funding for 7+yrs which is why its entrenched so deeply and used in nearly every PRO app you make money with (adobe, etc). IF it's not directly in the app, it's supported in a plugin. Only toms doesn't pit one against the other (cuda vs. opencl) because that would be UGLY and they couldn't keep loving opencl without looking like fools.
Click to expand...

The moment they can use both, people use OpenCL. Because FP64 is so much faster on an AMD GPU.
 
I don't know what the fuss is about. Proprietary is always great for the company for a while, but bad for the industry and consumer. What AMD is saying is absolutely true and I for one wish that consumers would widely start taking such considerations into account. But then... "smart consumer" is an oxymoron.
 
History repeat itself, 3DFX (later bought by NVidia) used Glide which have been replaced by DirectX... Honestly at that time Glide was much better thanDirectX.
Now it seems CUDA/PhysX will be replaced by OpenCL.
Just surprised that OpenGL couldn't replace DirectX for games.
 
Just no, CUDA wont get replaced by OpenCL. Although I think CUDA shouldnt be used for games (actuallys its used the most on professional applications). Those saying FP64 is faster on AMD GPUs is true but not every app needs it, so yeah, CUDA wont be dying anytime soon but OpenCL wont stop growing either
 
PhysX sure does save a lot of RnD money for companies wanting a half decent physics engine in their game, though. The only downside to having a propriatory physics engine is the dampening down of innovation and competition.
 
@ radiovan

What I was meaning to say is that when most people have hardware powerful enough, then the physics in games will no longer be used merely for optional cosmetic effects and will start being used for critical gameplay affecting aspects of the games. You didn't say something like that.
 
@radiovan

a game where destruction is the main focus, causing just absolute chaos, that extra that physics adds, adds ALOT to the overall enjoyment in games like that.
 
In other news, "Nazis OK, says Hitler", "PCs useless, says typewriter manufacturer" and "humans overrated, says dinosaur". C'mon guys. Was anyone expecting anything close to a fair and balanced statement from a head of Nvidia's main competitor? Me neither.
 
It is sad that PhysX never got anywhere as it is a really neat addition to games... but because you cannot really emulate it on a console, most games do not support it at all. I think that with the next gen consoles being AMD based it is a no-brainer that they will have some equivalent done via openCL, or even a CPU emulation now that there is power enough to do it.
CUDA on the other hand is still a rather large industry standard. I have needed CUDA based cards for several years now for doing real time color correction, green screening, and other little video filters and transitions. It is good stuff. But, there are finally a few openCL alternatives, and they are quickly catching up in quality and performance, while entering the market with a nice low price. nVidia either needs to open up the CUDA platform so that it is free (or at least extremely cheap) to implement on other hardware... or it is time to enjoy that they had such a great cash cow for so long but admit that it is time to move on.

nVidia is a great company, and I have been a staunch supporter of theirs for a very long time... but if I was buying a GPU right now then I would be going with AMD. If someone as biased as I am is considering moving to the other side then they have a real problem. It is not an issue of raw performance and features... because they have it. The issue is that AMD has caught up in those areas, and offers the same package at a much lower price point.
 


NVIDIA made the only move they could there. If PhysX enjoys widespread adoption on the new consoles, it means that ports will undoubtedly fare better on NVIDIA cards. I very much doubt that NVIDIA would make PhysX work on PCs with AMD cards and thus get rid of another reason to buy their hardware. This is merely their way of attempting to spoil the party, and besides which, the margins on the console side aren't exactly huge. I doubt AMD will let it slide, though - they might view this as anti-competitive if it gains traction.

I don't imagine that NVIDIA will be charging for the "privilege" of supporting PhysX on the new consoles, either.

Roy is wrong about the death of PhysX, and the professional market still values CUDA, so any change will take a lot of time. OpenCL simply isn't being pushed hard enough right now.
 
The micro-stutter issue is still not fixed. The effects have been lessened, but until they finish the job, consider AMD on the hook especially for the multi-GPU customers.

"Unlike our competitor, who’s obsessed with launching consoles in the mobile market, we still love PC gamers and we’re absolutely committed to them. That’s never going to go away. Nobody should have any doubt that we’re committed to GPUs..."

I began to doubt AMD when I crossfired (or rather micro-stuttered) my 5850s three years ago. Even today, after the August 1 driver update, the multi-GPU micro-stutter issue is a work-in-progress. Even with the improvements they've made with micro-stutter issue on a single screen, AMD still hasn't even approached a micro-stutter fix for their multi-GPU customers with Eyefinity setups. Definitely a lack of commitment to people spending the most money on their GPU solutions.

I hope AMD customers keep them on the hook, because if they become complacent with the August 1st not-quite-ready-for-prime-time micro-stutter patch, AMD will be fine with it as well. AMD did not say they would have things kind of worked out with the July (not August mind you) driver release. They said the micro-stutter issues would be addressed. As we all know, they are just kind of address. They've made progress, yes, but still did not deliver.

Let's hope AMD really is committed to its PC customers. I'm tired of not having choices. If AMD really commits to solving the multi-GPU issue, I will consider buying their tech again.

As for now, I've been using 3 generations of Nvidia SLI setups (580s, 680s, 780s) with no microstutter issues. It's no question to me that Nvidia is committed to its PC customers.
 
The thing with the propriety standard of CUDA is that nVidia made and is still making a ton of money through it and has certain system environments locked to it, at least for the near future. As nVidia is one of the main governing bodies of OpenCL I am sure they have no issue transferring over to it in time. While I agree with the comments regarding Intel calling their CPUs, APUs... I do think that Roy Taylor (nice spelling in the article Tom's) should at least wait until his company turns a profit before actively criticising other companies.
 


And yet we still see threads posted by ATi/AMD users who want to know how to get it to work with an AMD/ATi card, strange that.
 
" Nvidia's CUDA is doomed, and PhysX is an utter failure. Why? Because the industry doesn't like proprietary standards."

And yet companies like Microsoft and Apple *debatably the worst proprietary companies of all times * are some of the largest companies in computers.

Don't get me wrong, I don't run Apple or Windows (I run Linux Debian on GNOME) but to even try to sell this with the above statement is just a pure lie
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom