Report: Nvidia Prepping Maxell-based 750 Ti for February

Page 3 - 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.


I seem to remember another company, back in the mid to late 90s. It was named 3Dfx. It had a TON of money in the bank and was on its way to winning a lawsuit against nVIDIA.

It went out of business and sold its assets to nVIDIA. They also settled the IP patent claims out of court with nVIDIA (grating nVIDIA shares to top 3Dfx investors in exchange).

I seem to remember proprietary technologies such as the Glide API as well as the method which the Voodoo4/5's used in order to provide Cinematic effects being proprietary. I also remember the company, 3Dfx, thinking it was better than the industry and that it didn't have to follow trends but rather the industry followed their trends. So 3Dfx stubbornly refused to work on 32bit color (believing that 16-bit was enough). This reminds me of the nVIDIA of today. Still releasing proprietary technologies at a time when Open solutions are the industry trend (OpenCL, Mantle, FreeSync etc).

PhysX is dead (you'll begin to realize this more and more as Mantle titles are released). CUDA is dead with OpenCL all but replacing it under most consumer grade applications. G-SYNC will likely suffer the same fate as well.

So what does nVIDIA have? They have no CPU, a weak compute architecture (compared to the competition) as well as proprietary technologies that are just not gaining much traction in the industry in order to gain a near monopolistic advantage over competitors.

Seems to me nVIDIA isn't looking towards the future one bit. Seems to me they're living in the here and now. AMD, on the other hand, took a heck of a lot of risks with their APUs, HSA, Bulldozer/Vishera architectures, pushing Mantle, GCN, adopting OpenCL instead of their own Stream/Brook+ compute languages etc.

Seems these risks are about to pay back big time.
 


Mantle and GCN are open? How so? Mantle doesn't even work on older AMD cards so how is it open?
 


Not working on older AMD cards is =/= to not being open. Also, older cards are not supported because it would require a driver update and additional support built into Mantle that isn't financially intelligent considering the smaller number of cards that are even out there to support (per AMD).

Mantle is an open-source API and thus is an open platform; this is why people keep saying Nvidia could support it if they wanted to, because technically they could via drivers if they chose to, in much the same way they currently support various versions of DX.

The Glide API was proprietary BECAUSE no other solution existed, not because they wanted to keep it proprietary, necessarily.

Nvidia could easily have updated Glide and tried to run with it, but DirectX (and OpenGL) came along and it didn't need to.

There's nothing to say that Nvidia could not also provide an API similar to AMD's as an *option* for developers. There's also nothing substantiating the bad issues brought up even here on Tom's articles suggesting that this would be in any way worse than DirectX.

In fact, DirectX might be stifling innovation where an Nvidia-specific API and AMD-specific API would allow better cross-platform support (like Linux, for example) for developers.

I think Mantle is clearly unique in the sense that it hasn't REALLY been done before; Glide doesn't count. Glide existed out of necessity, OpenGL came along out of necessity and we ended up with the all-in-one-not-quite-as-good-as-to-the-metal-API DirectX being the most universally supported, mostly due to it's compatibility with Windows being ensured by MS.

I think we need to stop the Nvidia vs. AMD argument. They both have unique features that bring value to the table and they will ALWAYS try to one-up the competition and leverage that unique feature because that's how you survive in a competing market.

Mantle might be great, G-Sync might be great, and there's surely more to come. We have hardly seen enough to know how this will play out, but there's a strong chance it won't be another 3DFX.
 


A driver update and additional support? That sounds really difficult and are saying that no one is using any card older than the 7xxx series?
 

amd's majority gpus should be vliw5 (and vliw4, to a lesser extent) and nvidia's should be fermi. gcn is a small minority in pcs. overall dominant gpu should be intel's hd graphics, hd3000 and 2500. mantle doesn't support any of them.
 


But its open and can be used by Nvidia (which don't have GCN cores) so why can't it be used on older AMD cards (which also don't have GCN cores)?
 

i just re-read two articles from two sources, amd.com and anandtech, who discuss mantle - neither mention it being open but it being more or less tied to gcn. iirc it was someone in the mantle unveiling event that claimed that mantle can be used by nvidia. but i knew about these already. afterwards amd didn't reveal much more information on mantle. i just don't have enough information to claim that it's "open". according to what i know, no low level api can be open unless nvidia incorporates gcn in their cards (in case of mantle implementation). if it can be, it's not a low level (at least not that "low") api. any amount of "abstraction" will eventually lead to a "directx".
 


Whichever way you look at it Mantle is AMD & GCN only and no amount of BS from AMD fans will change that any time soon.
 


I have no idea what this is supposed to mean. But the majority of Nvidia and AMD cards are not necessarily their last generation.

And you don't plan a product launch (Mantle) during a time when you have a new generation coming out and have just released the last of the current generation around 2 generations ago; it doesn't make sense.

Their choice to support the older cards (if you actually read the interview w/AMD right here on Tom's) comes down to the cost and reality of supporting older drivers: it just doesn't make sense to throw money at it when they want you to move away from older cards and buy the new ones. There's no secret here: if they supported older cards then these users would get even MORE life out of those cards, delaying even further their next AMD GPU purchase.

http://www.tomshardware.com/reviews/amd-ama-toms-hardware,3672.html
There is also nothing to say that Intel couldn't chose to support Mantle in their graphics drivers, however it wouldn't make a lot of sense with the way their integrated solution already performs w/DirectX.




I keep reading the same misinformation over and over on Mantle.

1) IT'S AN API - What it supports isn't even the right terminology. What it is SUPPORTED BY would be far more accurate.
2) A graphics API can be supported via driver changes IF that card manufacturer decides to support it (i.e. Nvidia). It has NOTHING to do with AMD supporting Nvidia cards.
3) GCN is simply AMD's GPU architecture. There is nothing that says the API features require GCN to actually function, only that AMD will only be supporting it via their GCN chips. This does not translate to "Nvidia must now use GCN architechture" any more than OpenGL meant ATI had to use the same chips as 3DFX/Nvidia.
4) OPEN in software terminology means the code is open-source. It doesn't in any way suggest that it is magically compatible with every piece of hardware ever created. It means the code is available without paying for it to be studied so that the code framework can be clearly understood and properly supported by developers who wish to use it.

The API is an "Application Programming Interface" and in the context of graphics hardware it handles the instructions sent to the GPU driver essentially translating what the game or other graphical program is requesting.

An API that is "close to the metal" refers specifically to the fact that Mantle is designed with the specific functions of the card in mind; it's a better translator than DirectX. DirectX has a far more "generic" set of calls (instructions) which are more loosely in control of the driver than a device-specific API. This means they are less efficient at making requests to the GPU than Mantle would be.

On this page, the AMD reps answer the question fairly clearly: http://www.tomshardware.com/reviews/amd-ama-toms-hardware,3672-9.html

Q. How's Mantle going to be compatible with Fermi/Kepler/Maxwell etc. or Intel's HD graphics? Will you be forced to maintain backwards compatibility with GCN in future?

A. Mantle depends on the Graphics Core Next ISA. We hope that the design principles of Mantle will achieve broader adoption, and we intend to release an SDK in 2014. In the meantime, interest developers can contact us to begin a relationship of collaboration, working on the API together in its formative stages. As for “backwards compatibility,” I think it’s a given that any graphics API is architected for forward-looking extensibility while being able to support devices of the past. Necessary by design?
Click to expand...

ISA: Instruction Set Architechture. If GCN ISA support was baked into Nvidia drivers (essentially translating GCN input to Kepler input) then it could, in fact receive instructions via the Mantle API.

It's essentially the same concept as the classic pass-it-on effect: you tell someone who tells someone and so on, and once the message is received it's not entirely accurate, but maybe close enough (this is DirectX).

Mantle is the equivalent of walking right up to the desired recipient and telling them.
 


So much for price/performance eh? It seems as if AMD are the ones trying to force people to give up their old cards and spend money just so that they play a couple of games with this new API.
 
Something can't be low level and not hardware specific at the same time. Mantle is either a mid level API that AMD is going to make drivers for to make it feel it has low level performance or they are shooting people a line of BS towards those who don't know how APIs work.
 


I agree. I don't see how this could be true low-level if they plan to support it with future cards. It would be fairly restrictive. It might be TOO hardware specific to the extent that game integration would result in bad compatibility after future updates and it would also mean that all GCN GPUs would need to be extremely similar, which technically they are, but it would stifle innovation in GPU design for the foreseeable future.

I think it's fair to say it's definitely lower level than DX due to driver integration but not true to-the-metal.
 



So why did you disagree with my earlier comment about Mantle being GCN only? :heink:
 
What I mean is that I think AMD has taken the position that it is restricted to the GCN architecture because they've built it with the GCN ISA at the center, not because it's technically impossible for other hardware to take advantage of it.

If, for example, they created a go-between API to "translate" GCN instructions to those specific to any other GPU architecture, then other GPUs could be supported.

In AMD's case they've chosen not to do this as it would also mean major driver revisions.

In Nvidia's case it would require Nvidia to support the GCN ISA in the same way.

As AMD themselves have said, it is entirely possible for another GPU manufacturer to support this.

It should just be clear that an API can be supported in many ways; the limit to GCN is really a myth because it can be supported. But just like Google isn't going to write code to support Maps API for iOS and that is instead up to Apple, the same can be said for Mantle.
 


Well if its possible to run Mantle on non GCN GPU's then why not enable it on older AMD cards? It would go a long way to convince the sceptics that Mantle is not proprietary to GCN wouldn't it?
 


See, this is where my answer to your last question gets fuzzy and some context is needed:

To support it on older cards, AMD would need to do the same thing Nvidia or Intel or anyone else would need to do to support their GPUs: write code that would essentially translate the Mantle ISA to their ISA. Chances are, this extra step would result in Mantle not being quite as efficient as on the GCN chips it's designed for, but that's really my own speculation and only a real-world test would tell us for sure.

If indeed it DID bump up older cards 30-40% in performance, then it prolongs the life of that older GPU, which AMD would not want from a business sense standpoint. If it performed poorly on older cards then they've wasted money on R&D and the publicity might be a bit negative. It's kind of a lose-lose for them.

GCN is really their GPU flagship right now, so from a marketing perspective, they want to leave the old cards old and focus buyers on GCN. They don't want to explicitly state "Mantle could work for anyone and everyone" because it's not really directly up to them in some cases and it would not be true on their own older cards w/out additional development, which they don't want to invest in.
 


So why bother with ANY PR that insinuates Mantle is an open API or even that it could be open when they can't even open it up for their own cards and customers?
 
Open API means the source code is open; it doesn't cost money to access the code or ISA. This means anyone can develop around this API, rather than license it fromAMD before being able to develop for mantle.
It IS open because the code is available.
Open does not automatically mean "universal" in terms of support, but because of code openness, in theory support increases.
 


It's NOT open because the source code is NOT available, not yet and if it ever is made available it still requires GCN cores to work. What AMD reward program are you on by the way? I only ask because your posts ignore common sense and seem to be pushing the AMD PR BS.
 


Sorry, I wrote that wrong: Open API means the API *INSTRUCTION SET* is open. This means anyone can develop for it.

Open-source code is really only relevant if you want to *modify* the API itself.

Having an open API still allows GPU manufactures to adapt their ISA to take advantage of Mantle.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom