How much more can one go with graphics cards. I am satisfied with this card and will be getting one. Looks like once the drivers are sorted it'll make even more gains.
Printable View
How much more can one go with graphics cards. I am satisfied with this card and will be getting one. Looks like once the drivers are sorted it'll make even more gains.
Sorry I think my post was unclear, my bad! What I meant is that we are comparing two cards right now, at this point in time Nvidia does not have HBM and AMD does so why should we be going ' Oh but Nvidia dont have HBM so AMDs card is terrible as it still doesnt meant Nvidias non HBM cards power usage', Nvidia CANT release a card with HBM as they havent done the tech and are actually not allowed to use it so right now you must compare the two cards that are out as a whole so ignore HBM as a power usage because AMD did the work and got it into their card and Nvidia didnt, Nvidia lopped off all their compute performance yet its needed in cases and will be more useful in DX12 moving forward, they saved power that way but AMD used HBM so apples oranges really!Quote:
Originally Posted by shaithis
HBM does reduce power usage and its a great tech hence why AMD should be praised, not their fault Nvidia didnt invest and work hard on this particular tech (same as GDDR5... no surprises). Efficiency is about the output power vs the input power, this is subjective in GPUs as there are multiple use cases but yes Nvidia is best at gaming power efficiency but not by much it seems.
No offence but you seem to be on the mindset that AMD are sitting on their asses but really AMD and Nvidia are both focusing on on the next generation where there is an actual process node change so merely competing on a level playing field overall with the fury vs the 980ti is a 'win' for both AMD and Nvidia, you cant go wrong with either party now.
Also to just add a bit of confirmation regarding the importance of compute performance, Nvidia are highly marketing Pascal as 10x more powerful in COMPUTE and that they have made big strides in this area for double-precision, if Nvidia are focusing this hard on compute for Pascal then surely it is confirmed as compute power being significantly more important now which just again shows that AMD were already on a ball they just didnt carve it out several generations ago :P. And a little thing for power effciency, Pascal seems to be produced with the 16nm process but AMD arctic islands will be on the 14nm so it will be interesting to see what both camps provide in power/performance :D.
nVidia have been working on HBM for quite some time and to be honest, I don't think it's going to cause them headaches that AMD got an exclusive for one generation of cards. In waiting they won't be constrained by VRAM size like AMD have done and the next gen will be fast enough to show some true gains from using HBM (I really cannot get across how upset I am that these cards do not shine like a star at 4K where we are supposedly bandwidth-limited)
AMD sitting on their asses? How could anyone ever accuse them of that? It's not like they've been selling the same parts over and over for years is it? ;) It's also not as if slapping HBM on a card was a massive technical achievement......all the hype over these cards and they are essentially slightly modified old tech......and that's before we talk about what they have (not) been doing in the CPU world.
Compute is important to those that need it. Cards should be gaming or compute......we all know what a jack of all trades is. I truly hope BOTH camps realise that and going forward produce silicon with and without a focus on compute capabilities.
Therefore proving how much you know about it. :laugh:Quote:
Originally Posted by shaithis
Well, it doesn't. But...Quote:
Originally Posted by shaithis
Eh??Quote:
Originally Posted by shaithis
Any grain of evidence? AMD clearly made gains in power management on Fiji; HBM alone doesn't cover it considering the number of shaders and decrease in consumption over Hawaii. Plus as others have said, AMD may be on a different node to Nvidia, and although they're more secretive about it the architecture may change significantly for Arctic Islands (assuming that codename is correct). You have no sensible way to predict how cards, probably over a year from now, will compare. Besides completely random speculation of course.Quote:
Originally Posted by shaithis
Or completely misinterpreting Nvidia's marketing material for Pascal, assuming those '10x' or NV-Link gains have anything to do with gaming performance. You have to be very careful with theoretical performance claims to make sure you're actually comparing the same figures. E.g. FP16 vs FP32 like they did on Tegra.
However if it's FP64 performance they're talking about and comparing to Maxwell, then it's not all that surprising; GM200 has 1/32 performance, GK110 has 1/3. That's a scale of 10x difference - see where this is going? They were up against reticle size limits on 28nm for Maxwell; a die shrink would likely allow them to re-add FP64 on the large compute dies. Just a theory of course.
Well, you also need to consider Amd's budget compared to Nvidias. Amd have worked to bring us Mantle and Hbm. They're trying to do what they can to improve things while up against the force of Nvidia and Intel, without being so anti-competative/anti-consumer.Quote:
Originally Posted by shaithis