Nvidia will answer Hawaii, Super Kepler + Maxwell in q1 2014

[Lanek]

Maxwell brings only a small increase in performance against Kepler, until 20nm production ready. Maxwell has great power efficiency, has Huma functions and is more efficient, its advantages are NOT Performance at first. It is tweaked Kepler with few new features.

huum, hUMA is the AMD and HSA name.. for Nvidia the name is a bit different.

[boxleitnerb]

Maxwell brings only a small increase in performance against Kepler, until 20nm production ready. Maxwell has great power efficiency, has Huma functions and is more efficient, its advantages are NOT Performance at first. It is tweaked Kepler with few new features.

Small is 10%, 15%, 20%? At the same TDP, of course.
Or phrased differently: How large is the efficiency gain of Maxwell@28nm vs Kepler@28nm?

[zalbard]

huum, hUMA is the AMD and HSA name.. for Nvidia the name is a bit different.

Different name, same thing in principle.

[Sushi Warrior]

They are reporting record margins. So not only are price cuts possible but big ones at that. I would hope that yields would have improved at this point to for gk110 since it has been about a years since these chips have been shipping to customers.

Price cuts are absolutely possible, but I just frankly don't think Nvidia will. Too many consumers willing to pay inflated prices for "locked" and/or slower hardware, because of brand image/support/features/green is a cool colour. Not sure how much it would take for them to budge.

[raghu78]

They are reporting record margins. So not only are price cuts possible but big ones at that. I would hope that yields would have improved at this point to for gk110 since it has been about a years since these chips have been shipping to customers.

Nvidia's record margins are due to selling Titan at USD 1000 and GTX 780 at USD 650 - 700. the reason is lack of competition. With Hawaii AMD is bringing competition and Nvidia will cut prices to remain competitive. Hawaii at USD 600 and Titan performance looks to be the expected price and performance. that would force Nvidia to release a GTX 785 with 2688 sp or 2880 sp for USD 650 - 700 and cut prices on GTX 780. the fact that the desktop flagship Geforce chip has 20% of cuda cores (576 cc) and 1 complete SMX disabled means yields are nowhere near good for GK110. GTX 780 is the only GK110 chip available in real high volume. Titan was priced to keep demand low.

[AbortRetryFail?]

~snip~

They are reporting record margins. So not only are price cuts possible but big ones at that. I would hope that yields would have improved at this point to for gk110 since it has been about a years since these chips have been shipping to customers.

Book-cooking

Year-over-Year net revenue is down 19% for Q2 -- earnings per share down 16% including a $750 million payment to Goldman Sachs who returned 36.9 million shares on May 16, 2013, in the first round of a stock-repurchase.

When stock is repurchased, the price of outstanding shares is supposed to go up. It didn't.

[tajoh111]

I think they are lasering off way more cores than they have to for gk110. I have noticed lately, asic quality of gtx 780 chips have tended to be better than gtx titan chips for overclockers. I have seen titan chips with asics in the low 60's and gtx 780 in the high 70's. I think because titan were released relatively early and were the left over chips from the supercomputing project, I think it when it was more difficult to get a 2688 part chip.

I am guessing as yields improved and when the gtx 780 came along, they could have given the gtx 780 way more parts, but needed to preserve the gap between titan and the gtx 780. Especially with a 350 dollar price difference and the 7970 not being competition. So to justify the price difference of the titan and gtx 780, a lot of gk110 was lasered off I suspect more than they had too.

I haven't seen a titan or keplar with an asic quality in the 80s which leads me to suspect it doesn't take top notch bin to make a titan or gtx 780.

[GoldenTiger]

Price cuts are absolutely possible, but I just frankly don't think Nvidia will. Too many consumers willing to pay inflated prices for "locked" and/or slower hardware, because of brand image/support/features/green is a cool colour. Not sure how much it would take for them to budge.

You ever wonder if the reason people pay the higher prices isn't because they think "green" is a "cool color", but rather that the performance, features, support, and overall (on the whole, nothing's perfect in tech) driver quality tends to be better?

[KiSUAN]

You ever wonder if the reason people pay the higher prices isn't because they think "green" is a "cool color", but rather that the performance, features, support, and overall (on the whole, nothing's perfect in tech) driver quality tends to be better?

Working in hardware distributor for years, I can tell you that's not true, most people buy hardware as if they were buying ice cream, they eat the same flavor all their life and in the same shop, never dear to taste another flavor and god forbid if they have to go to another shop.

[tajoh111]

Book-cooking

Year-over-Year net revenue is down 19% for Q2 -- earnings per share down 16% including a $750 million payment to Goldman Sachs who returned 36.9 million shares on May 16, 2013, in the first round of a stock-repurchase.

When stock is repurchased, the price of outstanding shares is supposed to go up. It didn't.

The outlook on Tegra is grim kind of. Tegra sales were down something like 70 percent which highly affected net profit. Also GPU desktop market went down to for both companies. This doesn't mean margins decrease just less production is needed and lower cost.

Stock is still up 25 percent since the beginning of the year.

If we exclude the tegra division, the GPU division is up q/q and probably year over year slightly. The tegra chipset division has tanked hard this year and really puts a hurt on their revenue statement since they were not even breaking even last year when sales were much better.

[PedantOne]

huum, hUMA is the AMD and HSA name.. for Nvidia the name is a bit different.

of course, but it is the same thing

[Sushi Warrior]

You ever wonder if the reason people pay the higher prices isn't because they think "green" is a "cool color", but rather that the performance, features, support, and overall (on the whole, nothing's perfect in tech) driver quality tends to be better?

I feel like you completely ignored this part

brand image/support/features/green is a cool colour.

Performance for certain models is higher. Marketing and brand image also plays a giant part, think about Thermi.... lost on just about every front besides absolute performance, still sold quite a bit.

[Mungri]

Price cuts are absolutely possible, but I just frankly don't think Nvidia will. Too many consumers willing to pay inflated prices for "locked" and/or slower hardware, because of brand image/support/features/green is a cool colour. Not sure how much it would take for them to budge.

I doubt there will be any GK110 price cuts for a long time, if they are selling well then they wont get reduced in price.

UK prices on GTX 680s and 670s are recently amazing though, but theres not many left but I already got my pair. The GTX 780 and above dont entice me with their price tags.

People are also so quick to forget how inflated the prices on AMD CPUs and GPUs were when they had the tech lead - 7970 launch price was ridiculous and they always make such a huge deal out of beating Nvidia's previous gen, but then a few months later Nvidia pull out a card thats cooler, less power consumption, lower specified and performs pretty much the same, so AMD are forced to drop their prices to compete but now only Nvidia's prices are to blame.

Also I paid ?350 for an AMD dual core CPU back when they were thrashing the Pentium 4 so much, but then after Intel gets back on top you get top performance quad cores for just ?150, and then AMD are forced into undercutting that price because their CPUs were slower.

AMD / ATI arent simply cheaper because they care about the customer, they are cheaper when they need to be to remain competitive, otherwise they wouldnt be able to sell anything.

Another thing that people always overlook with graphics cards is that Nvidia have always had vastly better AIB partners / warranty supports, while outside of the main hardware companies (Asus, MSI, Gigabyte), the rest of AMD's partners tend to be terrible for build quality and after sale service.

[Sparky]

Oh the "nvidia drivers are superior" stuff again

I've had good results from most drivers from both sides, and some buggy ones from both sides. As of right now I'm a little grumpy with nvidia's drivers every time I close IE on my work computer as IE crashes most of the time instead of closing cleanly. Thanks nvidia.

[Mungri]

Oh the "nvidia drivers are superior" stuff again

TBH if you use two or three cards then it is entirely true.

AMD's crossfire support is attrocious, they havnt put much effort into it for a long time, and I remember that when I had a pair of 4870s, they wouldnt even work in crossfire for a month or two until the manufacturer emailed me an updated bios. My 5770s wouldnt run benchmarks without severe graphical corruption, though each card was fine individually and I tried several crossfire connectors and had them on two motherboards.

Nvidia SLI on the other hand simply works without a single issue from the moment of installation, and microstutter has also been shown to be a lot lower on SLI compared to Crossfire. I havnt had a single issue with several SLI setups, but plenty of problems with setting up and getting crossfire working. At the time I didnt realize how bad it was though until I tried Nvidia setups.

For single card use, I wouldnt imagine that theres anything different between the two.

Also another thing is that I'm not sure if AMD support crossfire AA. I've tried out SLI AA in many games and it works in any game that allows driver overrides for AA up to 64x SLI AA + Transparancy options. This is very handy for games that dont get much gain from multi GPUs, or those which dont need that much performance.

[Sparky]

Granted I never use more than one card. Extra cost, more cause for frustration, all for what? When my single card gets outdated enough to start having issues I find it better to get the latest gen single card.

I've had friends both with crossfire and SLI in the past and both have had misc random issues. They are using single cards again lol.

[Mungri]

I've never had any issues with SLI tbh, GTX 460, 560 and now 680 SLI work flawlessly. Plug them in, install drivers, and everything works without an issue, unlike my experiences with multiple crossfire setups.

AMD definitely have issues with crossfire. That doesnt mean that the same issues exist for SLI.

Extra cost, more cause for frustration, all for what?

Because 2 GTX 680s currently cost a little less than a single 780, at least in the UK. GTX 670s are much less, can get those for ?180 each.

[Sparky]

Difference again I guess. I would have one GTX680 (or AMD equivalent) and not bother with a 780. 680 would be plenty fast. I only upgrade once every few generations anymore. Unless there is some massive change that makes me need an upgrade sooner I'll probably be running this 7950 for the next 3 years or so. At that point it should be worth upgrading again.

[BrowncoatGR]

huum, hUMA is the AMD and HSA name.. for Nvidia the name is a bit different.

How exactly is Nvidia going to implement common memory access without having access to the CPU memory controller?

[Ace123]

I'm not exactly sure what everyone's problems with either drivers are. I have been using crossfire since my dual x850s with that crossover cable in the back. Crossfire worked on every single one of my rigs regardless of what version driver I have used all the way up to my 7990. I have never had a problem with a crossfire configuration that didn't turn out to be a problem with my windows installation or some damn update or some fouled up situation from hard resetting my rig a hundred times.

On the other end of the spectrum every Sli configuration since my dual 7600gs has worked just as flawless and I've had even more Sli configurations that crossfire.

People are so quick to blame nvidia or Sli drivers for there stupid problems. Maybe your computer is crappy, power supply is bad, windows install is fouled, IE explorer is messed, updates are screwing it up.

[zalbard]

How exactly is Nvidia going to implement common memory access without having access to the CPU memory controller?

GPUs already have access to system memory through drivers. What currently happens is, GPUs copy stuff back and forth between system memory and VRAM.
What they will do soon is map system memory as an extension of VRAM. How exactly that is going to work, only people implementing this can tell you.

[mutantmagnet]

I feel like you completely ignored this part



Performance for certain models is higher. Marketing and brand image also plays a giant part, think about Thermi.... lost on just about every front besides absolute performance, still sold quite a bit.

While he accidentally ignored that part you did start off your sentence saying the hardware was slower :|

[Piledriver]

Oh my, like the Super Saiyan kepler about to come out wasn't hilarious enough, now Maxwell is an APU with huma



Don't buy AMD kids, nvidia's woodscrew editions are always around the corner.

[BrowncoatGR]

GPUs already have access to system memory through drivers. What currently happens is, GPUs copy stuff back and forth between system memory and VRAM.
What they will do soon is map system memory as an extension of VRAM. How exactly that is going to work, only people implementing this can tell you.

Well yeah. AMD can do this with APUs because the GPU and CPU are on the same die and share the memory controller. Nvidia otoh have to go through the pci-e bus so if they managed to work around that, that would be quite amazing

[Lanek]

Well yeah. AMD can do this with APUs because the GPU and CPU are on the same die and share the memory controller. Nvidia otoh have to go through the pci-e bus so if they managed to work around that, that would be quite amazing

You point to the major difference between hUMA ( AMD ), and the system of Nvidia, actually in reality, AMD allready have enabled Unified memory acess on GCN ( all HD7xxx ).. basically a desktop GPU ( AMD ) will pass today through the PCI express bus for do it. Anyway, the subject is complex and i have absolutely not the time right now.