what is double precision useful for?

[GRAFiZ]

I know this is straying it a little off topic, but it's still quasi-relevant due to the fact DP plays into pricing and interest.

I do believe the problem with the Titan was the fact its pricing was to try to lure more gamers into a workstation/compute-grade card, but still too high for most gamers to afford. And what it offered as its strongest points 99.9% gamers didn't need or could have cared less about (double precision, 6GB of VRAM).

Id say the sweet spot for most high-end gamers is $450 to $600 dollars (most being in the singly GPU board), with hardcore enthusiasts willing to dump $700 to $900 per card (most being in the dual GPU boards). At it's release, the Titan cost a whopping $1300 per card.... so it was way outside the ballpark of interest of even the elitist of gamers, and a single gpu board.

I remember hearing the arguments of 6GB of VRAM being the smoking gun for its sale for guaranteed future proofing, however in reality, for gamers on mainstream resolutions (and even 4k), 6GB is just ridiculous, and likely wont be needed as the standard for another few years at least. Which in that case, it was unnecessary futureproofing. By the time it was needed, there'd be 3 or 4 generations of newer cards that far outpower the Titan in every way/shape/form (and there already is... the 780ti, not even a year later). You had 3 years into the future worth of VRAM, but 3 years behind on the performance, it would have been humorously out of touch with what is required for ultra-level gaming as far as performance goes 3 years from now despite having sufficient VRAM. I mean the next few years will be 4k, which all of todays 2GB to 4GB cards will handle with ease (well, more so the 3GB and 4GB vram cards... I think the 2GB cards will start having issues with 4k especially when AA comes into play)

And never once did I hear gamers bring up Double Precision. I doubt many even know what it is. It literally plays no role in gaming as of current, or for a while (if ever). So there's a feature the target audience doesn't even use, nor care about.

If Nvidia marketing is listening (and I know they are... they browse these forums from time to time), I'd suggest going forward with the Titan line to drop the gamer strategy for it, and focus it more on budget Quadro hardware, less for compute (because of the non-ECC RAM), more for design/modelling/vid editing and especially data mining with that DP. Heck, with Quadro's going for $3000+, it could easily be sold for $2000 and still be considered a budget Quadro if marketed to the right group, and still making more sales at $2000 to the target audience vs. less sales at $1200 with the gaming audience.

The problem is they're two entirely different cards. The Titan really isn't ideal for design/modelling over a 780 or 780ti, because it doesn't provide the stability and certification of a Quadro, or the OpenCL compatability of the quadro series, nor does it provide as many Cuda cores for rendering as say a cheaper 780ti.

So, it's sort of floating in a no mans land.

Nvidia makes a Quadro Titan, it's called the K6000.

Really the Titan is only ideal for someone who wants more Vram than the 780 series provides, or someone who needs DP but isn't willing to buy a Tesla add on card.

Otherwise it serves no real purpose anymore... which is part of why I believe you never see them in stock anywhere.

(I should mention, I own a Titan, it's in my gaming rig, and I love it!) ...but I bought it used for less than a 780ti. Otherwise I'd have purcased a 780ti for the slightly better gaming performance.

 

[Cerb]

The problem is they're two entirely different cards. The Titan really isn't ideal for design/modelling over a 780 or 780ti, because it doesn't provide the stability and certification of a Quadro, or the OpenCL compatability of the quadro series, nor does it provide as many Cuda cores for rendering as say a cheaper 780ti.

So, it's sort of floating in a no mans land.

The Titan packed a lot for the money. However, for most who would have bought a Titan for pro 3D work on a budget (it is and was fine for working on models and such, where the VRAM and GPU power were slaved to the CPU), the 780 Ti will suffice, and is already fairly popular.

 

[Beavermatic]

(I should mention, I own a Titan, it's in my gaming rig, and I love it!) ...but I bought it used for less than a 780ti. Otherwise I'd have purcased a 780ti for the slightly better gaming performance.

Actually just sold both my titans and bought two 780ti's in exchange, lol. Kind of a lateral move trading off VRAM for performance, but eh, it was something to do and killed boredom.

 

[Imouto]

The problem is they're two entirely different cards. The Titan really isn't ideal for design/modelling over a 780 or 780ti, because it doesn't provide the stability and certification of a Quadro, or the OpenCL compatability of the quadro series, nor does it provide as many Cuda cores for rendering as say a cheaper 780ti.

You don't need a powerful card for design/model. You need them for rendering, you may actually want a little render farm. For design/model a good CPU is far more important than the GPU.

Really the Titan is only ideal for someone who wants more Vram than the 780 series provides, or someone who needs DP but isn't willing to buy a Tesla add on card.

And this is why you will never see a 780 or 780Ti with 6GB as the VRAM size is the only selling point of Titan and if rumors hold any water of the next Titan Black Edition.

It's a shame that AMD doesn't worth a penny in this field because the 7970s Vapo-X/Toxic 6GB packed everything you're looking for in a graphics card for rendering. I hope that getting the new Mac Pro changes this but I have little to no faith at all.

 

[Beavermatic]

I hope that getting the new Mac Pro changes this but I have little to no faith at all.

Heard they were having some cooling issues with the new mac pro ashtray design, and that its not so great for expansion, just doesn't have sufficient space to do so. I'd say that'd pretty much kill the deal for anyone looking to stack up on multiple cards in there for such demanding tasks where one card just wont do it.

 

[Rakehellion]

No, it's not. They even admit that much: they make the 780 Ti, Titan, K6000, K20, K20X, and K40 from the GK110. Every spec but DP performance matches that.

Every generation is based on the same design, yes, but gaming cards are just Quadro rejects.

Single-precision has meant 32-bit and double-precision 64-bit ever since IEEE determined that back in 1985. And for all intents and purposes, for quite a while before then,

16-bit formats have always been called something like half-precision, and I don't think anyone was really using it way back in the GeForce 2 era (back then the relevant parts of the pipeline were still fixed point)

16-bit was pretty standard in the early days of GPU-accelerated gaming.

eg Dreamcast and AFAIK some later PowerVR hardware used a 32-bit float z-buffer. Although only inside the tile, for most cases that was effectively the same thing.

So one-off niche hardware uses floating point Z-buffers, but not typically the PC.

 

[Rakehellion]

And never once did I hear gamers bring up Double Precision. I doubt many even know what it is. It literally plays no role in gaming as of current, or for a while (if ever). So there's a feature the target audience doesn't even use, nor care about.

It isn't something gamers ever need to worry about. Number precision is strictly an issue for the programming and design team.

 

[lamedude]

I remembers some arguments over 16/24bit precision during the GeForceFX days.

 

[Imouto]

Heard they were having some cooling issues with the new mac pro ashtray design, and that its not so great for expansion, just doesn't have sufficient space to do so. I'd say that'd pretty much kill the deal for anyone looking to stack up on multiple cards in there for such demanding tasks where one card just wont do it.

Care to provide a link with these issues?

Anyway I couldn't care less about Mac Pro having overheating issues, I just want its users pushing AMD to release better drivers and software. Everyone wins with this, everyone.

 

[Beavermatic]

http://arstechnica.com/apple/2013/06/a-critical-look-at-the-new-mac-pro/

the forementioned thermal core heating concerns and the lack of expendability and impossibility of upgrading or switching GPU's is mentioned here.

More so, if you start reading up on the thermal core of people giving them the run down, it appears to have an inability to balance heat despite what apple originally said, causing one side to become suspiciously warmer than the other.

Apparently they had to dumb down the CPU and GPU selection because of heating issues, mentioned in the articles below:

http://www.primatelabs.com/blog/2013/06/thoughts-on-the-new-mac-pro-benchmarks/

http://www.macrumors.com/2013/06/19/apples-new-mac-pro-begins-showing-up-in-benchmarks/

I give Apple an "A" for effort in trying something new as far as cooling solution goes. Unfortunately, when it comes to your most highend workstation and server class offering, you DONT do implement this sort of thing on your $4000+ line of data crunchers. You use what you know is a adequate cooling solution that is overbuilt to handle heavy load. And this new thermal core apparently missed the point. Looks cool? Maybe. But they apparently sacrificed reliable proven cooling solution for style. That might have a been a cool thing to test on your home desktop lineup, but putting it straight into the Mac Pro was a really, really bad decision. And its starting to show.

Just looking at the design as a systems engineer standpoint, I can't see how it wouldn't have heating problems. The extended heatsink just isn't going to pull heat away fast enough from the high performance CPU/GPU boards in there, that core fan can suck air all it wants, its not helping what appears to be heat entrapment from the cramped walls of the casing, board, and heatsinks. That's why we see desktop gpus with their own independent airflows less than an inch from the cards themselves, and intake/exhaust revolving directly over the card itself either with or without heatsinks that are far closer to the surface of the card.

Like I said, it'd had been something nice to try on a lesser system, but not something that produces that much heat, nor a system that requires heating to one of the top priorities for its dependability. The ashtray inspired wind tunnel design would be nice if this was a Dyson Vaccum... not so much a high end workstation.

Also, the fact there is no NVidia option, nor anyway to swap gpu's because of its proprietary board design, is probably the greatest disadvantage of this machine. No CUDA potential... what where they thinking? AMD's gpu's might be fine for media creation and graphics design, but what about compute.. which is one huge aspect to the Mac Pro line audience.

 

[Imouto]

snip

You serious? It depends on the CPU choice.

The guys in both links have no idea about what they're talking about. They're comparing a 12 core chip with a 6 one.

Every single CPU option for the new Mac Pro is running at full specs and the GPU too. Also it looks like they don't throttle at all if we go by Anandtech's review.

 

[Beavermatic]

These things just started shipping, what last month? So I guess time will tell on the heating as more pople get their hands on them.

I get the design, suck all air to the center, cold air intake from bottom, out the top, heat rises, yada yada. But the fact the interior motherboard area has no apparent ventilation, and just the heatsink tube does is a concern. Like I said, that's a long drag for a heatsink alone without direct cooling to each device other than the heatsink itself. Sure it works, but how well compared to independent intake and exhaust, guess we'll find out.

But my concerns are further validated here:

http://www.macworld.com/article/208...ew-apples-new-mac-pro-really-is-for-pros.html

Not only are they dinging it for its lack of expandability and upgradability as a workstation, they also state the outside of the can was fairly warm to the touch. You can see the interior tube "traps" the components between it and the exterior wall. The tube has no ventilation for the interior space where the boards are, rather just cooling for the heatsink interior wall. What heat is trapped in there from the boards that the heatsink isn't able to transfer is stuck between those walls, significantly warming the case to a noticeable point where even the author mentions it.

The reason why the boards are trapped between the outside can wall and interior tube wall with no airflow in that space is obvious... they don't want debri and liquids spilled around it to be sucked up or fall down and get onto the boards.. rather it'll just hit the tube wall. While that provides a certain level of safety to internal hardware, it has its drawback of trapping heat buildup on components that are already running hot from passive cooling only to begin with and no direct airflow.

Also, while the cooling fan is noted to be unusually quiet, I can't think it's anything too high performance in order to keep decibals down. Just a thought.

 

[jimhsu]

There are (very) few instances where double precision in general (not talking about graphics here) makes sense in gaming -- generally, they tend to be similar to the scientific applications (in other words, "simulation"). I believe EVE Online uses double precision internally, which is necessary to reconcile astronomic distances with ship-to-ship distances. My favorite upcoming space sim Limit Theory also did a switch from single to double precision, which reduced noticeable floating point errors. Note however that neither of these applications offload double precision calculations to the GPU, although such a potential certainly exists.

Also, Minecraft (another simulation game, no surprise there) represents player position as a double precision number, but some other stuff is in single-precision. Hence, strange things like the Far lands crop up.

 

[JDG1980]

No CUDA potential... what where they thinking?

Apple is killing CUDA like they killed Flash. Good riddance.
Companies like Adobe will finally have incentive to do OpenCL right, now that they will be facing pressure from the designers and graphic artists who favor Macs.

 

[Anarchist420]

the 780 and 780 Ti use different chips than the titan and titan black do (GK110-300 for the 780, GK110-400 for the titan, GK110-430 for the black, and GK110-425 for the Ti). i think it's kind of ridiculous that they went through the trouble to make new chips just to reduce double precision performance.

sorry to resurrect this old thread

Not 100% of the time, eg Dreamcast and AFAIK some later PowerVR hardware used a 32-bit float z-buffer. Although only inside the tile, for most cases that was effectively the same thing.

the dreamcast used a Power VR CLX2 and did 32 bit floating point w-testing.

 

[Deders]

There's no switch inside the GTX 780 chip that makes it slower at double precision math. That's like saying The GTX 580 is slower than the 780 because of a switch.

Are you sure about this?

I don't believe the Titan and 780ti have completely different chips... I'm pretty sure it's the same foundation with different features enabled/disabled.

Financially it wouldn't make sense to design two entirely different GPU's that mirror each other in nearly every way possible.

In fact, no way Nvidia did this, it would cost them hundreds of millions of dollars.

Rather, they design one GPU with many features and have several variations fabbed for them.

The Titan and 780Ti use the same chip. With the 780ti, they just enabled the previously disabled SMX which now gives it another 192 cores and (I think) 19 more shaders, cut the VRAM in half, upped the clock speed thanks to a slightly new cooler (though resembles the Titan cooler), added DX11.2 support, and reduced DP to 1/24 from 1/3.

For all intents and purposes, they could have just kept everything intact, enabled the extra SMX, added DX11.2 support, and just called it a Titan 2. Though I believe they saw the Titan was too expensive and outside the reach of most gamers wallets, so while cranking up the horsepower, they had to neuter it in some areas that would be less of interest to a gamer to get the price down, and to still give relevant value of the current Titan lineup to those still wanting to drop $1200 per card.

Which is why I said the Titan should have never been labelled as a gaming card period, and just a value Quadro-series card. Though it'd had been a tough sell for those wanting to do compute on non-ECC ram of the Titan (and if so, risky despite its DP performance), it'd still be great for media/video/modeling/graphics design where ECC doesn't play such a significant role as quantity of VRAM compared to the Quadro line pricing.

This is the Titan GK110 green boxes are single precision shaders and the orange ones are double. The 780 uses exactly the same chip with 2 disabled clusters (compared to Titan) and the orange DP shaders disabled. Whether this is done in hardware or not, no one has enabled them yet on a 780.

 

[Anarchist420]

This is the Titan GK110 green boxes are single precision shaders and the orange ones are double. The 780 uses exactly the same chip with 2 disabled clusters (compared to Titan) and the orange DP shaders disabled. Whether this is done in hardware or not, no one has enabled them yet on a 780.

i dont know how they could damage exactly 672 DP units without damaging anything else given the design. but then i heard that the transistor density isnt very high, so perhaps not a whole lot of laser precision would be required.

 

[Cerb]

i dont know how they could damage exactly 672 DP units without damaging anything else given the design. but then i heard that the transistor density isnt very high, so perhaps not a whole lot of laser precision would be required.

They don't damage them. Somewhere on the chip, or in the firmware (drivers have gotten hacked enough in the past that there's no way it's done in there), there's a setting for whether or not they can be used, or how often they can be used. The sections of the chip controlling what executes then won't allow it, or run it slower, or however it gets done on some given part (I don't know if they disable whole SMs-worth, or if they just duty-cycle throttle them, to go from 1/3 SP to 1/24 SP). Any damage done to make it happen will be a fuse/bridge designed to be cut from the starr.

 

[Headfoot]

Typical ECC implementations can correct 100% of single bit errors, and detect 100% of 2-bit errors. More extensive errors may not be detected or corrected reliably.

However, the exact capability varies with hardware, bus width, etc.

The use of DP and ECC are not related, and there are plenty of reasons why you might need one and not the other. Although in general, DP means serious work of some kind, which means ECC.

DP gives you more significant figures, and therefore reduces the effect of rounding errors. (Rounding errors can crop up where not expected. For example the number 0.1, when converted to binary is actually infinitely long - it is a repeating number, like 1/3=0.3.... So, 0.1 will always be rounded off. Add enough 0.1s together, and the rounding errors will combine and you will get a noticeable discrepancy)

Certain types of computation, where the result of one computation is used in the next, over and over can accumulate errors. This is commonly found in scientific simulations, where the next step in the simulation depends on the previous step solution.

There also certain types of computation which by their nature require high precision at an intermediate step - sometimes it is possible to use a different method to work around this requirement, but sometimes you just have to use higher precision calcualtions. This is a problem that comes up in mathematical techniques like FFTs and matrix inversions.

For gaming, the need for precision and computational stability is low, and therefore DP isn't required. This is the case with most graphics. Although, I wrote a medical imaging app which had a volume rendering function - true volume rendering requires a render pass for each plane of voxels at each distance away from you. This app would use 200-300 pass rendering, with alpha blending on each pass. This worked fine with single precision floats, but with half precision (16 bit), you could start to see faint artefacts appearing where the rounding errors had built up.

In general, scientific and other industrial/engineering work uses DP to ensure that rounding errors don't pile-up and create a noticeable error. Financial work shouldn't use floats at all (not even DP) because a fixed point system which can avoid rounding errors completely is better.

ECC is used anywhere where you have to be sure that your data is correct. Any kind of file server, business data, scientific/industrial work, etc. An alternative, however, is to run the computation multiple times and see if the results are the same (this approach is used by distributed scientific projects such as folding @ home).

This is a great answer.

The only places DP actually makes any economic or hardware-trade-off sense in gaming at all (currently) is in FFT and other reverb/convolution for sound effects for 2 reasons. 1) As Mark has well explained above, the result of one calculation is the input for the next calculation and so on. 2) Humans are much less perceptive of small errors in visual data (slightly off color) compared to slight errors in sound (hissing, clicking, distortion)

 

[Deders]

i dont know how they could damage exactly 672 DP units without damaging anything else given the design. but then i heard that the transistor density isnt very high, so perhaps not a whole lot of laser precision would be required.

If lasers can be used to make the chip in the first place....

 

[BFG10K]

agree with that considering how long we were stuck with only partial precision z-buffers. and due to that, we never really saw any DX9 games that had anything like the original unreal's hugely spaced out levels.

Huh? You do know the version that shipped in Unreal 1 primarily targeted Voodoo 1/2 cards under Glide right? Cards with 16 bit z-buffers.

32 bit fixed point log depth buffers and D32FS8, the latter for when screen space linearity is was a necessity should've been allowed in DX 9 games, but microsoft wouldnt allow it in their games.

Except Unreal was later patched to have a Direct3D renderer and many, many DX9 games have very large draw distances (e.g. Serious Sam HD, Far Cry 1, Painkiller, Fallout 3, Call of Juarez 2, etc).

"Microsoft wouldn't allow it in their games". Seriously, stop making up rubbish.

 

[Exophase]

32-bit integer depth buffers is just his latest arbitrary obsession, something basically no one else cares about.

 

[BFG10K]

32-bit integer depth buffers is just his latest arbitrary obsession, something basically no one else cares about.

The funny thing is, z buffer precision and DP support aren't even related. :awe:

 

[Exophase]

The funny thing is, z buffer precision and DP support aren't even related. :awe:

Nope, he just wanted to revive this six month old post to tell me Dreamcast had a W buffer because I said Z buffer

 

[Anarchist420]

Huh? You do know the version that shipped in Unreal 1 primarily targeted Voodoo 1/2 cards under Glide right? Cards with 16 bit z-buffers.

except only the voodoo1/2 cards had 16 bit w-buffer support as well and that is what unreal 1's glide renderer used. they also used power VR's proprietary API iirc.

Except Unreal was later patched to have a Direct3D renderer

and with depth hacks if 24 bit z-buffer formats are used. the g400 had 32 bit int z-buffer support for a reason. 32 bit log z-buffers emulate w-buffering better than 32 bit float z-buffers do.

many, many DX9 games have very large draw distances (e.g. Serious Sam HD, Far Cry 1, Painkiller, Fallout 3, Call of Juarez 2, etc).

and matrix path of neo and serious sam 2.

32-bit integer depth buffers is just his latest arbitrary obsession, something basically no one else cares about.

games no longer use 16 bit formats for a reason, so what makes you think 24 bit formats will be used forever? just wondering

The funny thing is, z buffer precision and DP support aren't even related.

except if someone wants to replicate a 32 bit int z-buffer in software/compute, then 32 bit float precision wont be enough.