Time to move along folks. Not a suggestion...

To be honest, I've never read a post from Humansmoke that isn't logically planned and argued. Likewise, Xzibit generally does the same. But the argument about FP is pretty irrelevant.
Two guys with good GPU history knowledge, slightly ignoring literal freedoms and being obtuse. But kudos to both for knowing their stuff.
As for 880, it has to be more powerful than 780 and if it's not faster than 780ti, will need to come in an appropriate price point. What will its FP be? Doesn't matter, gamers don't need it. What matters in the market now is power draw/performance ratio. For 4k we need better power efficiency for the dual gpu's it looks like we need to run them till maybe a couple generations away?
I know folks say power consumption is irrelevant to the gamer but it isn't to the manufacturer. Mobile markets are dictating the trend and whoever gets the best power efficient architecture in their gfx will win. It's the only reason AMD aren't buried on the CPU front, with their APU's beating Intel's on board gfx solutions.

I did not forget AMD and it was irrelevant to this topic , not the matter discussed here, all fabless chip makers have this problem also ; They have to use what available at them . You had been missed one simple point , AMD is not skipping 20nm and nvidia will be at disadvantage all time 16nm did not realised by TSMC than ,it was the point.

arbiterOne thing you forget AMD will stuck at the same proc if they release a new gpu soon as well. So they would be stuck with same issue as well, so its not just problem for nvidia but could be bigger problem for amd.

the54thvoidTo be honest, I've never read a post from Humansmoke that isn't logically planned and argued. Likewise, Xzibit generally does the same. But the argument about FP is pretty irrelevant.
Two guys with good GPU history knowledge, slightly ignoring literal freedoms and being obtuse. But kudos to both for knowing their stuff.
As for 880, it has to be more powerful than 780 and if it's not faster than 780ti, will need to come in an appropriate price point. What will its FP be? Doesn't matter, gamers don't need it. What matters in the market now is power draw/performance ratio. For 4k we need better power efficiency for the dual gpu's it looks like we need to run them till maybe a couple generations away?
I know folks say power consumption is irrelevant to the gamer but it isn't to the manufacturer. Mobile markets are dictating the trend and whoever gets the best power efficient architecture in their gfx will win. It's the only reason AMD aren't buried on the CPU front, with their APU's beating Intel's on board gfx solutions.

FP was a reference to value. Similar to Titan brand marketing and the infamous Z towards more then just gamers to justify its high price. Some think that Nvidia making 50+ on margins wouldn't allow them to sell at a lower MSRP price or that of similar in the past.

If there was no value in FP64 TITAN brand wouldn't exist. I was simply implying that during Tahiti the value was always there since the launch of 7970->7990 until Titan at a more value oriented granted you weren't tied down to CUDA. Which is a benefit for Nvidia which can lure you into spending more $ if your locked into their CUDA Eco-System. I know some green faithful don't even want to look at the other side but it was there if you weren't bias.

The comparisons people are making are just wacky especially when they include GK110 Titan because then your being hypocrite when up-holding its value and not seeing GK104 similar down-side when comparing it to Tahiti. That's my grip.
By all means compare GK110 780/Ti all day long in a gaming context since its better and by the looks of it those people are more interested in swinging there favorite color purse at someone.

I will be looking at Maxwell but in a more cautious way with the talk of V1 28nm and V2 20nm.

GhostRyderMakes more sense than releasing an equal product with less VRAM and performs almost exactly the same on average (Except when you take resolutions into account)...

They only have to be more or less equal, but then I guess AMD felt the need to release a GHZ edition for a reason. But then people don't just take resolution into account but feature sets and the like, and as Tech Report had shown, higher FPS doesn't always equal a better experience... people hated hearing that too.

Both have come a long way regardless.

GhostRyder5-6months...Try almost a year later dude...

GTX 680 Released: March 22, 2012
GTX Titan Released: February 19, 2013

Yea they released Titan as a 1k card almost 11 months later, obviously they had no problem releasing a 1k Desktop grade video card. If they had wanted to get that card out sooner they would have been happier to and charged accordingly, but they had enough trouble even getting the GTX 680 out which was out of stock and basically required camping your computer night and day to get one.

You're right I got my dates wrong, but once again they were perfectly happy selling K20 \ K20X for thousands a pop. Once your high end pro market is happy you can start to trickle that tecnonlogy down to the consumer, even coming almost a year after the 680 it still had the market to itself for another what 8 months?

the54thvoidAs for 880, it has to be more powerful than 780 and if it's not faster than 780ti, will need to come in an appropriate price point.

I think a price realignment is in order anyway. The margin between the 780 and 780 Ti is thin as it is, and certainly doesn't warrant the $150 price differential, especially as cards such as the Gigabyte GHz Edition sits at 780 Ti (stock) performance and can be had for $480 after MIR, and if you shop around as I did for mine, then you can knock another forty off that. IF you split the difference between the 780 and 780 Ti, and say for arguments sake that the 880 sits between them, then it either sells for $549/599 or it takes the more usual $500 price point and the 780 drops officially down to $400-450, and even that may be a little high since the GTX 870 presumable would be around the same performance level if previous salvage part performance is applicable. The only non variable at the moment seems the GTX 770/760 pricing, since for all the talk about GM 204, there hasn't been much (if any) real info on the GM 206 which would EOL the 770/760. If the GM 206 parts are far enough off not to warrant rumours then Nvidia would need stability in the higher volume mainstream. It is pretty much why I thought that the 780 is the odd man out. Newer B1 silicon cards have considerable OC headroom, so how low can you realistically price them?

the54thvoidWhat will its FP be? Doesn't matter, gamers don't need it.

Exactly. Double precision is a workstation feature for the most part and becomes a liability for power consumption and die space - two aspects that are much more important - power consumption because most sales are to OEM's. Low power means being able to cheap out on the PSU, and die space as foundry pricing takes a hike (probably important for both AMD and Nvidia if they plan to basically optically shrink for the next process node.

the54thvoidWhat matters in the market now is power draw/performance ratio. For 4k we need better power efficiency for the dual gpu's it looks like we need to run them till maybe a couple generations away?

I'd guess further out than that. Both AMD and Nvidia need to keep discrete graphics alive and well, and both need to keep distance between themselves and Intel's efforts. Generally software runs ahead of hardware to drive graphics sales, and I don't see that changing. Even if gaming stayed at 4K for a while, it only takes the addition of path tracing (or ray tracing), and the almost certain addition ofvoxel based global illumination(UE4 almost had it except it requires too much graphics horsepower for current architectures) to bring the next generations of cards to their knees.

FluffmeisterThey only have to be more or less equal, but then I guess AMD felt the need to release a GHZ edition for a reason. But then people don't just take resolution into account but feature sets and the like, and as Tech Report had shown, higher FPS doesn't always equal a better experience... people hated hearing that too.

Both have come a long way regardless.

The GHz edition 7970s and the likes are no different then the normal counter parts. All that really changed was the core clocks being bumped to the 1ghz mark essentially to match nvidia doing the clock speeds so high on the 680. Anyone could have upped a normal 7970 to the same levels (not including the heavily binned non-reference models).

FluffmeisterYou're right I got my dates wrong, but once again they were perfectly happy selling K20 \ K20X for thousands a pop. Once your high end pro market is happy you can start to trickle that tecnonlogy down to the consumer, even coming almost a year after the 680 it still had the market to itself for another what 8 months?

That is the definition of the professional market. Professional cards always cost more because if the feature set, software, ram, and the fact they are built and rated for 24/7 use. They are meant to be beaten up and keep working under the most heavy of tasks hence why they charge a fortune for them. The teslas also do not give output (k20/x) and are designed for straight up compute.

XzibitFP was a reference to value. Similar to Titan brand marketing and the infamous Z towards more then just gamers to justify its high price. Some think that Nvidia making 50+ on margins wouldn't allow them to sell at a lower MSRP price or that of similar in the past.

If there was no value in FP64 TITAN brand wouldn't exist. I was simply implying that during Tahiti the value was always there since the launch of 7970->7990 until Titan at a more value oriented granted you weren't tied down to CUDA. Which is a benefit for Nvidia which can lure you into spending more $ if your locked into their CUDA Eco-System. I know some green faithful don't even want to look at the other side but it was there if you weren't bias.

The comparisons people are making are just wacky especially when they include GK110 Titan because then your being hypocrite when up-holding its value and not seeing GK104 similar down-side when comparing it to Tahiti. That's my grip.
By all means compare GK110 780/Ti all day long in a gaming context since its better and by the looks of it those people are more interested in swinging there favorite color purse at someone.

I will be looking at Maxwell but in a more cautious way with the talk of V1 28nm and V2 20nm.

It's because people think that separating the market out completely is the thing to do...I remember people buying 3gb GTX 580's like they were hot cakes for professional work simply because they were so good at it with a decent value. Truth is the only reason nvidia did that was to create more niche markets (titan). The professional cards always have and always will have their reasons for existing and being priced the way they do because they come way more prepared for that work. Desktop GPUs always carry some risk trying to use them for professional work hence why I think Titan branding is foolish because you get basically similar attributes the GTX 580 had normally without a load of extras (including 24/7 rated) at a premium.

You were correct as per usual in your argument.

Well, I wouldn't ordinarily trust Videocardz as any kind of legitimate news outlet, but they seem to have picked up on a vendor (Gigabyte) spokesman's interview at a Chinese event
From Google Translation of the original Expreviewarticle:

Gigabyte declared in an interview with the good news, this year in September they will launch a new G1 game graphics, and is based on the latest NVIDIA GTX 800 series flagship product...

If you follow the link. It refers to 880 GM204. We are more likely to be seeing a repeat of the Kepler cycle (880 is GM204 then is 970) and wont see a full Maxwell until next cycle in 2015.

Worst case scenario = a refined Kepler with a 800 series designation.

Best case scenario = Full Maxwell on 28nm (no gimps)

Atleast we got two months of speculations. What if Maxwell comes with a built in Alien receiver. My bad that was Tesla. Maxwell should come with a camera and a picture of little green men.

If Kepler is anything to go by, Maxwell is going to huge success for the big meanies from Santa Clara.

Don't worry guys, at least you have Tonga to look forward to.

GhostRyderThe GHz edition 7970s and the likes are no different then the normal counter parts. All that really changed was the core clocks being bumped to the 1ghz mark essentially to match nvidia doing the clock speeds so high on the 680. Anyone could have upped a normal 7970 to the same levels (not including the heavily binned non-reference models).

Of course, but AMD felt the need to respond, because the GK104 based 680 was more than enough to compete, I feel like I'm banging my head against a wall here.

GhostRyderThat is the definition of the professional market. Professional cards always cost more because if the feature set, software, ram, and the fact they are built and rated for 24/7 use. They are meant to be beaten up and keep working under the most heavy of tasks hence why they charge a fortune for them. The teslas also do not give output (k20/x) and are designed for straight up compute.

Hmm bit of a random statement there, nothing of which changes what I said.

Can't believe the GK110 is like 2 years old already, what a monster.

FluffmeisterIf Kepler is anything to go by, Maxwell is going to huge success for the big meanies from Santa Clara.

It wont just be desktop, since the latest Nvidia driver branchidentifies nine different Maxwell (N16E-) SKU's, which if they conform to Nvidia's usual nomenclature equate to GTX 940M to GTX 980M models. AFAIK, the top dog GTX 980M arrives October.

HumanSmoke and Xzibit can leave this thread now. Thread has been purged of off-topic nonsense. Move along.

FluffmeisterOf course, but AMD felt the need to respond, because the GK104 based 680 was more than enough to compete, I feel like I'm banging my head against a wall here.

First of all, the bump in clocks was more to conform with the whole "ghz race". Similar to cpu's in the past it was a race and nvidia decided to knock the clocks. The mere 75mhz was just so there was no longer a claim of nvidia being the only one with a reference 1ghz on the core clock. It also proved to be more than enough to best it in a clock to clock ratio (Since both overclock about the same).

FluffmeisterHmm bit of a random statement there, nothing of which changes what I said.

You did not get my point...The Tesla series of cards are not normal and do not conform with the similar laws the desktop and quaddro series follow. In most cases they do not even have the ability to output video to a monitor (Recent C Series have a DVI but not the K20's). They also are deigned in such a way that make it very hard to run in normal or even many professional environments without modification (passive coolers) or special rack mount servers. The K20X for instance has a massive passive heatsink opened up and designed to receive airflow from blower fans inside a machine.

The Primary Focus of these cards are as follows:
Large Scale calculations (Floating Point)
Cuda/OpenCL
Large Scale Image Generation

They are essentially with those series cards almost just saying "Here is a powerful GPU, have fun we will see you later". Your not getting the same type of package as with a Quaddro or Desktop card...

The Teslas are a special breed of cards designed with super computers in mind and do not need to have certain attributes designed for those in mind. They are meant for people to program things to utilize their GPU cores for calulations and have professionals spend time working on them. Nvidia can release GK 110 chips on these even if they are not ready for the mainstream because even with a poor early binning process they do not expect many sales of the cards. It is a very limited market (Even in the Oak Ridge Super Computer has 18,688 GPUs that are K20X but that is still an insignificant amount of GPU's out there) and Nvidia knows that which is why putting out a GK 110 chip early to a very niche market still meant they could work and improve the chip for the main market. I still call upon my last quote "they had enough trouble even getting the GTX 680 out which was out of stock and basically required camping your computer night and day to get one". They were not ready and even the K20X still did not have the full powered core (K40 does) because the process of creating those chips was still difficult like it was for GK 104. If they had been fully ready to release the chip, they would have done so even releasing gimped GK 110 chips (Like the 780) but they were not ready to push it out onto the market (Just like AMD was not ready with Hawaii or else they would have done the same). But they were not ready and had not perfected the binning process yet and a company likes to be prepared with the best product. They do not like having to release products that do not meet standards, they do not want to waste money and maximize profits and getting a bunch of poor quality chips that cannot run at the fullest power is a sure fire way to waste money.

FluffmeisterCan't believe the GK110 is like 2 years old already, what a monster.

This suprises you how? Most GPU's, CPU's, or other chips all have existed out in the development for quite some time (A year or so) but does not mean its ready to be used. There are exceptions of course but most GPU's are made awhile in advance and go through rigorous testing including working on simple things like making the GPU in a cost effective way with having the least amount of failures (Or poor performing chips). Nvidia nor AMD does not just plop out a chip the month they announce it, its not like they experiment and the chips appear with a chemical reaction and they go "BY GOD WE HAVE DONE IT!!! Quick make the annoucement", its a long and tiring process that includes much testing and refining.

GM 204 is the same, its takes along time and they have been working on it for quite some time. They were well aware at a point they could not drop down a node and began working on the GM 204 using the old process and working on it. It has existed for quite some time and we will see all the work they have put into it very soon.

Again hating on GM 204 saying its a poor GPU is going to show to be foolish. GM 204 is going to beat the GK 110 by a decent margin probably at least similar to how the GTX 680 beat the GTX 580. Even though its on the tick cycle where they introduce the new architecture and save the final chip until they have improved/refined the whole process, your going to get a better performing chip. Until we have more information however, most of this is still speculation. The only thing I am sure of (Unless something really wierd happens) is that it will be the top performing single GPU chip from Nvidia once its released at the time!

I know all this, it seems we have our wires crossed here.

You implied if the GK110 was ready they would release it, I and HumanSmoke said it was but they were fulfilling big buck contracts first which made much more sense then rushing it to the consumer market.

www.techpowerup.com/forums/threads/nvidia-to-launch-geforce-gtx-880-in-september.203661/page-3#post-3144739

www.techpowerup.com/forums/threads/nvidia-to-launch-geforce-gtx-880-in-september.203661/page-3#post-3144743

I then said:

Fluffmeister^ Exactly, and once those contracts were fulfilled and yields gradually improved what did we see some 5-6 months later.... *drum roll*.... the $1000 GTX Titan

You corrected me on the release date, but I was referring to the thousands of GK110 installed and up and running at Oak Ridge, so no I wasn't referring to the 680 release at all, and was right in the first place, so yeah shrug.

The comment about the age of GK110 is merely to emphasize how competitive the chip is, even against new AMD silicon, so again my point is simply they didn't need to rush it to the consumer market that's all... yes i really believe that.

FluffmeisterI know all this, it seems we have our wires crossed here.

You implied if the GK110 was ready they would release it, I and HumanSmoke said it was but they were fulfilling big buck contracts first which made much more sense then rushing it to the consumer market.

First of all it was not ready which is why it was NOT released for the normal consumer market or even the average professional market. Professionals and super computers do different things and program/work on the devices constantly. They are able to work more closely with the machine and most programs and such are custom for the machine which is part of the reason Teslas exist as they do. They can release something not yet ready for the consumer market for professionals because they know it wont sell in a high enough quantity and that the professionals do not need as much in terms of extras. The K20 and K20X did not even have the full chip which is even more reason it was not ready (K40 was full version). It was the tick cycle, and they decided to release the GK 104 so they can test things much more in the field and work on the GK 110 silicon. They were not ready and if GK 110 was fully ready they would have been more than happy to wreck AMD with the chip if it was ready like AMD would have done had Hawaii been ready to Nvidia.

Just because a chip exists does not make it ready for the average market or even the average professional market (Hence why no Quaddros or else they would have catered to all the professional markets and not the very strict niche market). If they had released a Quaddro card this would be a different discussion but as with the Desktop cards the GK 110 chip came out much later for reasons of more testing, improving binning, and improved software among other things. Hawaii was the same way and is why we saw Tahiti released first over the Hawaii chips just like we saw GK 104 over GK 110. They are the chips that can be made ready on the new architecture and they are used to test the fields while they put work onto the better chips and improve the processes so they can be ready for the next release.

FluffmeisterYou corrected me on the release date, but I was referring to the thousands of GK110 installed and up and running at Oak Ridge, so no I wasn't referring to the 680 release at all, and was right in the first place, so yeah shrug.

Ok well either way GK 110 being in a super computer working among professionals does not make it a consumer release or provide that it was ready for the big time...

FluffmeisterThe comment about the age of GK110 is merely to emphasize how competitive the chip is, even against new AMD silicon, so again my point is simply they didn't need to rush it to the consumer market that's all... yes i really believe that.

Again you are saying that like this chip is two years old and the AMD silicon is a few months old which is not the case. These chips may have been made at some different dates and exacts are hard to pinpoint except by the executives at both companies. But the fact remains both are probably alot closer in age than you would expect (Or seem to think)...

Just like GM 204, it was probably existed for much longer than we give credit for...

I am also done arguing this at this point...

Again your going off on a bit of a tangent here with your walls of text, I'm talking about working silicon here being shipped and used by paying clients.

Besides I never said it was ready for the consumer market, I said they didn't need to rush it to the consumer market, different things.

I'm glad your done too, for my own sake.

PS. Hi Xzibit!

the54thvoidBut the argument about FP is pretty irrelevant.
What will its FP be? Doesn't matter, gamers don't need it.

I am a gamer and I need my fully-enabled FP rate! Be it 1/3 or 1/4 but not so ugly crippled to 1/8 or even worse to 1/24.

Nvidia (and AMD but AMD is the smaller evil) guys are indeed arses and money-grabbing jerks. And some guys make it sound as if everything they do should be justified.

After all, HumanSmoke's argument would have made some sense (about the die space and power consumption) if they actually didn't cripple in the driver and used different dies but they use the same die for professional and consumer cards!

Sony Xperia SI am a gamer and I need my fully-enabled FP rate! Be it 1/3 or 1/4 but not so ugly crippled to 1/8 or even worse to 1/24.

Nvidia (and AMD but AMD is the smaller evil) guys are indeed arses and money-grabbing jerks. And some guys make it sound as if everything they do should be justified.

After all, HumanSmoke's argument would have made some sense (about the die space and power consumption) if they actually didn't cripple in the driver and used different dies but they use the same die for professional and consumer cards!

The point he was trying to make was clearly that there are almost no games (if any) that benefit from FP64 performance. I would also say it's far more likely that even though they are the same die's the die's the performed better became Tesla's where as the lower performing (slightly defective but still functional) die's became Titans. In the world of semiconductor manufacturing not all parts are created equal.

xenocideThe point he was trying to make was clearly that there are almost no games (if any) that benefit from FP64 performance.

I know that GTX 680 and its professional iteration K10 actually do have one and the same FP performance. It is hardware castrated from the very beginning. My question is - how do nvidia sell this crap as a Tesla K10 and who would buy 1/24 double precision?



As a gamer, I didn't say that I need my fully-enabled double precision for games. There are multiple other applications where I would be glad to use the same card for.

Anandtech have a pretty nice showcase page called "Compute: What You Leave Behind?" in the GTX 680 review.

www.anandtech.com/show/5699/nvidia-geforce-gtx-680-review/17

You will see the PrimeGRID Genefer 1.06, and also AESEncryptDecrypt, SmallLUX GPU, Civilisation V.

I will leave you for your own conclusions.

ScatlerAs if Nvidia ever released top end gpu's at reasonable prices. When you have "top" performance you ask top dollar for it /sarcasm.

The price of GTX 480, 580 and 680 was 500 $, which was not unreasonable. They went bullshit with the whole Titan/780 thing.