Nvidia GeForce RTX 2080 Ti and RTX 2080

[CAT-THE-FIFTH]

But people moaning about people moaning the price is high is even worse. In the realworld switch on the TV and watch the news. People moan about price increases all the time - nobody in the realworld cares outside enthusiasts on tech forums who seem to just want everyone to keep quiet about companies since they seem terrified they might offend them.

Nobody in the real world cares how much margins Tesco makes as long as they can afford to keep food on the table.

Nobody cares about all the margins of high street retailers when buying off the internet, This has meant so many jobs are being lost,despite shops probably making worse margins as they have to employ more people,pay for shops,pay more taxes,rates,etc and internet retailers like Amazon using legal loopholes to avoid more tax,etc. A shop cannot compare in price due to the higher fixed costs.

Look at people moaning about fuel going up,rail fairs going up,etc. Maybe there are real reasons why these are needed,but guess what?? People don't care.

Most consumers in the real world care less about companies and more about themselves. That is the REAL WORLD.

In fact on this very forum the hypocracy is terrible - people justifying that US tech companies should charge more,whilst moaning that Marmite went up. Last time I checked Marmite was made in the UK and the company which does it has far more people here than say Intel,Nvidia or AMD.

Where is all the justification for people to buy British and pay more for British?? Yeah,but some product made for relative peanuts in China,sure lets think of a poor company and its margins,and have most of that money flow out the country.

Its the same thinking supporters of microtransactions said - they attacked people who criticised them including famous YT channels,saying games cost more to make,shareholders needed to be considered and people should just suffer in silence,just like sycophants.Yet when EA went too far,people moaned plenty,boycotted the game,and EA then turned around to investors saying they didn't need microtransactions to make decent money from the game,which made all the excuse makers just clam up.

Companies should know entirely what people think of them just like people should have no fear of critcising politicians.

Edit!!

Lets look at tech forum logic - the Nvidia GTX970 mislabelling. People should have just shut up about that - what did all the complaining lead to?? Eventually Nvidia having to correctly label their products,and having to pay $30 back to many GTX970 buyers after a court agreed they had mislabelled the product:

https://www.pcgamer.com/uk/heres-how...70-settlement/

[kalniel]

People moaning about moaning is even worse Then there's the moaning about the moaning about the moaning etc..

Why not simply stop commenting about other posters, let's keep the discussion to the tech and tech companies.

[Corky34]

That article makes it sound like the Tensor cores are a dedicated function, which makes some sense. AI isn't that sensitive to calculation precision, but data sets can be huge and some calculations are bandwidth limited making fp16 twice as fast for a given bandwidth and potentially faster if you avoid paging training data on and off the card because your vram can store twice as much compared with fp32. So tensor operations really want fp16; but frankly nothing else does. Adding fp16 support to the fp32 and fp64 cores makes them more complex and so potentially slows them down when nothing else uses fp16. I can see reasoning for this being a dedicated unit.

AFAIK all the 'cores' within Turing/Volta are dedicated function units, that is they're all mixed precision arithmetic logic units, what I'm less sure about is the width of those units (INT/FP 64,32,16,8,4), i can't remember if it's in the Volta/Turing white papers or the research paper publish by Citadel LLC (or exactly where i read it) but IIRC the data that makes up the tensor matrices (4x4 grid of 4bits of data, 2x2 of 8bits, etc) are gathered from register files and assembled into a single FP/INT 16/32 bit data block that's then processed by a 32bit MP ALU.

If i get the time I'll try to confirm that but having recently tried to explain something similar on another forum recently and it descending into an argument because the person i was discussing it with couldn't understand that there's a big difference between what something does and how it does it, it's probably best not to hold your breath.

EDIT: It was in the research paper publish by Citadel LLC (PDF Link), it seems i got it a little wrong as from what i can tell the matrix is compiled by NVCC at runtime, others (with more knowledge than i) may have a different reading of it..

Before invoking the matrix multiplication, programmers must load data from memory into registers with primitive wmma::load_matrix_sync , explicitly. The NVCC compiler translates that primitive into multiple memory load instructions. At run time, every thread loads 16 elements from matrix A and 16 elements from B.

There's also other references in the paper that talks about how matrices are constructed so it's worth a read IMO.

[CAT-THE-FIFTH]

People moaning about moaning is even worse Then there's the moaning about the moaning about the moaning etc..

Why not simply stop commenting about other posters, let's keep the discussion to the tech and tech companies.

Its Friday...thank goodness the weekend is coming.

[aidanjt]

Gamers whinging about this need to learn that this is the real world.

You mean the target market of GeForce cards?

[CAT-THE-FIFTH]

BTW,as a side note I have started a new thread for raytracing just to talk about the tech itself and any implementations(similar to the one I started over on OcUK too):

https://forums.hexus.net/pc-hardware...ml#post4013891

So any contributions will be greatly valued.

[badass]

You mean the target market of GeForce cards?

Yep. The target market that moans about a companies behavior then buys from them any way.

[badass]

But people moaning about people moaning the price is high is even worse.

True. However moaning about moaning about moaning is worse than moaning about moaning. Moaning about moaning is worse than just moaning.

[badass]

True. However moaning about moaning about moaning is worse than moaning about moaning. Moaning about moaning is worse than just moaning.

And don't get me started about moaning about moaning about moaning about moaning. That's worse than moaning about moaning about moaning which is worse than moaning about moaning. Moaning about moaning is worse than just moaning.

[watercooled]

Not sure I agree with that - simply because I think its bad business to sell capacity on a higher performance, lower power, lower area per transistor process for the same price. However, lets just say perhaps it does

Fair point - I would imagine it's somewhat more expensive than the vanilla 16nm node now, comparing like for like, but I doubt it's substantially more and I would estimate (speculation) it's still cheaper than 16nm was on release. In my mind I was dismissing that it was as expensive as an entirely new cutting-edge node would be, which is somehow how I read your comment!

Agree with you on the rest.

[watercooled]

Yep. The target market that moans about a companies behavior then buys from them any way.

And furthermore directly blame AMD for the situation. Bemoaning the lack of competition is one thing but they tend to be the same people who dismiss anything made by AMD anyway, then blame them when they focus on other market segments instead!

[CAT-THE-FIFTH]

And furthermore directly blame AMD for the situation. Bemoaning the lack of competition is one thing but they tend to be the same people who dismiss anything made by AMD anyway, then blame them when they focus on other market segments instead!

What is even worse is the same gamers making arguments cards need to cost more. Then moan when companies say oh sure BRB...then push up the prices since their potential customers seem to be justifying price increases. Then they bemoan the prices as being high.

Then the same gamers say I won't spend the extra money but end up getting a worse higher margin card even if they don't spend more. That card which might have been profitable at £200 is now £300 since people said they needed to be priced more.

They are literally repeating what Rollo said and don't realise it,who said stuff needed to cost more,stuff is expensive to develope,etc. I saw him on loads of forums before Hexus making those arguments. Now people are just repeating what NV FCG were repeating for years!!

Last time I checked it was marketing which was meant to do that not customers. It's like Turkeys voting for Christmas.

If I was a company I would charge as much as possible,and like I said many years ago,when all the experts argued with me,this would happen and it happened. The same people now are bemoaning the increases. BAHAHAHHAAHA!

I wonder if British Gas could get away with it?

[watercooled]

TBH I suspect the current competitive landscape in the GPU market is what has allowed Nvidia to take this sort of a risk - if AMD currently had a product competitive with the 1080Ti in current games, and for a reasonable price, it would be even harder for Nvidia to convince people this is a worthwhile purchase at the asking price.

Given GPU development times we likely would have seen something similar sooner or later, but I've heard people suggest it wasn't originally intended to be a 16/12nm product? Maybe that has something to do with it?* If so, Nvidia really got lucky as any competition would have likely ate well in to their margins with such large dies, which is probably at least partly why they're shipping so many die variants this generation - past a point, die fusing probably doesn't make as much sense, especially on a fairly mature process.

*Then again, architecturally it seems very similar to Vega apart from the RTX cores, so in theory development time should be somewhat shorter if so.

/speculation

My point being, I can't see how it would have sold well at close to current prices in a more competitive environment. Either they got really lucky or there's far more breathing room in the margins than people seem to be believing.

But again, you can't blame the competition for not competing, when you don't consider them even when they are actually competitive, and/or split hairs over whatever the latest marketing buzzword happens to be.

[CAT-THE-FIFTH]

TBH I suspect the current competitive landscape in the GPU market is what has allowed Nvidia to take this sort of a risk - if AMD currently had a product competitive with the 1080Ti in current games, and for a reasonable price, it would be even harder for Nvidia to convince people this is a worthwhile purchase at the asking price.

Given GPU development times we likely would have seen something similar sooner or later, but I've heard people suggest it wasn't originally intended to be a 16/12nm product? Maybe that has something to do with it? If so, Nvidia really got lucky as any competition would have likely ate well in to their margins with such large dies, which is probably at least partly why they're shipping so many die variants this generation - past a point, die fusing probably doesn't make as much sense, especially on a fairly mature process.

/speculation

But again, you can't blame the competition for not competing, when you don't consider them even when they are actually competitive, and/or split hairs over whatever the latest marketing buzzword happens to be.

The problem is WC,the excuse makers years ago,thought they were being clever when they looked at certain very conveniently "leaked" documents and then on forums were saying things costs more,so its justified companies now added more tiers,etc and upping pricing massively. I told those experts to be wary and they still argued,and I told them you wait and see. Year on year revenue and margins went up. Yeah,they were wrong and I was right since its a trick used elsewhere. Marketing is far more clever than people think and market research has moved on a lot.

I mean not only Nvidia too,but Intel also had people saying on forums competition was BAD.

Also expecting AMD to save anyone is not realistic. If AMD themselves can see Nvidia by using simple marketing,can make their own consumers believe they need to be charged more 5+ years ago due to "costs",then do you honestly think AMD is going to think,wait a second we should massively undercut them?? Why did AMD say they didn't want to be the "budget brand" - because the Turkeys voted for Christmas.

Why use a tactic of massively undercutting Nvidia?? If it didn't work before and now so called gamers are literally parrot repeating the same stuff Rollo was saying 10+ years ago to pay more for stuff,its happy days.

Instead of your own customers wanting better pricing,they just gave you a blank cheque to up prices. So what did they expect....prices to drop?? LMAO.

Polaris wasn't another HD4870 moment - AMD just poked Nvidia slightly not to provoked too much of a price war. Navi will be like Polaris - a gentle poke. Maybe a bit cheaper,but why bother being "too" cheap.

Why should they??

I knew this would happen years ago,and I wished to be wrong yet I was sadly right. People asked for this to happen really.

What to do,even if I keep within my own budget range,I am still enabling them(or either longterm stop playing any newish games).

Edit!!

General comment

Oh also something else - also in general I have seen the view a lack of competition means massively higher prices and its justfied. If market research sees that,then guess what a company is going to do??

Well if that is really the case then think back to the days of the 8800GT. ATI had the craptastic 2900XT which barely competed with the 8800GTS 640MB. Fast foward under a year,and ATI just made another cheaper 2900XT in the form of the HD3870. Nvidia responded with the 8800GT. Nvidia was utterly dominant.



The HD3870 was slower than the 8800GT 512MB,8800GTS 512MB,8800GTX,8800 Ultra,and barely quicker than an 8800GTS 640MB,and traded blows with the 9600GT,and 8800GT 256MB. ATI was probably more screwed competitively than AMD is now. They could barely get over £150 for a single GPU card!!

So 4 faster cards,and 2 cards which traded blows.

Yet the 8800GT 512MB was incredible value for money. If this was 2018,something like the 8800GT/8800GTS 512MB would have been priced much more,because "costs" and "lack of competition".

That should have no bearing on how much YOU as a consumer pays for any product. That is the problem of the company TBH.

[watercooled]

The state of the 'enthusiast' PC market is quite pathetic in more than a few areas now. It seems to be less about interest in hardware, engineering and actually thinking about a build, and more about just throwing as much money at the problem as possible like it's some sort of designer fashion accessory to brag about, and filling the result with hideous RGB lighting, hiding anything interesting-looking in the process.

But like I said, it comes to something when even sites I'd typically associate with being quite Nvidia-apologetic are finding it hard to say anything objectively positive about this release, forums in general seem to have little nice to say, and even share price has taken a dip. It's all a bit silly going 'but but but raytracing... revolution... something' when there's literally nothing to back it up. Sorry for sounding like a broken record but I'm just so surprised the reviews didn't carry anything to sell the big 'selling' features of the card - the things which supposedly make it worth the money. I was really anticipating reading some stuff about it and was ready to pick apart what was actually being benchmarked, where apples were being compared to oranges, etc. It's all just turned out a bit boring really.

I mean sure I would most likely have been ranting about something different then, but at least there would have been something new and interesting to discuss about the technology rather than just pricing and its justification!

[DanceswithUnix]

There's also other references in the paper that talks about how matrices are constructed so it's worth a read IMO.

Ta, I will try and have a proper read but a quick skim through the tensor part it looks interesting.

As for width, I got the impression that pre-Volta the ALU was mixed fp32+fp64+int but now the integer processing is stripped out into its own execution unit so address calculations can be done in parallel with the floating point. ISTR that fp16 was removed from gpus early on, around Riva TNT, and no-one missed it until now. I think in Vega AMD put the data type back in, but a dedicated fp16 matrix multiply-accumulate would seem sensible.

The bit I noticed from that pdf was the huge matrix size required to hit high performance, when AIUI most AI just uses 4x4 multiplies.

Edit: As for people complaining about cost of these cards, don't ever get into performance cars. Yachting is right out