nvidia NVIDIA Controls AIB Launch and Driver Distribution - HardOCP

[Dylanc1500]

14 minutes ago, S w a t s o n said:

U wot? I watched that shit live it said $999. Are you telling me $999 is for the plastic shroud version, I havent even seen a card at $999 then, could have sworn FE was the only one

I don't know what "U wot?" is, my apologies. Also note: I'm not saying whether or not I agree with the pricing as many arguments can be made in either direction.

 

It said starting at $999. It wasn't said on stage what the price of the Founders Edition would be, but they did show the prices on their website at the time of launch of the 20 series.

 

I know I will most likely get reemed for this sentiment, because the more active people tend to not agree and that is perfectly fine. That being said, Nvidia and board partners can set what they want the prices to be or start at, however, retailers don't always follow suit and price them for whatever people will pay.

[S w a t s o n]

5 minutes ago, Dylanc1500 said:

I don't know what "U wot?" is, my apologies.

 

It said starting at $999. It wasn't said on stage what the price of the Founders Edition would be, but they did show the prices on their website at the time of launch of the 20 series.

 

I know I will most likely get reemed for this sentiment, because the more active people tend to not agree and that is perfectly fine. That being said, Nvidia and board partners can set what they want the prices to be or start at, however, retailers don't always follow suit and price them for whatever people will pay.

Well that's not a sentiment it's a fact but it's generally not in the downward direction, as selling below MSRP/MAP is usually highly discouraged even if product isnt moving. Generally that effect means that those $1200 AIB cards turn into $1500 AIB cards if another shortage hit

[leadeater]

33 minutes ago, S w a t s o n said:

Founders Edition that is, because 2080ti at $999 seems to actually be the floor unlike pascal FE. Every AIB card seems to be 1100-1200 which is a higher premium than usual too

Yea I was referencing xx80 and MSRP with that, I was half expecting having to explain the difference between manufacturer MSRP and street prices at some point. Maybe that has been brought up but but not read over to page 4 yet.

[Dylanc1500]

13 minutes ago, S w a t s o n said:

Well that's not a sentiment it's a fact but it's generally not in the downward direction, as selling below MSRP is usually highly discouraged even if product isnt moving. Generally that effect means that those $1200 AIB cards turn into $1500 AIB cards if another shortage hit

Ya, I know it's a fact. Most tend to get very emotionally driven though when that comes up, which I understand it can be frustrating as a consumer (especially when prices jump three times MSRP within a few weeks), so I tend to stray away from conversations that people get emontional about.

 

I completely agree with you though, it does mean retailers have the possibility of shoving a cactus up the 'ol tooter. The only way for people to force retailers to drop their prices is to hold off on buying till prices start to fall, or only purchase the cheapest cards leaving the more expensive cards to not sell, driving the price down.

 

I'll say this, from what I know in the grapevine they are trying to move product lines around. Take that how you will, I fully expect skepticism and that understandable.

 

@leadeater not to mention board partners can set their own suggested price causing a further increase from NVidia's at retail.

[S w a t s o n]

3 minutes ago, Dylanc1500 said:

I'll say this, from what I know in the grapevine they are trying to move product lines around. Take that how you will, I fully expect skepticism and that understandable.

I believe it, before turing announcement I expected Titan T with full turing die and 3k price like Titan V, I can see them still doing that and keeping Titan as the full die card only with a focus on compute like Titan V

[ZacoAttaco]

Gamers Nexus released a news video, basically NVIDIA still has so many 1000 series cards after the mining bubble burst that the plan is to have them sell alongside the new 2000 series RTX cards. He mentioned the different price range will appeal to different audiences and that the GTX name will still be used in the future.

 

I'm glad that for the time being at least the 2000 series isn't the only option. Unless the performance is insane, I wouldn't be interested in the 2000 series, simply too expensive so it's cool to see the 1000 series will still be around.

[Dylanc1500]

7 minutes ago, S w a t s o n said:

I believe it, before turing announcement I expected Titan T with full turing die and 3k price like Titan V, I can see them still doing that and keeping Titan as the full die card only

Don't forget they brought out the 1070ti. There is now more gap in price between the 2070 and the 2080, than there was with the 1080 and 1070.

 

We shall see...

[leadeater]

37 minutes ago, Blademaster91 said:

The cost of GDDR6 combined with a much larger die and throwing tensor cores on, the cost is probably close to $100 if not maybe more. Since there are still tons of 10 series cards left, what does Nvidia do? Launch Turing cards at the same MSRP's? I think that would make the new gpu's appear not worth pre-ordering or much of an upgrade especially when they're heavily marketing ray tracing tech. I complained about the prices right after seeing the RTX news, but with RAM inflation and increased production cost the prices had to go up.

The Tensor cores themselves don't cost extra to make, not fab wise anyway. That comes under the architecture development cost which all architectures have, no architecture cost nothing to make so having Tensor cores or RT cores will themselves not increase the price. If they were a huge development effort, significantly more than previous generation architectures then a product price increase to cover that would be expected, we don't actually know what the cost of development is of Turing, Volta, Pascal etc or where the cost recovery breakdown is in the product stacks etc are.

 

GTX 780 was the last time MSRP was increased, $499 to $649, previous few generations were all $499. 980 was sort of a price decrease at $549 where the 980 Ti was $649 (780 Ti was $699). Basically GPU prices hover at the same/similar levels for a while then step up after a few generations, as expected re: inflation. There can be other factors too like significant die size increase like you mentioned, fab price may drop to help that but really large increases will always cost more.

 

To me it just seems a bit early to do that inflation adjustment but maybe it was the best time to do it when you combine that with GDDR6 and the die size increase, $749 may just not have worked out and it's better to go to $799 now and not do a double price increase.

[S w a t s o n]

1 minute ago, leadeater said:

 

Yea, between the die size and gddr6 price premium I think that increase is actually reasonable, the 2080ti tho... still being a cutdown chip and charging titan x/xp pricing when titan x/xp got the full cuda core count is fucked.

 

interesting note, P100 had less cuda cores than P40 but more transistors, i had to figure this out when double checking titan x and xp cuda core counts. both have full chip but p40/titanxp(GP102) has more cuda cores than p100/titan x. (GP100)

[leadeater]

2 minutes ago, S w a t s o n said:

interesting note, P100 had less cuda cores than P40 but more transistors, i had to figure this out when double checking titan x and xp cuda core counts. both have full chip but p40/titanxp(GP102) has more cuda cores than p100/titan x. (GP100)

Yea that's the extra space from the FP64 units being cut out of GP102, depending on workload some were better off getting high end GP102 products over GP100.

[S w a t s o n]

3 minutes ago, leadeater said:

Yea that's the extra space from the FP64 units being cut out of GP102, depending on workload some were better off getting high end GP102 products over GP100.

Ah I thought those were cut out after the fact but they actually just designed them out and filled up the space, well played nvidia

[Dylanc1500]

11 minutes ago, leadeater said:

Yea that's the extra space from the FP64 units being cut out of GP102, depending on workload some were better off getting high end GP102 products over GP100.

Evidenced by the fact that the ~5.3 teraflops max of FP64 on the GP 100 (mezzanine card having a higher base clock and boost clock ) vs 367 gigaflops max on the GP102. At the same time the GP 102 had ~11.7 teraflops max of FP32 vs ~10.6 teraflops max on the GP100 (mezzanine card, again).

[leadeater]

2 minutes ago, Dylanc1500 said:

Evidenced by the fact that the ~5.3 teraflops max of FP64 on the GP 100 (mezzanine card having a higher base clock and boost clock ) vs 367 gigaflops max on the GP102.

The FP32 performance of the PCIe cards is an interesting one too, 8071‒9340 vs 10007–11758. Not a bad increase, for a cheaper product with more VRAM as well.

[Dylanc1500]

1 minute ago, leadeater said:

The FP32 performance of the PCIe cards is an interesting one too, 8071‒9340 vs 10007–11758. Not a bad increase, for a cheaper product with more VRAM as well.

lol I just finished that. 

 

12 minutes ago, S w a t s o n said:

Ah I thought those were cut out after the fact but they actually just designed them out and filled up the space, well played nvidia

The way it's meant to be played...

[asus killer]

#raydisaster

 

#raypocalise

 

 

[leadeater]

6 minutes ago, Dylanc1500 said:

lol I just finished that. 

Ninja edit 

[Dylanc1500]

2 minutes ago, leadeater said:

Ninja edit 

I would've said clumsy early submit, but I'll take that lol ?

[S w a t s o n]

3 minutes ago, leadeater said:

Ninja edit 

plsopls

I am the one who edits Sky- I mean leadeater

[Quadriplegic]

Nvidia does what Nvidia wants
 

 

[Dylanc1500]

10 minutes ago, asus killer said:

#raydisaster

 

#raypocalise

 

 

#rayramano

 

2 minutes ago, Quadriplegic said:

 

 

Off topic: I like that song. It's fun to play drums to.

[Taf the Ghost]

1 hour ago, S w a t s o n said:

so $999 for this? https://www.amazon.com/GeForce-Turbo-Type-C-Graphics-TURBO-RTX2080TI-11G/dp/B07GK2LWDL

ahahahahaahaha

 

1080ti FE launch price: 699

2080ti FE launch price: 1199

Holy fuck I didn't realize the gap was that large, what the fuck is jensen smoking. Titan only existed from 700 series onward so it really isn't a given that there should be a 1000+ price tier. If they were gonna charge titan pricing they should have at least made it a proper titan card with double precision and full 12gb of ram.

 

In some ways, Nvidia is being more transparent. The Ti was the Titan, but they launched the Titan first at a high price and Ti later. Now, it's just the Ti at the start for Titan prices. Why? Well, because they can.

 

For the most part, though, Nvidia is simply forcing some constant Performance per Cost metric on the market. At MSRPs, you're paying almost exactly within a continuum from the 1060 through the 2080 Ti. We'll even see a Titan T at some point, with the full TU100 die part for probably another 3k.

[Taf the Ghost]

1 hour ago, leadeater said:

The Tensor cores themselves don't cost extra to make, not fab wise anyway. That comes under the architecture development cost which all architectures have, no architecture cost nothing to make so having Tensor cores or RT cores will themselves not increase the price. If they were a huge development effort, significantly more than previous generation architectures then a product price increase to cover that would be expected, we don't actually know what the cost of development is of Turing, Volta, Pascal etc or where the cost recovery breakdown is in the product stacks etc are.

 

GTX 780 was the last time MSRP was increased, $499 to $649, previous few generations were all $499. 980 was sort of a price decrease at $549 where the 980 Ti was $649 (780 Ti was $699). Basically GPU prices hover at the same/similar levels for a while then step up after a few generations, as expected re: inflation. There can be other factors too like significant die size increase like you mentioned, fab price may drop to help that but really large increases will always cost more.

 

To me it just seems a bit early to do that inflation adjustment but maybe it was the best time to do it when you combine that with GDDR6 and the die size increase, $749 may just not have worked out and it's better to go to $799 now and not do a double price increase.

I think it's important to remember that since they moved beyond the 499USD price category, the lower tier cards are still the massive sellers. The xx70 outsells the xx80 & Ti class about 2:1, if the Steam Survey is representative of the Nvidia market. And the 1060 & below sell about 5:1 or higher compared to the high end of the market.

 

Which is to say, Nvidia has finally started acting like they're in the fully dominant position in the market. They're the only true high-end game in town, it's a smaller market that will simply have to pay for the performance. Nvidia is acting in the way you would expect a monopoly to act, which is really what is driving the ire. After a full year of insane prices in the GPU market due to Mining, Nvidia is pretty much expecting the consumers to just pay for the privilege of having better performance.

 

Know what? They will. Once Navi finally launches, they'll drop prices a little, as Navi should undercut where the majority of the market is, but that's probably not until Q3 2019.

[Taf the Ghost]

Just spitballing, but I expect uplift of around 15% per CUDA core. Then, with more CUDA cores, the uplift will be between 15-30% depending on Game & Resolution. Then Nvidia is going to want 300USD for the 2060 6 Gb in November.

 

There's going to be a lot of "you should look for a used 580 or 1060" suggestions in our future here.

[S w a t s o n]

1 minute ago, VegetableStu said:

I'd even say used 1070 and 1080, but I'm guessing there might be people who might hold on to it for this gen ._. (I mean I would)

Considering the AIBs were still accepting massive shipments of 10 series cards and will be selling them alongside 20 series used 10 series cards are going to end up DIRT CHEAP by the time Ampere rolls around

[Taf the Ghost]

1 minute ago, VegetableStu said:

I'd even say used 1070 and 1080, but I'm guessing there might be people who might hold on to it for this gen ._. (I mean I would)

Like the 970, I don't expect the 1070 to hit the used market in great numbers for a long while, relative to the number sold. It's untouchable for 1080p/60Hz gaming, and that's like 80% of the market still. The upcoming Navi should push the Mainstream AMD GPUs into roughly the same performance category, which is pretty much as far as things need to go for the vast majority of the market.

 

Unless someone is going up to 4K gaming, there's just no reason for them to upgrade. We're going to be in a weird situation where CPU upgrades will probably be more important than GPU for a couple of years.