Possible Nvidia 3070 Ti/Super Spotted

[Jet_ski]

On 9/2/2020 at 11:23 AM, GravityHurts said:

My thoughts

I think that super/ti versions of their cards with more vram would make sense for Nvidia to release. I was underwhelmed by the amount of vram especially in the 3080 and I think other people were too.

Wouldn’t the cards do more with less RAM now that they can access storage directly and come with faster VRAM?

[TheDankKoosh]

On 9/2/2020 at 11:23 AM, GravityHurts said:

My thoughts

 

I think that super/ti versions of their cards with more vram would make sense for Nvidia to release. I was underwhelmed by the amount of vram especially in the 3080 and I think other people were too.

With Nvidia's color compression and other vram saving tricks I doubt 10gb will be a limiting factor in the next 4-5 years at 4k in the vast majority of titles, I play at 1440p with high refresh and the 3080 looks to be perfect for me tbh. 4k is honestly too expensive for the good monitors, meanwhile a good 1440p 144hz panel can be had at $300.

[porina]

nvidia holding high VRAM cards in reserve is likely due to ongoing rumours that AMD will go large with their VRAM. If you want to have more VRAM options, seems like you'll have to wait for AMD to make a move first.

[Fatih19]

18 minutes ago, porina said:

nvidia holding high VRAM cards in reserve is likely due to ongoing rumours that AMD will go large with their VRAM. If you want to have more VRAM options, seems like you'll have to wait for AMD to make a move first.

Probably this, there's already precedent. AMD slid their cards to Nvidia's original Turing lineup perfectly, making them better for the same price. They of course did a preemptive strike by teasing "Super" before AMD announced Navi.

 

I think this is a contingency plan by them.

[RejZoR]

AMD used to have high VRAM capacities. Remember Radeon 390X with at that time eye watering 8GB ? Competitor, the GTX 980 only had 4GB at the time and GTX 980Ti 6GB if my memory serves me well. But we know memory isn't everything and while 390X benefited somewhat from it, it was more of a marketing thing imo.

[zachsa999]

On 9/2/2020 at 10:32 AM, EJMB said:

I am expecting a 3060 and a 3080 Ti though to cover all the price ranges several weeks after "Big Navi"

I also doubt a 3070 Ti, wouldn't it hurt the 3080?

A similar card with more memory? Never will hurt a card with more transistors and similar memory

[porina]

1 hour ago, RejZoR said:

AMD used to have high VRAM capacities. Remember Radeon 390X with at that time eye watering 8GB ? Competitor, the GTX 980 only had 4GB at the time and GTX 980Ti 6GB if my memory serves me well. But we know memory isn't everything and while 390X benefited somewhat from it, it was more of a marketing thing imo.

I don't recall the details, or if anyone actually tested it, but I recall claims around that era that nvidia had more/better compression which meant they needed less VRAM. Does anyone know when this might have applied? Does it still apply, or do both sides do similar now? Are there any test results on this?

[RejZoR]

47 minutes ago, porina said:

I don't recall the details, or if anyone actually tested it, but I recall claims around that era that nvidia had more/better compression which meant they needed less VRAM. Does anyone know when this might have applied? Does it still apply, or do both sides do similar now? Are there any test results on this?

No, that was for the memory bus itself as I don't think we're actually compressing the whole VRAM content even today. AMD had 512bit memory bus on R9 290X as they weren't using any compression and NVIDIA was using framebuffer compression which allowed them to have a narrower memory bus. I think it was 256bit at the time on GTX 980. It was a bit weird with AMD at the time. They did release Tonga GPU if you remember it, that did use framebuffer compression and 256bit memory bus, but was released after R9 290X and while R9 390X was technically newer, being a rebadged R9 290X meant it was never treated with framebuffer compression goodness because it actually predated Tonga. R9 Fury then used framebuffer compression, which was odd decision given it already had HBM memory which had stupendous bandwidth and there wasn't even any real need for it.

[Bombastinator]

On 9/2/2020 at 10:36 AM, Mondas42 said:

Just seen this......hmmmm

 

https://www.techpowerup.com/gpu-specs/geforce-rtx-3080-ti.c3581

Not seeing a date on it.  Is there one?

[Vode]

On 9/5/2020 at 4:23 PM, porina said:

I don't recall the details, or if anyone actually tested it, but I recall claims around that era that nvidia had more/better compression which meant they needed less VRAM. Does anyone know when this might have applied? Does it still apply, or do both sides do similar now? Are there any test results on this?

Yeah Maxwell was the gen where color compression really started have a serious effect. They improved upon it with Pascal.

[Mark Kaine]

On 9/5/2020 at 4:15 AM, Jet_ski said:

Wouldn’t the cards do more with less RAM now that they can access storage directly and come with faster VRAM?

No.  just because it works like that in theory doesn't mean devs will actually target such configurations

 

On 9/5/2020 at 2:43 PM, porina said:

nvidia holding high VRAM cards in reserve is likely due to ongoing rumours that AMD will go large with their VRAM. If you want to have more VRAM options, seems like you'll have to wait for AMD to make a move first.

Also this, if someone doesn't totally need a new card right now it's better to wait and see what happens the next 1-2 years as it's likely Nvidia holding better cards with especially more Vram back... 

 

[Bombastinator]

3 hours ago, Mark Kaine said:

No.  just because it works like that in theory doesn't mean devs will actually target such configurations

 

 

Also this, if someone doesn't totally need a new card right now it's better to wait and see what happens the next 1-2 years as it's likely Nvidia holding better cards with especially more Vram back... 

 

So the conventional wisdom still holds: if you don’t absolutely need new gear, don’t buy it.

[porina]

4 hours ago, Mark Kaine said:

Also this, if someone doesn't totally need a new card right now it's better to wait and see what happens the next 1-2 years as it's likely Nvidia holding better cards with especially more Vram back... 

For my comment you replied on, I'm thinking it might be more like 1-2 months before AMD reveal their cards, and it remains to be seen how quickly nvidia might respond to that. If AMD make a big play on VRAM quantity, I think nvidia could respond pretty quickly given the rumours are already in place for some time.

[Mark Kaine]

21 minutes ago, porina said:

For my comment you replied on, I'm thinking it might be more like 1-2 months before AMD reveal their cards, and it remains to be seen how quickly nvidia might respond to that. If AMD make a big play on VRAM quantity, I think nvidia could respond pretty quickly given the rumours are already in place for some time.

Yeah, right I was just guesstimating, the sooner the better and let's hope it actually happens...  

 

 

[Bombastinator]

1 hour ago, porina said:

For my comment you replied on, I'm thinking it might be more like 1-2 months before AMD reveal their cards, and it remains to be seen how quickly nvidia might respond to that. If AMD make a big play on VRAM quantity, I think nvidia could respond pretty quickly given the rumours are already in place for some time.

I haven’t a base clue what AMD will do.  They’ve got a ticklish problem to solve: how to put a card into the market Without Nvidia crushing it.   It’s a problem AMD hasn’t been able to solve so far and I don’t know how, or if, they are going to try. 

 

 They may wait until the ti cards are far enough into production that they can’t be changed, but doing that might mean waiting forever because if Nvidia doesn’t have to put a ti card up crush AMD with they may not do so at all.  This has historically been their only purpose as far as I can tell.  AMD forces Nvidia to produce less abusive products.  It might be that the AMD product doing that this year is the console APU.  There has been speculation that the biggest card Nvidia can come up with is an 80cu gpu that can’t pass the 3080.  Since historically Nvidia has been able to push the gpu die one below a given die into the performance envelope of the die above it (which means I’m this case that a 104 powered 3070 should be able to produce very near what is now considered 102 performance) and the 3090 is a 102 card showing that the 102 die in the 3080 can be pushed to that level (and likely beyond) the only actual market openings I can see are either not only beyond the 3090, but well beyond it, or for AMD to play Nvidia’s game and produce a card positioned to compete in, say, the 3060 space but that can be updateable to beat a 3070, or some such thing.  Basically they’ve got to build a card better than a 3070 that comes removably gimped, and sell it for less than 3060 money.  A pretty steep slope.   
 

No doubt someone will come out with an alternate to that or find some flaw in the data therefore and roundly claim that I am a complete fool in all things for a small disparity again. None the less this is what it seems to me to be the purpose of the ti/super/whatever it happens to be named. A way to reduce prices on a card without actually having to admit they did so. Doing it requires that all Nvidia cards be gimped from the start.  Selling cards with deliberately reduced functionality seems to be how things are done in this space. There seems to me to be suspicion at the moment that the gimping this year is memory.   Not as profitable as simply grossly overcharging as was done the previous cycle, but memory costs money, if perhaps not all that much.  AMD already announced through Microsoft and Sony an apu that has Nvidia reducing prices: Jaguar2.