Nvidia Post-Volta (Ampere?) Rumor and Speculation Thread

[snarfbot]

Wow Nvidia really loves gouging the consumer thats wack as hell.

So not only now are they charging 1000+ dollars for the high end product, but they are selling their mid level chip in the 3080.

Hopefully big Navi destroys it and they have to slash prices and stuff, cuz that's ridiculous.

 

[Benetanegia]

Wow Nvidia really loves gouging the consumer thats wack as hell.

So not only now are they charging 1000+ dollars for the high end product, but they are selling their mid level chip in the 3080.

Hopefully big Navi destroys it and they have to slash prices and stuff, cuz that's ridiculous.

What are you talking about? Nvidia has been selling the Gx104 chip in their GTX xx80 SKU for over 5 generations now. If those rumors are true they actually created a bigger GA103 chip now for 3080 with Ampere. The complete opposite of what you claim...

 

[snarfbot]

Yea and it launched at $499 if you're talking about kepler. What matters is they are selling the mid sized chip as the 3080 instead of a cut down full size chip.

If those rumors are true it will have fewer shaders than a 2080ti, do you think they are going to charge less for it? Knowing Nvidia, doubtful. Maybe they'll shock me, I would be thrilled.

 

[DegustatoR]

Since when does price depend on the number of "shaders"?

 

[snarfbot]

that's like saying since when does price depend on performance

 

[DegustatoR]

that's like saying since when does price depend on performance

Since when does performance depend on the number of shaders?

 

[snarfbot]

Since when does performance depend on the number of shaders?

shh

 

[Benetanegia]

Yea and it launched at $499 if you're talking about kepler. What matters is they are selling the mid sized chip as the 3080 instead of a cut down full size chip.

If those rumors are true it will have fewer shaders than a 2080ti, do you think they are going to charge less for it? Knowing Nvidia, doubtful. Maybe they'll shock me, I would be thrilled.

It's not the mid sized chip. It's the second biggest chip.

 

[snarfbot]

It's not the mid sized chip. It's the second biggest chip.

Well that's one way of looking at it.

Another way of looking at is that Nvidia is going to sell a 3080 in July and dependent on competition from AMD could replace that product within 6 months with a far superior version and continue to sell what was once called the 3080 as the 3070 or more likely the 3060ti.

It all comes down on price, and perhaps freqs as DegustatoR was alluding to.

 

[DegustatoR]

I wasn't "alluding to freqs". Pricing is a function of market competition only.

 

[Benetanegia]

Well that's one way of looking at it.

Another way of looking at is that Nvidia is going to sell a 3080 in July and dependent on competition from AMD could replace that product within 6 months with a far superior version and continue to sell what was once called the 3080 as the 3070 or more likely the 3060ti.

No, that's actually the only way of looking at it. There's a (rumored) lineup and the 3080 is based on the second biggest chip, like it has been the case in the entirety of the last decade, only this time is bigger than ever before. The rest is just pure unsubstantiated nonsense.

 

[trinibwoy]

No, that's actually the only way of looking at it. There's a (rumored) lineup and the 3080 is based on the second biggest chip, like it has been the case in the entirety of the last decade, only this time is bigger than ever before. The rest is just pure unsubstantiated nonsense.

GA103 aside, the rumors point to a pretty standard lineup.

GA104 at essentially the same config as TU104 would be really interesting. We can expect a bump in IPC, bigger and faster caches and of course a healthy upgrade in RT. That would probably come in around 15 billion transistors and 350mm^2. Not an unreasonable guess given Nvidia likely won't go too big on 7nm die sizes right out the gate.

 

[Leovinus]

GA103 aside, the rumors point to a pretty standard lineup.

GA104 at essentially the same config as TU104 would be really interesting. We can expect a bump in IPC, bigger and faster caches and of course a healthy upgrade in RT. That would probably come in around 15 billion transistors and 350mm^2. Not an unreasonable guess given Nvidia likely won't go too big on 7nm die sizes right out the gate.

Correct me if I'm wrong, please, but hasn't nVidia teamed up with Samsung for their 7nm efforts? So far as I'm aware their 7nm implementation hasn't met with a lot of success. Though I haven't personally seen any comparisons between the expected power/performance uplifts to be expected of their 7nm as compared to TSMC's 7nm process. I just recall the rumours floating about production difficulties and performance issues.

 

[Kaotik]

Correct me if I'm wrong, please, but hasn't nVidia teamed up with Samsung for their 7nm efforts? So far as I'm aware their 7nm implementation hasn't met with a lot of success. Though I haven't personally seen any comparisons between the expected power/performance uplifts to be expected of their 7nm as compared to TSMC's 7nm process. I just recall the rumours floating about production difficulties and performance issues.

I'll correct you then
NVIDIA will use Samsung, yes, but majority of chips will still come from TSMC. It's probably the same story as with Pascal, lowest end chips from Samsung and rest from TSMC.
The confusion came from the fact that when Samsung & NV collaboration was announced, they didn't talk about TSMC yet. They confirmed a bit later that TSMC will continue to be important partner to NVIDIA, and then quite a bit later that TSMC will in fact still build most of the chips.

 

[Leovinus]

I'll correct you then
NVIDIA will use Samsung, yes, but majority of chips will still come from TSMC. It's probably the same story as with Pascal, lowest end chips from Samsung and rest from TSMC.
The confusion came from the fact that when Samsung & NV collaboration was announced, they didn't talk about TSMC yet. They confirmed a bit later that TSMC will continue to be important partner to NVIDIA, and then quite a bit later that TSMC will in fact still build most of the chips.

Thanks for that! It must be nice being TSMC at the moment.

 

[w0lfram]

There is a chance that Turing for gaming is not 7nm.

Perhaps that is why the GA103 seems so odd..? 10nm?

 

[DavidGraham]

New There is a chance that Turing for gaming is not 7nm.

NVIDIA said next gen is 7nm at TSMC.

 

[DegustatoR]

I don't see anything "odd" in proposed specs of GA103, besides the fact that I'd expect them to make wider GPUs on 7nm than what's suggested by that leak.

 

[Rootax]

I don't see anything "odd" in proposed specs of GA103, besides the fact that I'd expect them to make wider GPUs on 7nm than what's suggested by that leak.

I'm sure they have bigger gpus, but they don't need them for the gaming segment right now, if ever.
Maybe rdna2 will change that

 

[w0lfram]

NVIDIA said next gen is 7nm at TSMC.

I have not heard Nvidia mention their entire "Ampere" portfolio, is on the same process node. Nvidia has said, their next datacenter/a.i./server farm GPU would be on 7nm, they did not say that their gaming GPUs & smaller dies, would be on 7nm. There is mention of Nvidia shopping TSMC's "other than" 7nm nodes, though.

And since Ampere for DataCenter, binned down to gaming GPU would take 2-3 months, from AMpere's August release, would make Ampere for gaming too far away. So, I personally think Jensen has a SUPER + incoming with Ampere-like architecture, just on an advanced TSMC 12nm node. OC'd to hell, because of really good 12nm+ thermals and efficiencies. And priced just right for gaming.