GeForce RTX 3090 gets 24GB GDDR6X and GeForce RTX 3080 to have 10GB GDDR6X

Yeah but what's the die size of a 2080Ti on 7nm? And how much more does it cost to manufacture it on 7nm?

 

Really its how much does the 3080 cost vs Big Navi but I digress. Nobody knows cost per mm² except a couple people on the planet as this is the most guarded information. I have no hard answers but we certainly aren't looking at high costs for a 505mm² chip vs what high end GPU's sale for today. Tons of room for margins so I wouldn't even worry about that side of it because 505mm² is still not a large chip approaching the max reticle size.

That all is to say I would only worry about Navi performance as chip cost should be perfectly within margins. We should expect more than 2x performance since its double the CU's with whatever tweaks made it into RDNA2. How much more than 2x I am not going to speculate.

 

We do have some public numbers on gate costs and the price is basically stalled as of 16nm. So essentially a x number transistor chip on 7nm costs identical (probably slightly more) as the same chip on 16nm. The node allows you double said chip (without increasing reticle size), but the manufacturing cost also essentially doubles because the cost per transistor didn't move and you're doubling the number of transistors.

Now I agree that there are large margins to play with, so in reality the MSRP of the card won't double but it's still going to increase rather significantly. From the leaks, it sounds like the reason why Nvidia possibly went with Samsung is because they were able to negotiate a really good deal on 10nm as opposed to TSMC's 7nm. Which might have been done in an effort to maintain margins, I imagine Nvidia knows the market will only bear so much of a price increase. Like in order for Nvidia to improve performance from the 2080Ti, it needs a fuckton more transistors - there is no way around this.. but it no longer has price scaling from new nodes. So the best they can do is shop for a really cheap node to keep manufacturing costs down, but the new node isn't as good as 7nm so they design a new cooler, raise the TDP by 50-100w, release a youtube video about how revolutionary their new design is and kind of make it all work through marketing.

I think AMD is probably in a better position but I doubt Nvidia cares much - it's clout/marketing/mindshare/etc essentially buys them time to maneuver. They are almost certainly going MCM with Hopper.

As for the end result I couldn't careless. I'm buying a GPU at $1000, if the closest comes in at $800 then I'm saving $200 but I want something that can do 4K in Cyberpunk and close to max settings @ 60fps. Will probably end up with a 3080.. if AMD launches relatively soon I might consider them but idk, the black screen issues plagued a couple people who I convinced to buy AMD and now it's kind of turned me off AMD's GPUs again.

 

No argument I just don't think cost on the higher end is even a consideration not since last year Turning hikes. So much margin built in these days. I have a feeling AMD negotiated a really good deal between Xbox, Playstaion, CPU's, GPU's, and other APU's because they had some high volume buying power.

Anyhow I'm only worried about how Navi performs that will be 99% determining factor of AMD's success as there is tons of wiggle room to adjust pricing for the different high end tiers.

 

So we expect rdna2 and ampere have the same performance impact switching RT on like previous generation of nvidia, because developers know those games are minority. OK then, so why the jump to this generation's boat if we cannot play ray traced games with more than 60fps 4k?

You are very pessimists about RT performance of new gpus, I would understand this ten years ago, now RT is becoming a real thing.

 

Unfortunately with developers catering their engines to take full advantage of the next gen console bandwidth the 3090 won't be a halo product with 24gb of vram. It will be a sku needed on PC to combat the ever growing games that are bandwidth sensitive, MS FS2020, Horizon Dawn Zero, etc. Newly released sandbox games are showing this trend and I don't believe it's an outlier. Any, and dare I say all, future console ported, open world, sandbox games are going to be extremely bandwidth sensitive.

The only way PC's can combat this is through increased vram, for one to at least match the ram on console.

 

3090 is like a Titan, 3080 OK but where is the 3080Ti?? There is a gap between a gfx card with 24gb and 10gb

 

3080Ti = 3090 That is if all the rumors are true.

 

No, there is already mention of 3090 with 12gb vram so 3080ti will be a cut down version.

 

The 3080-Ti is the product at 1100$, can't you see it ?

 

They may be holding back the 3080Ti to see what big Navi brings to counter act.

I really think the 3090 is the "Titan".

 

Unless we get a second 3080 with more gb, this gap in memory amount is impossible to justify. The power difference won't be so huge that they can classify 3080 as 4k gpu & 3090 as 8k gpu.

 

I really, really wanted a 3080 but it's now in the air as i wanted more than 10gb of Vram. If they are gonna double up the vram sizes on future 30XX cards then that makes it even harder to pull the trigger. Just 2gb more and i would of been happy. I mean we already have 11gb cards available and that can be used up in some titles @4k.

 

Amd needs to come up with a 12-16gb navi cards at competitive price. It will be just enough. Thst gap in vram between 3090 and 3080 just shows which cards nvidia wants you to buy. If you would to consider one probably smart thing is to wait for the 20gb version if it ever appears.

 

the 3080ti could be the 20gb version.

It's all a mystery... dern this manufacturers!

 

I noticed in a recent Digital Foundry video Alex Battaglia claimed the Series X GPU had about half the machine learning capability of a 2060, he also thought AMD wouldn't have an alternative to DLSS. I'm not really sure how he came to that conclusion, though.

 

Reviews and benchmarks will be exciting for sure, but damn these prices are just insane. I'm starting to feel like a PC peasant now waiting for a $300 3060.

 

If 3060 ends up 300$ thats actually pretty good.

 

10GB is not really enough. NVIDIA will refresh those really quickly if there's no 20GB version available right from the start. Performance/VRAM ratio is not optimal for future. 3090 would be fine 'till it breaks.

 

3090 for 60fps+ 4k ? then how about 3080 and below ? can all those cards also capable do that ?

lets do simple calculation based previous generation, if 2080ti 4k RT can only do ~30fps... by capable running at 4k/60fps its already 100% increase, right ?
each generation if we look avg performance increase (based fps) is like 30%~40%
how much performance increase u expecting ? like expecting 200% increase is somewhat realistic to you ?

so is this generation is all about RT 4k @60fps ?
as far for nvidia, that not what they target/decide imo, i mean game developers can (if they want) create games that need more performance than 3090 capable off