GPUs & CPUs & Enthusiast hardware: Questions, Discussion and fanboy slap-fights - Nvidia & AMD & Intel - Separe but Equal. Intel rides in the back of the bus.

If we're going based on TOPS (which isn't accurate but it's the only thing that's been revealed so far), the 5070 is going to sit between the 4080 and 4090 in performance.
Yeah we are not doing that. This is the company that will claim 4x generational performance uplift.

5070 12 GB BTW
 
If we're going based on TOPS (which isn't accurate but it's the only thing that's been revealed so far), the 5070 is going to sit between the 4080 and 4090 in performance. If that's the case, then that's pretty brutal for AMD because that means there's effectively a $549 ceiling for all of RDNA4 and realistically they need to be $100 lower than that to get *any* sales.

Maybe
9070 XT - $450
9070 - $399
9060 - $299

Can AMD even make money at those prices?
1736220908802.png

Nvidia is not being exactly truthful with their comparison you are losing 12 GB of VRAM (but is GDDR6X and not GDDR7), the memory bandwidth is cut out and the AI TOPs if someone still care about it is also cutted out still, maybe with the faster VRAM it could reach the 4090, i need hard data before and yes AMD will need to start selling at a loss because they dont have any way to reach the 5070 at this rate

Edit: or the meta will end with AMD just taking the 4080-4070 market but they will have to again sell it so low that would make intel look reasonable
 
4090 = 16,384 CUDA cores, 1008 GB/s bandwidth, 24 GB
4070 Ti Super = 8,448 cores, 672 GB/s, 16 GB
4070 Ti = 7,680 cores, 504 GB/s, 12 GB
5070 = 6,144 cores, 672 GB/s, 12 GB
4070 = 5,888 cores, 504 GB/s, 12 GB

Maybe 5th gen tensor cores are amazing, but the debate should be about whether the 5070 can even beat the 4070 Ti or 4070 Ti Super in gaming, not the 4090.
 
4090 = 16,384 CUDA cores, 1008 GB/s bandwidth, 24 GB
4070 Ti Super = 8,448 cores, 672 GB/s, 16 GB
4070 Ti = 7,680 cores, 504 GB/s, 12 GB
5070 = (based on leak) 6,144 cores, 672 GB/s, 12 GB
4070 = 5,888 cores, 504 GB/s, 12 GB

Maybe 5th gen tensor cores are amazing, but the debate should be about whether the 5070 can even beat the 4070 Ti or 4070 Ti Super in gaming, not the 4090.
1736222001313.png
Official tech specs
 
I’m just really curious how Nvidia has addressed the input lag issue if they are going to continue the expansion of their frame generation tech.

If they didn’t people aren’t going to like their games feeling sluggish.
 
I’m just really curious how Nvidia has addressed the input lag issue if they are going to continue the expansion of their frame generation tech.

If they didn’t people aren’t going to like their games feeling sluggish.
AI heuristics predicting your input and acting on it ahead of time. I'm only half-joking, I'm pretty sure this has been explored.
 
4090 = 16,384 CUDA cores, 1008 GB/s bandwidth, 24 GB
4070 Ti Super = 8,448 cores, 672 GB/s, 16 GB
4070 Ti = 7,680 cores, 504 GB/s, 12 GB
5070 = 6,144 cores, 672 GB/s, 12 GB
4070 = 5,888 cores, 504 GB/s, 12 GB

Maybe 5th gen tensor cores are amazing, but the debate should be about whether the 5070 can even beat the 4070 Ti or 4070 Ti Super in gaming, not the 4090.
I'd wager Nvidia is including DLSS 4 frame generation into the "beating the 4090" metric. We won't really know until third party benchmarks.
 
So this may be nvidia’s answer to the added latency with more frames.

 
  • Thunk-Provoking
Reactions: Vecr
View attachment 6829797
The little AI box is so cute, remind me of the chinese mini pcs you see in aliexpress
you're joking but i do love the aesthetic, in general its sort of funny how GPUs and their multicolor light shows are now so many in the corporate world. then you have stuff like this that looks like something you'd get at staples, surprised they don't add cigarette burns to the outside.
 
So what's the deal? Why did AMD completely bail on RTG - discrete RDNA for their CES keynote time?

It feels like they're not only giving up the high-end but the midrange as well this generation ..

APUs forever .. APUs as far as the eye can see ..
 
Last edited:
  • Thunk-Provoking
Reactions: Vecr and Betonhaus
So what's the deal? Why did AMD completely bail on RTG - discrete RDNA for their CES keynote time?

It feels like they're not only giving up the high-end but the midrange as well this generation ..

APUs forever .. APUs as far as the eye can see ..
It’s not a coincidence AMD is focusing on the APU space, where Nvidia isnt even trying to compete (right now anyway)
 
So what's the deal? Why did AMD completely bail on RTG - discrete RDNA for their CES keynote time?

It feels like they're not only giving up the high-end but the midrange as well this generation ..
Their new discrete GPUs are literally being shown off at CES, such as this one:

What AMD didn't do is talk about performance, price, etc. because they are cowardly and needed to wait for Nvidia's announcement first. They will have a full product announcement later, probably within a couple weeks.
 
  • Like
  • Agree
Reactions: Vecr and Miller
32 GB VRAM on a single card is pretty good. Interestingly enough if the price is similar to the 4090. If the use case allows for multi card setups, it would finally be more cost effective to just buy two 5090s assuming that they can be fit in a case or open frame. The issue is that most CPUs these days only have 24 pcie lanes. With most setups only having a single 16x slot capable of pushing 16 lanes.
 
those “leaks” putting the 5080 at $1600 and 5090 at $2600 always looked fake and gay to me, can’t wait to see the exact same shit happen next gen too!
 
Back