AMD Navi 23 ‘NVIDIA Killer’ GPU Rumored to Support Hardware Ray Tracing, Coming Next Year

[Nablaoperator]

The Nvidia's 10K CUDA cores are actually half

I have not been convinced with that line of reasoning..
and yes i forgot to mention the elephant in the room - 512GB/s bandwidth.. how is that being explained away?

 

[4K8KW10]

The key lies here:

Radeon RX 6900 XT

https://www.techpowerup.com/gpu-specs/radeon-rx-6900-xt.c3481#:~:text=AMD has paired 16 GB,MHz (16 Gbps effective).

RTX 3090

https://www.techpowerup.com/gpu-specs/geforce-rtx-3090.c3622

 

[Raelgun]

I have not been convinced with that line of reasoning..
and yes i forgot to mention the elephant in the room - 512GB/s bandwidth.. how is that being explained away?

Looking at the radeon VII, bandwith only truly matters for large assets and high resolutions granted this is where these cards should be used, but faster Vram is more desirable for just games that don't have 4K texture assets or higher (this is not 4K resolution).

HBM memory has huge bandwith, but isn't actually fast.

 

[LoadsaMoney]

LLLLLLLLLLLLLLLLLLLLadies and Gentlemen, my names Paul, and in this redgamingtech.com video yadda yadda yadda......................

 

[shankly1985]

Of true AMD has been pushing TDP hard to get those clock speeds up

where are all the fans who've been shouting about 2.5ghz and 250w... oh so it's actually 2.2ghz and 340w - looks like AMD got worried along the way and put the extra budget into pushing clocks way above peak efficiency so now they conpete with Nvidia. I wonder how AMDs cooler will handle 340w compare to Nvidias that runs at 68c, that will be quite interesting

68c put the box it doesn't I have watched a good amount of benchmark video that show the GPU hitting 80c

 

[FoxEye]

Big Navi's theoretical performance puts it right at the top, it all just depends on how well RDNA2 scales with more CUs and frequency, it's the first architecture that's meant to have no leftover GCN so it could therefore scale a lot better, it also depends on the actual real world perf/watt improvement, whether it's 50%, 60% plus like RedGamingTech has suggested, which I'd be inclined to agree since he's gotten the Infinity Cache correct, or whether it's lower.

Nobody has a fn clue where Big Navi's performance is, theoretical or otherwise.

It's all 100% pulled from ass-land.

 

[4K8KW10]

Nobody has a fn clue where Big Navi's performance is, theoretical or otherwise.

Don't underestimate Nvidia's spying abilities

 

[FoxEye]

Don't underestimate Nvidia's spying abilities

It always makes me laugh when people say, "We've analysed these rumours and..."

You might as well get your divining rods out mate, because that "analysis" is about as worthwhile as asking Mystic Meg. I'm sure she'd draw you a bar chart if you asked nicely.

 

[4K8KW10]

It always makes me laugh when people say, "We've analysed these rumours and..."

You might as well get your divining rods out mate, because that "analysis" is about as worthwhile as asking Mystic Meg. I'm sure she'd draw you a bar chart if you asked nicely.

There is no analysis, sorry, brother.

 

[EastCoastHandle]

Don't underestimate Nvidia's spying abilities

What do those same spies say about regulatories approving the Arm deal as they weigh in on their Ampere release?

 

[Journey]

Nobody has a fn clue where Big Navi's performance is, theoretical or otherwise.

Lets hope AMD have a few members of staff who are aware, or is it just going to be one big surprise to everyone on stage at the end of the month?

 

[Nablaoperator]

The key lies here:

Radeon RX 6900 XT

https://www.techpowerup.com/gpu-specs/radeon-rx-6900-xt.c3481#:~:text=AMD has paired 16 GB,MHz (16 Gbps effective).

RTX 3090

You got to back that with workload distribution..
I am still not convinced..

I believe this is how it works out qualitatively in games:
fp64 not used??
fp16 for normals/directions, post process effects that rely on low precision inputs
int used for lookups and loop controls (35% of total workload - Nvidia)
rest of the stuff in fp32 (looks like bulk of the %)

 

[4K8KW10]

Only AMD staff? I think quite enough people in the supply chain - both at TSMC and the board partners know everything already. At least they. Not to mention people like Charlie at S|A and the other old dogs.

 

[Nablaoperator]

LLLLLLLLLLLLLLLLLLLLadies and Gentlemen, my names Paul, and in this redgamingtech.com video yadda yadda yadda......................

Seemed less upbeat

 

[queamin]

It always makes me laugh when people say, "We've analysed these rumours and..."

You might as well get your divining rods out mate, because that "analysis" is about as worthwhile as asking Mystic Meg. I'm sure she'd draw you a bar chart if you asked nicely.

They are using these to see the future.

 

[4K8KW10]

Nvidia vs AMD: who is trolling who?

According to some AIB partners cited by Videocardz, Nvidia managed to troll AMD, as it switched the GA104 SKU with the GA102 model in the RTX 3080 GPU at the last minute, so now, apparently Big Navi (Navi 21) is only matching the RTX 3070. Other leakers believe that AMD is trolling the AIB partners misrepresenting Navi 21 as Navi 22, which has only half the CUs.

We are three weeks away from the official announcement of the AMD RDNA2 GPUs, which are supposed to be team red’s first real competitive GPU series in years, yet we have not seen too many trustworthy leaks that point to AMD’s intended performance targets. Although we know most of the GPU specs for the Navi 21, Navi 22 and Navi 23 models by now, the information is all over the place when it comes to direct comparisons with Nvidia’s cards. Is the Navi 21 faster than the RTX 3080? Is it slower than the RTX 3070? Is it only just as fast as the RTX 2080 Ti? Nothing too conclusive as of yet, and, to add more fuel to the fire, the guys over at Videocardz recently tweeted the possibility of Nvidia jebaiting AMD as per some info they managed to gather from AIB partners this week.

https://www.notebookcheck.net/Nvidia-vs-AMD-who-is-trolling-who.496922.0.html

 

[Zeeflyboy]

Ah, I see, thanks. I thought both were from VideoCardz...carry on...

-------------------------------------------------------------------



So, if this is on the PS5 then what's being used for PC??? I'm not sure what corteks provided is from the 6000 series but from the PS5.

eh? because it's a rectangle?

What coreteks was showing is quite literally more than twice the size of that chip.

 

[Twinz]

The 3080's price and TDP would seem to indicate that it's Nvidia who had to find a way to do better than they had originally planned.

No way Nvidia *planned* on needing over 300w to get the job done. These cards look like they would have been "fine" at 250w, but Nvidia got spooked. They brought their pricing back in line, turned the power up to 11, and launched before they had sufficient supply.

If using the 102 die for the $700 price point was a planned bait-n-switch the whole time, why then did they feel the need to go nuts on power? $700 for even just %20 over a 2080Ti would have been okay in a vacuum. But they felt the need to get *every* bit of performance the could.

Nvidia appears to be worried about something.

 

[Alexrose1uk]

I wouldn't necessarily listen to the AIBs unfortunately. They are the same ones who absolutely said the Radeon VII was never coming from memory and yet that turned up, even if only for 6 months.

As the first wave of AMD cards especially are typically inhouse reference designs, with potentially only one AIB, there's a good chance the AIBs don't have all the information yet.

The fairly recent pictures of boxes of Nvidia and AMD card's with incorrect information they have to put stickers over should highlight its not uncommon the AIBs don't get everything confirmed until fairly late in the day. Most of the AIBs unfortunately tend to be where the leaks come from, so if AMD are playing things close to their chests (and they do seem to be with both RDNA2 and Zen3), then that wouldn't be that unsurprising that they've not released all the info yet. Initial announcement isnt until end of the month. Release will probably not be immediately after than announcement either, so mid November would seem likely.

I suspect some of the AIBs won't get final information until just a few weeks before hand. Stops them screwing up and accidentally putting the info on their websites too early. That has ALSO happened a few too many times.


And yes, I would would suggest Nvidia are possibly worried for 3 reasons:
1) They dropped pricing - Nvidia usually RAISE pricing, not drop it
2) They launched way too early with virtually no stock, rather than waiting even a couple of weeks to give stock more time, despite knowing the competition had no new cards due out for at least a month or two.
3) They shifted the 3070 launch to the day after AMDs announcement; you could easily take that as they don't want the 3070's to be compared to AMD's offering before they've had a chance to make final price alterations etc.

 

[FoxEye]

Only AMD staff? I think quite enough people in the supply chain - both at TSMC and the board partners know everything already. At least they. Not to mention people like Charlie at S|A and the other old dogs.

Why do you think TSMC would have any clue at all about Big Navi's performance?

Just wonder why you think that. Sure they manufacture the chips, but why do you think that would translate into knowing the performance of the end product? Which includes AMD drivers... you think AMD are giving TSMS the drivers and asking them to build/test the cards?

I don't think they are.