• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD RDNA3 unveiling event

Gamers Nexus RDNA3 Q&A with the man who designed it.

Steve asks the one major question we all wanted to know and now we have the answer.

The question: Why does RDNA3 only have the single GPU logic chiplet but multiple memory chiplets.

The answer: The amount of data that would need to flow near instantaneously across is gigantic (many many terabytes per second). It's harder to do compared to the CPU and at this stage when we do it for the GPU it creates a lot of delays/latency because that connection can't handle all the vertices, textures etc in realtime that would need to be routed through it.

 
Last edited:
Grim5 said:
Gamers Nexus RDNA3 Q&A with the man who designed it.

Steve asks the one major question we all wanted to know and now we have the answer.

The question: Why does RDNA3 only have the single GPU logic chiplet but multiple memory chiplets.

The answer: The amount of data that would need to flow near instantaneously across is gigantic (many many terabytes per second). It's harder to do compared to the CPU and at this stage when we do it for the GPU it creates a lot of delays/latency because that connection can't handle all the vertices, textures etc in realtime that would need to be routed through it.

Click to expand...
I’m about to watch the video but just a quick take away from what you’ve said and maybe he does address it in the video. It is interesting that the issue holding it up isn’t about having the two GCDs operating coherently. Could it be that AMD has already solved this and memory bandwidth is now the last thing standing in their way of a glorious dual GCD design. Sounds like they may already have some parts in the lab.

Does AMD have any patents for fast interconnecting technology that anyone is aware that could solve this for them?
 
Last edited:
Chuk_Chuk said:
Does AMD have any patents for fast interconnecting technology that anyone is aware that could solve this for them?
Click to expand...
I did read something that said that the new interconnecting technology is 10 times faster than the old Infinity Fabric, but it may have been regarding the MCD's and the main die I can't quite remember. Will watch this later :)
 
Chuk_Chuk said:
I’m about to watch the video but just a quick take away from what you’ve said and maybe he does address it in the video. It is interesting that the issue holding it up isn’t about having the two GCDs operating coherently. Could it be that AMD has already solved this and memory bandwidth is now the last thing standing in their way of a glorious dual GCD design. Sounds like they may already have some parts in the lab.

Does AMD have any patents for fast interconnecting technology that anyone is aware that could solve this for them?
Click to expand...

He didn't rule it out, just say it's not easy. I still think it's coming, could be RDNA4 who knows, they already managed a 10x improvement in IF bandwidth for RDNA3 vs what Ryzen has and so the data link between the memory Chiplets operates at 9.2GB/s. The guy didn't say how fast it would need to be to support multiple GPU GCD chiplets; but also worth noting he specifically mentioned that you require high speed because you are transferring things like game texture so potentially we're talking hundreds of gigabyte/thousands of gigabytes so may required another 100x improvement. Ignoring gaming, multiple GCD chiplet design GPUs for things like servers would be easier since they're doing calculations and not passing large amount of bandwidth and latency intensive game assets
 
Last edited:
Grim5 said:
He didn't rule it out, just say it's not easy. I still think it's coming, could be RDNA4 who knows, they already managed a 10x improvement in IF bandwidth for RDNA3 vs what Ryzen has and so the data link between the memory Chiplets operates at 9.2GB/s. The guy didn't say how fast it would need to be to support multiple GPU GCD chiplets; but also worth noting he specifically mentioned that you require high speed because you are transferring things like game texture so potentially we're talking hundreds of gigabyte/thousands of gigabytes so may required another 100x improvement. Ignoring gaming, multiple GCD chiplet design GPUs for things like servers would be easier since they're doing calculations and not passing large amount of bandwidth and latency intensive game assets
Click to expand...
I think the last part is incorrect. in servers they are generally handling large datasets hence nvlink while it was around would be on all the quadros and they generally have more memory than the equivalent gaming version.
 
Trypta said:
Do they actually do that? I've booked a week off work from the 13th and the girlfriend is away, I'll be gutted if I can't get my hands on one!
Click to expand...
last i heard there is no main area you can buy anything from, someone made a post saying about coming to collect one and gibbo said store is not open
 
Firegod said:
Doesn't seem that low. In Nvidia's response to the melting cable issue, board partners told Gamers Nexus that they have sold 125,000 RTX4090 cards. So AMD's figures seem to align with Nvidia supply / sales to date.
Click to expand...

Don't forget that's their reduced 'sell in' constricted rate to the supply chain, so is going to be low effectively by definition.

I do wonder what their typical launch quarter supply for flagships has been historically.
Don't suppose anyone knows/has any sources to hand?
 
Last edited:
I'm willing to buy a 4090, but if Nvidia can't (or won't) sell me one, the 7900XTX may do the job for me, and the latest rumor is that AMD actually wants to sell their new cards.

Holding back supply will not push me into a 4080 or any of the Ampere stuff they want to move.
 
Last edited:
Twinz said:
I'm willing to buy a 4090, but if Nvidia can't (or won't) sell me one, the 7900XTX may do the job for me, and the latest rumor is that AMD actually wants to sell their new cards.

Holding back supply will not push me into a 4080 or any of the Ampere stuff they want to move.
Click to expand...
The 4090 is in stock at MSRP.

www.overclockers.co.uk

Palit GeForce RTX 4090 OmniBlack 24GB GDDR6X PCI-Express Graphics Card

Order Palit GeForce RTX 4090 OmniBlack 24GB GDDR6X PCI-Express Graphics Card now online and benefit from fast delivery.
www.overclockers.co.uk www.overclockers.co.uk
 
Grim5 said:
The answer: The amount of data that would need to flow near instantaneously across is gigantic (many many terabytes per second). It's harder to do compared to the CPU and at this stage when we do it for the GPU it creates a lot of delays/latency because that connection can't handle all the vertices, textures etc in realtime that would need to be routed through it.

Click to expand...
They should just get together with Microsoft and reboot the CF/SLI days through some common factors, so that it works and scales well across the board.
 
skullkid said:
didn't someone tear palit a new one with regards to how badly this card is built?
Click to expand...
Not that I know of, in all likelihood it will have superior cooling to the founders edition card. MSRP price is pretty poor though at the current exchange rate, it should be £1599 now. Notice how quick they were to put the MSRP up when the exchange rate went down, not so much in the other direction.
 
DeliciousStorage said:
Not that I know of, in all likelihood it will have superior cooling to the founders edition card. MSRP price is pretty poor though at the current exchange rate, it should be £1599 now. Notice how quick they were to put the MSRP up when the exchange rate went down, not so much in the other direction.
Click to expand...
www.youtube.com

GPU PCB Breakdown: Palit RTX 4090 GameRock OC // BIG OOF

Techpowerup review: https://www.techpowerup.com/review/palit-geforce-rtx-4090-gamerock-oc/TL;DW/text version: https://www.patreon.com/posts/73321881My Patreo...
www.youtube.com www.youtube.com

"i can't believe nvidia approved this" lol
 
You must log in or register to reply here.
Back
Top Bottom