Just a thought, whats stopping AMD adding another GCD to the current PCB. I know it wont double perf as they would likely have to drop clocks a bit to keep in a reasonably power target but anther 50% on top. As well as from what Ive read double stack the catch so even more through put. Where would that land in relation to the 4090 ?
-
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
- Thread starter CAT-THE-FIFTH
- Start date
AMD/NV's profit at this point in time is already off the scale, there's probably miles to manoeuvre in pp.AMD also have excess stock.
Neither of them can afford to write them off completely as losses, and whle they are getting cheaper, this is down from £630 to £560, they might have to get cheaper still to get people to buy them in the numbers they need to clear them out, AMD may have to take somewhat of a loss on them to avoid a total loss.
Gigabyte Radeon RX 6800 Gaming OC 16GB GDDR6 PCI-Express Graphics Card
Order Gigabyte Radeon RX 6800 Gaming OC 16GB GDDR6 PCI-Express Graphics Card now online and benefit from fast delivery.www.overclockers.co.uk
Drivers. Getting it to all work together, like a cohesive unit.
It could be their grand plan for RDNA 4.
Last edited:
Associate
- Joined
- 19 Sep 2022
- Posts
- 1,010
- Location
- Pyongyang
there seems to be a latency issue with mcm designs, especially with games using temporal techniques, we still have sli or crossfire to fall back on
it would be interesting to find out how the design scales though
it would be interesting to find out how the design scales though
Last edited:
I did wonder that, Though the expeience from their zen CPU's would have greatly helped I would expect. But you are proberbly right that it will be a RDNA4 thing. That may genuinely be a real pain in NV's side if AMD cracks it!
Associate
- Joined
- 31 Jan 2012
- Posts
- 2,173
- Location
- Droitwich, UK
Would having a separate chiplet for RT work? If so it would make sense if AMD were aiming towards that eventually, if they manage to have acceptable latency.
Workstation GPU's (CDNA) already do this, i don't know if it would work for gaming GPU's, it may need some more engineering work, but once cracked, if not already, holly poo.... RIP Nvidia.
Last edited:
Slow down buddy, I said could not would. I don't need leaker channels quoting me for the rumour mill on RDNA 4
Last edited:
Associate
- Joined
- 19 Sep 2022
- Posts
- 1,010
- Location
- Pyongyang
its a long term strategic vision, thats how nvidia wants to differentiate - i mean you got to look at their presentations and white papers, rasterization isnt a big priority for them and that shows in their die allocation as well. so its going to be rt+dlss for nvidia going forward. i personally find the prospect of dlss exciting, its supposed to be 4x more energy efficient than traditional methods, and looks like the likely upgrade path to 8k without running into energy barriers, the tech's still in its infancy but i find initial implementations promising
Had forgotten about CDNA. Latancy would also be an issue for those as much as it would be for gaming I would have thought so they may well have gotten over the top of the hill with it.
Could be interesting where they go with it .
For now separating the Memory Controllers and Cache from the main die is quite a significant thing in its self.
SDRAM does not scale anything like Logic with node shrinks, there is almost no scaling at all, so 1MB of Cache is the same size on 5nm as it is on 7nm.
If you can remove that from the Logic and put it on a cheap older node, because it doesn't matter anyway, then you can bring your costs down and yields up.
This is exactly what AMD Did, the Logic is on 5nm while the Chache is on 6nm, the expensive bit, the logic is only 300mm2^, compared to the 600mm2^ Nvidia die with similar performance which on top of that is also on an even more expensive 4nm, the cache around AMD's 300mm2^ logic die costs pennies.
That's how AMD are able to make a 4090 class card at $600 less retail and still make decent margins.
SDRAM does not scale anything like Logic with node shrinks, there is almost no scaling at all, so 1MB of Cache is the same size on 5nm as it is on 7nm.
If you can remove that from the Logic and put it on a cheap older node, because it doesn't matter anyway, then you can bring your costs down and yields up.
This is exactly what AMD Did, the Logic is on 5nm while the Chache is on 6nm, the expensive bit, the logic is only 300mm2^, compared to the 600mm2^ Nvidia die with similar performance which on top of that is also on an even more expensive 4nm, the cache around AMD's 300mm2^ logic die costs pennies.
That's how AMD are able to make a 4090 class card at $600 less retail and still make decent margins.
Last edited:
Soldato
- Joined
- 30 Jul 2012
- Posts
- 2,775
Only ever interested in the toxic range.
Forgot to add, what AMD could also do is grow that 300m2^ die into something larger by adding more shaders, if they need a wider bus and more cache to drive more shaders they can just "Glue" more cheap chips around it.
Brilliant!
I want to give you another perspective.
The RX 6700XT at 335mm2^ is larger than the Logic die of the 7900XTX.
What AMD have cracked here is huge.
www.techpowerup.com
The RX 6700XT at 335mm2^ is larger than the Logic die of the 7900XTX.
What AMD have cracked here is huge.
AMD Radeon RX 6700 XT Specs
AMD Navi 22, 2581 MHz, 2560 Cores, 160 TMUs, 64 ROPs, 12288 MB GDDR6, 2000 MHz, 192 bit
Last edited:
Last gen chips are 7nm, the new ones are 5nm so a 6700XT would be smaller @5nm. They also pulled out the IF cache and mem controllers. I expected AMD to have an IO die and several compute dies like Ryzen as that would have made them easy to scale, high end with 2+ compute dies, lower end with less. Maybe next generation will do that although GPU’s have more stages than CPU’s so it will be hard to decuple them without adding latency.I want to give you another perspective.
The RX 6700XT at 335mm2^ is larger than the Logic die of the 7900XTX.
What AMD have cracked here is huge.
AMD Radeon RX 6700 XT Specs
AMD Navi 22, 2581 MHz, 2560 Cores, 160 TMUs, 64 ROPs, 12288 MB GDDR6, 2000 MHz, 192 bit
Last edited:
- Joined
- 9 Nov 2009
- Posts
- 25,333
- Location
- Planet Earth
At least AMD launched the RX7900XTX at $100 below the price of the RX6950XT and has dropped pricing a decent amount on the RDNA2.0 dGPUs.
Nvidia,OTH,has barely reduced prices,and the RTX4060 was to become a £900 RTX4080 12GB and the RTX4070 is a £1300 RTX4080 16GB. Their new dGPUs are doing nothing to reduce the Ampere dGPU prices. They are not even responding to the AMD price cuts!
This is the cheapest 3080.
At $1.14 to £ that should be £613 + 20% VAT = £735, its £780. Nvidia are still charging higher than MSRP, its like an addiction, they just can't help themselves stubbornly trying to get us to agree to these inflated prices.
Inno3D GeForce RTX 3080 Twin X2 OC LHR 10GB GDDR6X PCI-Express Graphics Car
Order Inno3D GeForce RTX 3080 Twin X2 OC LHR 10GB GDDR6X PCI-Express Graphics Car now online and benefit from fast delivery.
www.overclockers.co.uk
Last edited:
Cheapest 3070, should be £525, is £549.
www.overclockers.co.uk
I hope Nvidia have $5 Billion tied up in this stock and it all sits on shelves until it turns to dust.
Zotac GeForce RTX 3070 Twin Edge LHR 8GB GDDR6 PCI-Express Graphics Card
Order Zotac GeForce RTX 3070 Twin Edge LHR 8GB GDDR6 PCI-Express Graphics Card now online and benefit from fast delivery.
www.overclockers.co.uk
I hope Nvidia have $5 Billion tied up in this stock and it all sits on shelves until it turns to dust.
Last edited:
Soldato
- Joined
- 16 Aug 2009
- Posts
- 8,065
Marketing, mindshare, call it what you will but its real reading general gaming forums the opinion was buy Nvidia doesn't matter what variety so long as its Nvidia. AMD are no good, drivers are crap, etc, etc just repeated verbatim. No questioning no "hang on a minute..." just rumour, hearsay, buzz if you will expressed as fact.