AMD Says RDNA3 50% Perf/Watt increase and mentions RDNA4

Published by

Click here to post a comment for AMD Says RDNA3 50% Perf/Watt increase and mentions RDNA4 on our message forum
https://forums.guru3d.com/data/avatars/m/248/248994.jpg
I hope RT performance gets closer to Nvidia's level. Otherwise I do appreciate the claimed performance/watt priority.
https://forums.guru3d.com/data/avatars/m/278/278016.jpg
Kaarme:

I hope RT performance gets closer to Nvidia's level. Otherwise I do appreciate the claimed performance/watt priority.
I personally believe that AMD's goal is the same performance with rasterization and 60 frames in 4k with rtx on, it is a given that Nvidia will be much faster with rtx on but if it manages this 60 AMD the benchmarks will not matter for me at least.
https://forums.guru3d.com/data/avatars/m/250/250418.jpg
Kaarme:

I hope RT performance gets closer to Nvidia's level. Otherwise I do appreciate the claimed performance/watt priority.
RT is still just something you toggle on or off. Hardware still needs to catch up allot so you can do everything in RT, then I'd worry about performance of AMD GPUs. I'm really exited for these GPUs, but sad I won't be able to afford any. The 50% performance/watt though, really exiting! Top cards can get up to 50% more performance and lower cards can run on PCIe power!
https://forums.guru3d.com/data/avatars/m/248/248291.jpg
Accordingly to TSMC, N5 vs N7, has a 40% power reduction, at the same clocks. Or a 20% higher clock, at the same power. AMD is probably going to use higher clocks at the same power. So maybe will see an increase in clock speeds closer to 3Ghz.....
https://forums.guru3d.com/data/avatars/m/248/248994.jpg
Horus-Anhur:

Accordingly to TSMC, N5 vs N7, has a 40% power reduction, at the same clocks. Or a 20% higher clock, at the same power. AMD is probably going to use higher clocks at the same power. So maybe will see an increase in clock speeds closer to 3Ghz.....
200-220W gaming consumption is acceptable for me. That's where 6700 XT and 3060 Ti sit. If the next gen uses that much for better performance, it would be nice. However, if the prices are outrageous, then I suppose I'd need to settle for a lesser model, which would have 6700 XT or 3060 Ti level of performance but wattage around 160-170, I'd guess.
https://forums.guru3d.com/data/avatars/m/242/242471.jpg
Remember older 5000-6000 series with tessellation, 6000 was somewhat ok, 7000 was good. I think it will be the same now with RT.
https://forums.guru3d.com/data/avatars/m/270/270008.jpg
Kaarme:

200-220W gaming consumption is acceptable for me. That's where 6700 XT and 3060 Ti sit. If the next gen uses that much for better performance, it would be nice. However, if the prices are outrageous, then I suppose I'd need to settle for a lesser model, which would have 6700 XT or 3060 Ti level of performance but wattage around 160-170, I'd guess.
Regardless of prices I will stay under 250w. I set money aside for a 3080 but never received one but didn't try too terribly hard. However I will not put a space heater in my PC as I have been down that road with SLI etc and no thanks I don't like to sweat while gaming.
https://forums.guru3d.com/data/avatars/m/282/282473.jpg
amd says rdna1 to rdna2 is +65% perf/wat while their most efficcient 6800 is around 45% best case scenario in reality, so let's wait and see
https://forums.guru3d.com/data/avatars/m/270/270008.jpg
I expect 2x performance and similar wattage draws to RDNA2 based models.
https://forums.guru3d.com/data/avatars/m/282/282473.jpg
JamesSneed:

I expect 2x performance and similar wattage draws to RDNA2 based models.
that would be +100% performance per wat
https://forums.guru3d.com/data/avatars/m/80/80129.jpg
Kind of curious if it's 50% across the board in games or it varies heavily based on workload. One of the biggest flags against MCM on GPUs was issues with scheduling.. some workloads would be really difficult to spread across MCM without some major scheduling issues.
https://forums.guru3d.com/data/avatars/m/282/282473.jpg
Horus-Anhur:

No. AMD said it's 50% performance per watt improvement. So their figure is very close to what we got. Of course performance efficiency will always vary with SKU, as clock speeds vary. https://www.thefpsreview.com/wp-content/uploads/2020/06/amd-rdna-2-perf-watt-improvement-june-presentation-1024x576.jpg
wrong again check their page for rdna2 https://www.amd.com/en/technologies/rdna-2 best case scenario would be 6800 at 45% 6700xt is about 28-30% over 5700xt
https://forums.guru3d.com/data/avatars/m/271/271560.jpg
+ 50% ? yes please. especially if that +50% is based on the refresh (doubtful). the rx6750 is a price point rockstar for 1440p, so it would be great for that market to get some love as 4k is still far too expensive (for most people) this could be the $500-600 target for a lot of people
https://forums.guru3d.com/data/avatars/m/282/282473.jpg
tunejunky:

the rx6750 is a price point rockstar for 1440p
here comes amd marketing departament well it's slower than 3070 and costs more so it's a pretty extravagant rockstar
RadeonRX 6950 XT RX 6750.png
https://forums.guru3d.com/data/avatars/m/248/248291.jpg
cucaulay malkin:

wrong again check their page for rdna2 https://www.amd.com/en/technologies/rdna-2 best case scenario would be 6800 at 45% 6700xt is about 28-30% over 5700xt
Yet all their press releases stated, at launch, a 50% perf per watt. Even their own public presentation of RDNA2. And that number was repeated multiples times, over and over again. Published in all sites, including Guru3d.
https://forums.guru3d.com/data/avatars/m/282/282473.jpg
Horus-Anhur:

Yet all their press releases stated, at launch, a 50% perf per watt. Even their own public presentation of RDNA2. And that number was repeated multiples times, over and over again. Published in all sites, including Guru3d.
I think amd.com is a good source
https://forums.guru3d.com/data/avatars/m/248/248291.jpg
cucaulay malkin:

I think amd.com is a good source
Dr. Lisa Su is a better source. At the minute 3:50 [youtube=oHpgu-cTjyM]
Untitled.jpg
https://forums.guru3d.com/data/avatars/m/271/271560.jpg
Denial:

Kind of curious if it's 50% across the board in games or it varies heavily based on workload. One of the biggest flags against MCM on GPUs was issues with scheduling.. some workloads would be really difficult to spread across MCM without some major scheduling issues.
i doubt a large variance in loads as that was one of the first places the uArch design team went. they have taken a long term generational approach to this for a long time with each previous iteration having (and failing to) to perform every task, especially heavy compute etc... nothing has made more of a difference for AMD than Infinity Fabric. Lisa Su focused on the transformational effect this has for the future of AMD. it has already knocked Intel on it's backside (and out of smugness) and next up is Nvidia (as far as AMD is concerned).
https://forums.guru3d.com/data/avatars/m/282/282473.jpg
Horus-Anhur:

Dr. Lisa Su is a better source. At the minute 3:50 [youtube=oHpgu-cTjyM]
yet their site says 65%, almost double the number whatever the case, they're both unreal numbers except for one off scenarios like 6800 vs 5700xt at 4K which is over 40% like I said, sku to sku, 6700xt achieved 30% over 5700xt and 6500xt was worse than 5600xt they like to have fun with numbers if this is indeed the same 50% they're claming with rdna3, it shouldn't be hard to extrapolate the performance based on 50% of rdna2 vs rdna1, the only unknown is the power limit. if they stay at 300w, they'll ofer 1.5x of 6900xt, and I bet they'll want more