16gb 512bit Big Navi sounds so nice. It will be an interesting end of the year.

Undying:

16gb 512bit Big Navi sounds so nice. It will be an interesting end of the year.

Highly unlikely though, cards haven't gone beyond 384bit in some time (except for HBM) because the complex routing requirements for 448 and 512bit cards is expensive and because more modules tends to be more expensive than higher density.

Astyanax:

Highly unlikely though, cards haven't gone beyond 384bit in some time (except for HBM) because the complex routing requirements for 448 and 512bit cards is expensive and because more modules tends to be more expensive than higher density.

I'm skeptical as well. The only way I see something like this happening is a real halo Titan like card but we would also see Titan like prices. Guess we will know soon enough.

Hmmm. This is getting interesting. So we had some rumors of BIG Navi JUST being 15% faster than the 2080ti in AMDoptimised benchmarks.. Then there is a rumor now of it using less than 250Watts then there is the rumor of it not having 72 but 80 CUs... And there has also been a sample found with 56CU... Here is what I think is going on. 56CU 7nm+ EUV 12GB 15% faster than 2080TI (mistaken for the top card) 72CU 7nm+ EUV 16GB 30-30% faster than 2080TI 80CU 7nm+ EUV 16GB 40% faster than 2080TI (HMB???? Beginning of next year when yields are up for full die usage????) But I might be mistaken. Those 15% rumors sounded a lot like a 56CU die... and even though AMD has put through optimisations and 7nm+ EUV, I still think 250watts for the 80CU SKU is too low of an estimate. But we will have to see. Might all be overhyped as usual...... But the simple math of doubling a 5700XT pans out... Navi2X... do the math yourself... Unless they have royally fucked up. (Again)

Cidious:

Hmmm. This is getting interesting. So we had some rumors of BIG Navi JUST being 15% faster than the 2080ti in AMDoptimised benchmarks.. Then there is a rumor now of it using less than 250Watts then there is the rumor of it not having 72 but 80 CUs... And there has also been a sample found with 56CU... Here is what I think is going on. 56CU 7nm+ EUV 12GB 15% faster than 2080TI (mistaken for the top card) 72CU 7nm+ EUV 16GB 30-30% faster than 2080TI 80CU 7nm+ EUV 16GB 40% faster than 2080TI (HMB???? Beginning of next year when yields are up for full die usage????) But I might be mistaken. Those 15% rumors sounded a lot like a 56CU die... and even though AMD has put through optimisations and 7nm+ EUV, I still think 250watts for the 80CU SKU is too low of an estimate. But we will have to see. Might all be overhyped as usual...... But the simple math of doubling a 5700XT pans out... Navi2X... do the math yourself... Unless they have royally ****** up. (Again)

your logic is fine if the assumptions based on the rumors are true.

don't forget that the lesser sku is about 33% chance of being a new mobile... or that they use the same chip for mobile and cut-down model. at 7nm+ the power savings are significant vs. old school radeon. and the mobile section (ryzen) has been proven to be a very lucrative space in a very short time. but maybe my wishes are getting ahead of their intent:p

Back in 2015 PCB manufacturers still knew how to make a 512-bit bus for GDDR, but now in 2020 it's a lost art.

It's not about the PCB, but about the ASIC itself. More bits = More die area dedicated to memory controllers = Bigger die = More lithography failures = More expensive to make. But, harvesting is an option here as well... the 12-GB 384-bit one might just as well be a 16-GB 512-bit one with some memory controllers disabled.

Cidious:

Hmmm. This is getting interesting. So we had some rumors of BIG Navi JUST being 15% faster than the 2080ti in AMDoptimised benchmarks.. Then there is a rumor now of it using less than 250Watts then there is the rumor of it not having 72 but 80 CUs... And there has also been a sample found with 56CU... Here is what I think is going on. 56CU 7nm+ EUV 12GB 15% faster than 2080TI (mistaken for the top card) 72CU 7nm+ EUV 16GB 30-30% faster than 2080TI 80CU 7nm+ EUV 16GB 40% faster than 2080TI (HMB???? Beginning of next year when yields are up for full die usage????) But I might be mistaken. Those 15% rumors sounded a lot like a 56CU die... and even though AMD has put through optimisations and 7nm+ EUV, I still think 250watts for the 80CU SKU is too low of an estimate. But we will have to see. Might all be overhyped as usual...... But the simple math of doubling a 5700XT pans out... Navi2X... do the math yourself... Unless they have royally ****** up. (Again)

Wait a second. AMD mentioned that RDNA 2 has 50% more performance per watt. That kind of improvement usually comes from either process enhancements, or uArch... or both... but we weren't told which applies to RDNA2. Doesn't this muddies the waters a bit... especially since 7nm+ EUV allows for another 20% higher density and 15% improved efficiency over regular 7nm? 2080ti is roughly 35-40% faster than 5700xt. If this 50% more perf. per watt jump is just on uArch level only (which is definitely possible), then 40CU's RDNA 2 version of 5700xt = just about 10-15% more performance than 2080ti (and about 20% reduction to TDP from EUV node). That would put RDNA 2 equivalent of 5700XT with 40CU's at 180W. Any GPU with higher CU's would be (more or less) proportionally faster, because in AMD's case, 10% increase in CU's seems to correlate with same increase in performance (as scaling on AMD is apparently a LOT better vs NV), but as we know, power expenditure rises exponentially depending on how much performance you gain (which depends on voltages, frequencies, and other factors - 5700xt for example can be undervolted to drop power consumption to 5700 levels and improve sustained boosting performance [or just drop core frequencies by 100MhZ without losing any performance and increasing efficiency in ADDITION to undervolting]). So, if that is actually accurate, then full blown RDNA 2 with 80CU 7nm EUV could end up being 50% faster than 2080ti at about 360W (with manual undervolting, probably around 20% lower TDP). AMD should really try automating the process for undervolting their GPU's out of the factory. I do not think HBM 2 will be used. It seems AMD is sticking with GDDR6 for now.

well if AMD want me to purchase another GPU it better have HBM. Not going backwards i hope they dont fail.

Martin5000:

well if AMD want me to purchase another GPU it better have HBM. Not going backwards i hope they dont fail.

I think you may be disappointed then. Now that AMD is splitting the Professional and Consumer cards. I have a sneaking suspicion that Navi 2 cards are going to be special.

At the same time rumors state that the competition (not Intel...) will have: 8/16GB (RTX3070/S) - 256bit 10/20GB (RTX3080/S) - 320bit 24GB (Ti) - 384bit So i hope for AMD that it will be more competitive this time, and can deliver performance not only in the mid-range, but for enthusiasts as well. (Because i doubt that nVidia is just compensating something with more GDDR6, but we will see in september 😛, and probably november from AMDs side)

Martin5000:

well if AMD want me to purchase another GPU it better have HBM. Not going backwards i hope they dont fail.

who cares about the details, if the end you get more fps?

deksman2:

Wait a second. AMD mentioned that RDNA 2 has 50% more performance per watt. That kind of improvement usually comes from either process enhancements, or uArch... or both... but we weren't told which applies to RDNA2. Doesn't this muddies the waters a bit... especially since 7nm+ EUV allows for another 20% higher density and 15% improved efficiency over regular 7nm? 2080ti is roughly 35-40% faster than 5700xt. If this 50% more perf. per watt jump is just on uArch level only (which is definitely possible), then 40CU's RDNA 2 version of 5700xt = just about 10-15% more performance than 2080ti (and about 20% reduction to TDP from EUV node). That would put RDNA 2 equivalent of 5700XT with 40CU's at 180W. Any GPU with higher CU's would be (more or less) proportionally faster, because in AMD's case, 10% increase in CU's seems to correlate with same increase in performance (as scaling on AMD is apparently a LOT better vs NV), but as we know, power expenditure rises exponentially depending on how much performance you gain (which depends on voltages, frequencies, and other factors - 5700xt for example can be undervolted to drop power consumption to 5700 levels and improve sustained boosting performance [or just drop core frequencies by 100MhZ without losing any performance and increasing efficiency in ADDITION to undervolting]). So, if that is actually accurate, then full blown RDNA 2 with 80CU 7nm EUV could end up being 50% faster than 2080ti at about 360W (with manual undervolting, probably around 20% lower TDP). AMD should really try automating the process for undervolting their GPU's out of the factory. I do not think HBM 2 will be used. It seems AMD is sticking with GDDR6 for now.

Maths :P 2080Ti is 35%-40% faster than 40CU RDNA1. (5700XT) If AMD goes for 50% per watt keeping the same power consumption on 40CU RDNA2 that means it would be +15%-10% faster than the 2080Ti. An 80CU RDNA 2 would be double the performance of the 40CU RDNA2 so +130% to 120% faster (over twice as fast) than the 2080Ti. Not 50% :P

Point is. I'll hold on to my cash a bit. I got a $1000 budget this round... I'm not willing to pay $1200 for a 2080ti like scenario. So I'll wait a bit and see what AMD does.. I do feel awkward towards their lack of solid features like DLSS3.0 and their driver development issues... but yeahhhh I'd like to not have to include the GPU into the loop also.. or even a second loop for it... so below 300 watts would be nice...

can u have 10gb or 12gb on a 256bit bus by doing some ram chips at 2gb per chip and 1gb per chip like 4x2gb and 4x1gb giveing 12gb or 2x2gb then 6x1gb which give 10gb??????

Undying:

16gb 512bit Big Navi sounds so nice. It will be an interesting end of the year.

For me it sounds VERY expensive. RTX2080Ti ++ expensive

devastator:

can u have 10gb or 12gb on a 256bit bus by doing some ram chips at 2gb per chip and 1gb per chip like 4x2gb and 4x1gb giveing 12gb or 2x2gb then 6x1gb which give 10gb??????

not on a pc where you don't have metal access to ram usage.

barbacot:

Yes, AMD mentioned...reality could be a different story though... There are a lot of AMD zealots who believe everything that Lisa Su says and think that she sweats holy water or NVIDIA zealots who think that the leather jacket is magic...It's a marketing battle right now between NVIDIA and AMD with intentional leaks, rumors, etc...very soon we will find out the truth.

Yep, would not be surprised if rumors big navi was only 10-15% better than 2080ti was to conceal the potential of a more powerful card. Each side is waiting to screw the other with surprises. Like what happened with the 5600xt when Nvidia rained down on AMDs parade with last minute price reduction of the 2060, thus forcing AMD to make up for it with frantic bios updates to boost performance. Initial reviews are important because they report perf/price comparisons of competition and recommend accordingly. [spoiler]https://i.pinimg.com/originals/c0/07/92/c0079249f3264dfc26f1f7da7d2094d2.jpg [/spoiler]

Astyanax:

not on a pc where you don't have metal access to ram usage.

i meant vram on gpu