You are going to be disappointed. 4080 will definitely not be 2x 3080 12 GB.
Considering the humongous process advantage this sounds about right. Well, maybe a bit on the high side since these are not AD102 based. They better be close to 2x faster or the perf/w improvements (or "improvements") are disappointing.
Yeah, it's closer to 60-70% faster which ain't bad at all.You are going to be disappointed. 4080 will definitely not be 2x 3080 12 GB.
Yeah, it's closer to 60-70% faster which ain't bad at all.
Actually, given the rumored power draw and a full 2 node leap we could see 2x performance and it not be that crazy. Not saying its a sure thing, but going from Samsung's 8nm (rebranded 10nm, itself a half node) all the way to TSMC 5nm is a massive jump. If you go back all the way to 180um in the early 2000s.. we've never seen either NV or AMD/ATi jump two nodes in a single gen.You are going to be disappointed. 4080 will definitely not be 2x 3080 12 GB.
There's that and one would think that architecture improvements would also help. I do expect quite large improvements at same power usage. It's been quite some time when we had this kind of situation. Interesting to see how things turn out...Actually, given the rumored power draw and a full 2 node leap we could see 2x performance and it not be that crazy. Not saying its a sure thing, but going from Samsung's 8nm (rebranded 10nm, itself a half node) all the way to TSMC 5nm is a massive jump. If you go back all the way to 180um in the early 2000s.. we've never seen either NV or AMD/ATi jump two nodes in a single gen.
To me it looks more like kopite doesn't know what he's talking about anymore. At this point his "monday: 420W, wednesday: we can expect 350W, saturday: another update now we're at 320W for the 4080" is just annoying. And all of this "there's another change" stuff is just bs although he won't admit it.
I can give you 5 different TGP/TBP values right now and I can guarantee you one of them will be correct. And for that I don't need to be some sort of "leaker".
If nothing else, to confuse leakers.While I do agree about kopite Nvidia are known to change specs right up to the announcement.
At this point I'm reasonably convinced that some Nvidia engineers are confused as well, if only by the numbers reported in the media.If nothing else, to confuse leakers.
If nothing else, to confuse leakers.
I never believed 420w for the 4080 to begin with. Firstly, it would not look good when it shows up as a con in all reviews. Secondly, Nvidia knows ppl may need to factor in a PSU purchase in the overall cost of the GPU. So whatever AMD equivalent to it with lower power draw will have an advantage. Same argument with 4070 at 300w+.
Actually, given the rumored power draw and a full 2 node leap we could see 2x performance and it not be that crazy. Not saying its a sure thing, but going from Samsung's 8nm (rebranded 10nm, itself a half node) all the way to TSMC 5nm is a massive jump. If you go back all the way to 180um in the early 2000s.. we've never seen either NV or AMD/ATi jump two nodes in a single gen.
It has 7680 FP32 and separate 3840 INT32, so If TFlops at stock boost is similar then It should be faster than 3090 Ti at 1080p or even at 1440p. Performance at 4k is questionable because of 1/2 bus and possibly lower number of ROPs.
That would be 40 TF FP32 at the stock boost number. Which is about what the 3090 Ti is. I do think that you can boost to over 3 Ghz, but obviously not at the stock TBP levels. So it has a decent chance of being faster than the 3090 Ti at 1080p at least.
It has 7680 FP32 and separate 3840 INT32, so If TFlops at stock boost is similar then It should be faster than 3090 Ti at 1080p or even at 1440p. Performance at 4k is questionable because of 1/2 bus and possibly lower number of ROPs.
I know It has more cache, just didn't mention It in my previous post.Yep my thoughts exactly and don't forget nvidia is effectively doing their own version of infinity cache. The 4070 has 12x the cache of the 3070 so that smaller memory interface likely won't matter too much sub 4k.
One of the most interesting things about this generation is going to be just how the cache impacts performance. It seems like the rest of the design is similar to Ampere and Turing (I hope I'm wrong though), so just what kind of difference the cache makes will be an interesting case study. Even at 4K, the advantage will surely not be zero, although the resolution is above the 'sweet spot' that accelerates 1080 and 1440 that we saw on RDNA3.
Samsung 8nm ain't just "rebranded 10nm", it's also very serviceable node. nVidia choose out of pure chase for higher margins, TSMC didn't want to budge on giving them a better deal on N7, nVidia/TSMC are not exactly the best buds in the industry anyway.Actually, given the rumored power draw and a full 2 node leap we could see 2x performance and it not be that crazy. Not saying its a sure thing, but going from Samsung's 8nm (rebranded 10nm, itself a half node) all the way to TSMC 5nm is a massive jump. If you go back all the way to 180um in the early 2000s.. we've never seen either NV or AMD/ATi jump two nodes in a single gen.
memory bw is not everything. look at rx 5700xt vs 6700xt and see what difference a ton of cache does, same thing will happen with 3080 vs 4080, because 3080 has 5MB L2 while 4080 has 64MB L2
4080 memory spec changed to be faster maybe. It is still less memory bandwidth than the 3080.
All of that extra cache is going to take a load off the memory system, so the total bandwidth doesn't need to be as high.
76 SM vs 70 SM is not a lot, true, but these SM are more powerful and clockspeed will be also significantly higher.I suspect memory bandwidth is still going to be an issue.
Worth reminding that the 4080 (as it stands) is only 8% more SMs than the 3080 12 GB.