Yotsugi
Golden Member
- Oct 16, 2017
- 1,029
- 487
- 106
Hence why the smaller one is G5.I reckon AMD won't hand out GDDR6 for cheap..
Yes, should be 2 dies this year.So am i correct that amds launching navi 10/11 and 12 this year ?
Hence why the smaller one is G5.I reckon AMD won't hand out GDDR6 for cheap..
Yes, should be 2 dies this year.So am i correct that amds launching navi 10/11 and 12 this year ?
Hence why the smaller one is G5.
The latter.The 36/40 CU one or the 16/20 CU chip ?
We don't know anything about it, but at least 20CU gfx10 part showed up in benches, so more plausible.If Navi 14, with 20 CUs is GDDR5, then it may be RX 580 performance at 75W, really.
That part was actually 10-15% faster than RX 580, IF I remember correctly.We don't know anything about it, but at least 20CU gfx10 part showed up in benches, so more plausible.
Somewhere there, yes.That part was actually 10-15% faster than RX 580, IF I remember correctly.
Who cares?And it was veeeeeeeeeeeery slow in compute versus the Vega parts.
People that wanted cheap FP crunchers once AMD puts their software together.Who cares?
The latter.
9Gbps G5 should do a-okay.
If Navi 14, with 20 CUs is GDDR5, then it may be RX 580 performance at 75W, really.
Why?It supposedly bringing no uArch improvements besides the node, which is both stupid and outright false.
Arent you contradicting yourself a little ?Why?
1) They are really busy making the console gpu's, not desktop ones - so they don't have much of the dev team to focus on this.
2) They've got to the 7nm shrink which is a load of work.
3) Over the last few years they've basically done a lot of releasing almost the same gpu on a smaller node with even faster memory.
What's so hard to believe that Navi is the same GCN(basically some variant of polaris) we've seen for years on a smaller node (7nm) with faster memory (GGR6)?
The console gpu's are for Sony/MS, and will be substantially different to what is released for the desktop - it's going to be some sort of APU with shared memory and a load of console specific stuff. Not to say they can't share work but the console gpu's come first because they are the ones with the hard contract. They aren't due out for over a year. Navi isn't going to be based on anything new in them.Arent you contradicting yourself a little ?
You say they are busy making console gpus, so you are making the assumption they are just shrinking "vegalis" to 7nm.BUT, these console gpus that they are so hard working on are NAVI`s! This is the same or almost the same architecture, so they must put all the new and exciting stuff on it, because sony and ms want more features and performance, thats why FP16 happened, and thats why larger caches and vrr are happening on navi.
Thats quite imaginative interpretation of a "NAVI architecture" that is implemented in all of this products.You do know that polaris in xbox one and playstation 4 is THE SAME BASE architecture that is in the polaris desktop cards right ?The console gpu's are for Sony/MS, and will be substantially different to what is released for the desktop - it's going to be some sort of APU with shared memory and a load of console specific stuff. Not to say they can't share work but the console gpu's come first because they are the ones with the hard contract. They aren't due out for over a year. Navi isn't going to be based on anything new in them.
What.However, it won't match GTX 1650 without that extra bandwidth advantage, on top of 16 vs 32 ROP deficit.
No, still 64, still 4*16, that was very-very optimal from the get-go.that would mean it contains more shaders per CU
Console APU is a just a slightly customized die that uses their base GPU IP.They are really busy making the console gpu's, not desktop ones - so they don't have much of the dev team to focus on this.
That's, that's not how the fabless world lives.They've got to do the 7nm shrink which is a load of work
Vega is pretty different.Over the last few years they've basically done a lot of releasing almost the same gpu on a smaller node with even faster memory.
Because I can read funky english letters in LLVM patches, and you can't.Why is it so hard to believe that Navi is the same GCN(some variant of polaris) we've seen for years on a smaller node (7nm) with faster memory (GDDR6)?
Literally the very same base IP.and will be substantially different to what is released for the desktop
20 CU Navi GPU will not be slower than GTX 1650. Even with low memory bandwidth.
What.
No, still 64, still 4*16, that was very-very optimal from the get-go.
RX 560 with 1024 Shaders performs similar to GTX 1050 with 640 shaders ( both at 128bit / GDDR5 )
GTX 1050 Ti with 768 Shaders is 15% faster than RX 560. GTX 1650 with 896 Shaders is 60-70% faster than RX 560
Even with 20CU / 1280 shaders, how can they bridge that gap ?
ByEven with 20CU / 1280 shaders, how can they bridge that gap ?
Correct me if im wrong but 560 has 1175mhz clock speed.Assuming this does 1750mhz ,and has 20% higher shader count its already on par, and 560 is GCN 4 and this is going to be GCN 6.
By
*claps*
changing the uArch!
Almost if their current single-issue TLP-centric design leaves a lot of perf on the table.
No idea what did they do to FF there.Will AMD still be employing 16 ROPs on the 1024/1280 shader cards along 32 ROPs on 2048/2560 shader cards ?
Right on money.I presume this 16/20CU chip is on 128 bit with 36/40 CU chip on 256bit bus.
Clock speeds, new architecture, focused on gaming optimized features. Or by increasing internal bandwidth of CU's due to implementing Caches all over the design.RX 560 with 1024 Shaders performs similar to GTX 1050 with 640 shaders ( both at 128bit / GDDR5 )
GTX 1050 Ti with 768 Shaders is 15% faster than RX 560. GTX 1650 with 896 Shaders is 60-70% faster than RX 560
Even with 20CU / 1280 shaders, how can they bridge that gap ?
Better SIMDs are a net win in vacuum.new architecture, focused on gaming optimized features
The thing's almost guaranteed to get better caches and waaaay better reg file.Or by increasing internal bandwidth of CU's due to implementing Caches all over the design.
Looking in "some" places, it appears that really is the case.Better SIMDs are a net win in vacuum.
The thing's almost guaranteed to get better caches and waaaay better reg file.