I imagine this is the reason AMD decided to launch the Radeon Pro Duo chip, to get rid of all the Fiji dies. The box even contains 1 Fiji chip as a souvenir As the ASP is 1499$ they sell it at 500/chip = same as Nano.
The Nano still might hold the lead as M-itx champ though. But Fiji never was a mass product, the 390 was.
If you think about, I don't see any reason why Polaris 10 would be slower than a 390X. It just doesn't make any sense. If 14nm provides 2.2X density advantage of 28nm, that would make a 14nm 232mm ~ 28nm 510mm. Hawaii is 438mm. That means, by comparison, Hawaii is actually SMALLER than Polaris 10 if they're both on the same 28nm node.
Also, Polaris 10 will use a much smaller memory bus (256 vs 512). That should free up some more die space for Polaris 10 for more cores. Rumors suggest Polaris 10 will use gddr5x. That will solve the bandwidth problem.
Less not forget about the higher clocks achievable through 14nm FF. Most 390/x should be able to hit 1100mhz. If Polaris 10 is able to clock upwards of 1300mhz, it might actually be able to reach Nano/fury level of performance.
Basically, if you just do the math, there shouldn't be any doubt that Polaris 10 will be at the very least, faster than a 390x.
The top bins, with Gddr5x, will be faster than 390x that bit I'm sure of. Whether it also bests the Titan X, or even 1070, depends on the clocks we'll see on 14nm.Basically, if you just do the math, there shouldn't be any doubt that Polaris 10 will be at the very least, faster than a 390x.
Don't forget, this generations high end card, the GTX 980 Ti launched at $649.
We already know about the pascal high end card, the P100. It's that card that should be launching at $699 not the midrange parts.
If you think about, I don't see any reason why Polaris 10 would be slower than a 390X. It just doesn't make any sense. If 14nm provides 2.2X density advantage of 28nm, that would make a 14nm 232mm ~ 28nm 510mm. Hawaii is 438mm. That means, by comparison, Hawaii is actually SMALLER than Polaris 10 if they're both on the same 28nm node.
Also, Polaris 10 will use a much smaller memory bus (256 vs 512). That should free up some more die space for Polaris 10 for more cores. Rumors suggest Polaris 10 will use gddr5x. That will solve the bandwidth problem.
Lets not forget about the higher clocks achievable through 14nm FF. Most 390/x should be able to hit 1100mhz. If Polaris 10 is able to clock upwards of 1300mhz, it might actually be able to reach Nano/fury level of performance.
Basically, if you just do the math, there shouldn't be any doubt that Polaris 10 will be at the very least, faster than a 390x.
Btw, does anybody want to have an educated guess as to why NV is paper launching now, with EARLY ACCESS "Founders Edition" at a $100 premium on the 27th and custom cards due at a later unspecified date?
Think about it.
More than likely, could also beat it handily with GDDR5x.The big polaris should go head to head with the 1070 right?
More than likely, could also beat it handily with GDDR5x.
Nowhere close, 256 vs 337 GBps, unless that's a typo?Here's the thing, the 1070 has GDDR5 8Gbps. As fast as Titan X (+5%).
Why do people jump to the conclusion that Polaris 10 is somehow bandwidth starved that it can't hit Fury X performance without GDDR5X?
Tonga's memory compression according to AMD is ~40% effective, onpar with Maxwell gen 2. Polaris has enhanced memory compression.
It also has a new Pre-fetch & cache feature that reduces bandwidth requirements.
It's Discard Accelerator removes most of the complexity in a given scene.
Polaris 10's bandwidth efficiency should be great.
Nowhere close, 256 vs 337 GBps, unless that's a typo?
It isn't, however at all resolutions (like 4K) it needs GDDR5x to beat Fury X, that or HBM which it won't have.
It still won't be enough to even match a fury Nano, when it comes to just the bandwidth, & recent benches show that HBM (basically mem bandwidth) can make a huge difference at high(er) resolutions.
We still dpn't know how well these new features perform, therefore effective bandwidth is not a number I'm looking at. Even ignoring the competition all I see is a gaping hole between, Vega with HBM & Polaris with GDDR5, the top & mid tier GPUs. I don't see how this gap will be filled without small Vega having GDDR5x, highly unlikely but possible, or big Polaris having GDDR5x.
That's not a typo. 1070 has low bandwidth but it's got Titan X equivalent performance.
You don't need raw bandwidth if you improve utilization and memory efficiency.
People seem to apply different rules, like Pascal must be better, it's got superior tech so it can achieve such performance with gimped GDDR5 on a 256 bit bus... but somehow Polaris can't, because we all know it's inferior or something.
It's crazy logic. Polaris is a big leap for GCN.
Some interesting questions raised ...
Maybe GTX 1080 availability will be very low because GDDR5X mass production is still not started.
http://www.bitsandchips.it/52-english-news/6972-some-questions-about-new-gp104-gpu
if pascal cant do async compute well then the cards have no future for gaming.
...so no FreeSync still?
No it doesnt sound ridiculous if achieving 20-25% better than their flagship. This is typical pricing trend since long ago (for both Nv and AMD), only interrupted with the lower than usual $329 for the 970. What were the intro prices for the 7950/7970, 290/x?Anybody else unimpressed with new midrange Nvidia GPUs, and much higher MSRP?
Yes quite a bit unimpressed and disappointed.Wished it would have been $300 for 1070 and $450 for 1080.It sounds ridiculous doesn't it?
Its crazy..Reference 1080 for 700USD(founder edition) and reference 1070 for 450USD.Anybody else unimpressed with new midrange Nvidia GPUs, and much higher MSRP?
Yes quite a bit unimpressed and disappointed.Wished it would have been $300 for 1070 and $450 for 1080.It sounds ridiculous doesn't it? But that's the price/performance jump i would have liked to see.
With the amount of hype that preceded the launch,nothing less than that would have impressed me.
The only thing i am definitely impressed by is Nvidia gave 8gb memory to the 1070.Even though its only GDDR5 but its 8gb of very fast gddr5 memory.
But again $400 for custom 1070 cards is a bit much i hope this we can agree on.