I believe Polaris 10 will be up to 75W TDP (no external power) and Polaris 11 will be up to 150W TDP
I believe Polaris 10 will be up to 75W TDP (no external power) and Polaris 11 will be up to 150W TDP
You know that's pretty damn close to 4X better perf/W than the R9 380, right?My guess is R9 380 performance at 50W.
That seems pretty reasonable for ref models, though I'm sure some will come with a larger cooler and 6pin. Polaris 11 is still hard to ascribe a power target for, since we don't really know anything about it. The 232mm² rumour is extremely weak, but if it did end up being that big I would expect the top model to come in at around that.
You know that's pretty damn close to 4X better perf/W than the R9 380, right?
You have to compare it to nano. Who's 2x the performance of 380 at same watts.
Erm...
According to this: http://www.digitaltrends.com/computing/samsung-mass-producing-hbm2/ Samsung started mass production of HBM2 4 GB chips. 4Hi-Stacks to be precise.
Later in the year they will go for 8 GB Hi-Stack chips. That means, that 4 GB configs may be possible in Q2, not Q3. I know Charlie said that HBM2 configs will be Q3 at least, but how accurate he can be? 1 stack brings 256 GB/s, right? 2 stacks of 2 GB HBM2 would made for 512 GB/s and cost effective option for mid range chip. Or am I misunderstanding something?
Edit: Also, HBM would account for smaller Memory Controller, and... more Compute Units in the GPU.
I have to calm down...
Maybe that GPU with 2048 Bit VRAM that was posted earlier IS Polaris 11?
http://images.anandtech.com/doci/9969/sk_hynix_hbm2_implementations.png Look at the 2 GB stack, and compare it to this: http://abload.de/img/4545vconw.jpg
Im sorry for hyping it up, but EVERYTHING so far falls in line perfectly.
That seems pretty reasonable for ref models, though I'm sure some will come with a larger cooler and 6pin.
You know that's pretty damn close to 4X better perf/W than the R9 380, right?
You forget some new posters claimed that Fury X will be 40% faster then Titan X and they diapered after Fury X release. Lack of hype control from AMD.Remember the 1.25Ghz Fiji from that site
And 3GB for 2048bit HBM also sounds odd.
What's with the (3GB DDR5 2GHZ 2048-bit) in the SiSoft benchmark? Can you explain this?Erm...
According to this: http://www.digitaltrends.com/computing/samsung-mass-producing-hbm2/ Samsung started mass production of HBM2 4 GB chips. 4Hi-Stacks to be precise.
Later in the year they will go for 8 GB Hi-Stack chips. That means, that 4 GB configs may be possible in Q2, not Q3. I know Charlie said that HBM2 configs will be Q3 at least, but how accurate he can be? 1 stack brings 256 GB/s, right? 2 stacks of 2 GB HBM2 would made for 512 GB/s and cost effective option for mid range chip. Or am I misunderstanding something?
Edit: Also, HBM would account for smaller Memory Controller, and... more Compute Units in the GPU.
I have to calm down...
Maybe that GPU with 2048 Bit VRAM that was posted earlier IS Polaris 11?
http://images.anandtech.com/doci/9969/sk_hynix_hbm2_implementations.png Look at the 2 GB stack, and compare it to this: http://abload.de/img/4545vconw.jpg
Im sorry for hyping it up, but EVERYTHING so far falls in line perfectly.
You forget some new posters claimed that Fury X will be 40% faster then Titan X and they diapered after Fury X release. Lack of hype control from AMD.
No doubt, and that doesn't disagree with what I said. There's also OC'd 750Ti's with good cooling and a 6pin power connector, BTW.I think AMD is targetting Polaris 10 at GPUs which can run off the motherboard PCI-E power. AMD saw Nvidia rake in the moolah with 750 Ti and I think there is a significant demand for such GPUs which can run off the motherboard PCI-E power especially OEM systems.
Tonga was a pathetic chip in terms of balance of resources. It had the front end of Hawaii with a 384 bit memory controller with 128 bits permanently disabled. AMD also has some serious efficiency problems (perf/sp, perf/watt and perf/sq mm) which they are going to address with Polaris.
http://www.pcper.com/news/Graphics-...a-384-bit-Memory-Bus-Not-Enabled-Any-Products
btw the system with Polaris 10 chip demoed at CES was drawing 55-60w lower than GTX 950 based system (all other components remained exact). GTX 950 has a TDP of 90w. I would say the Polaris chip was drawing 35w, could be even lower given that Nvidia has a much lower CPU overhead in DX11 than AMD GCN(even if Polaris improves that I doubt AMD's DX11 CPU overhead can match Nvidia).
http://venturebeat.com/2016/01/15/a...-to-full-graphics-immersion-with-16k-screens/
"We have two versions of these FinFET GPUs. Both are extremely power efficient. This is Polaris 10 and thats Polaris 11. In terms of what weve done at the high level, its our most revolutionary jump in performance so far. Weve redesigned many blocks in our cores. Weve redesigned the main processor, a new geometry processor, a completely new fourth-generation Graphics Core Next with a very high increase in performance. We have new multimedia cores, a new display engine. This is very early silicon, by the way. We have much more performance optimization to do in the coming months."
It was very early silicon with a lot of optimizations being worked on. The GPU was clocked at 850 Mhz. So yeah I think getting to R9 380 perf is definitely possible at 50-55w.
Do you see this kind of thing from Nvidia users? Did anyone claimed nonsense like this with Maxwell or Pascal?LOL. You expect AMD to have some level of control over forum trolls? This may be one of the most absurd things I've read this month
What's with the (3GB DDR5 2GHZ 2048-bit) in the SiSoft benchmark? Can you explain this?
Do you see this kind of thing from Nvidia users? Did anyone claimed nonsense like this with Maxwell or Pascal?
Tonga may be the red-headed stepchild of AMD's current lineup, but its efficiency isn't out of line with the other GDDR5-based GCN GPUs. I would have compared it to Pitcairn, but you suggested Tonga.
Look at it another way. The 750Ti is designed from the ground up for efficiency, and shares many of power saving features of Maxwell like a lack of a hardware scheduler, etc. It pretty much tops the charts for current GPUs in efficiency terms. A 380 is 83% faster than a 750Ti at 1080p, which consumes 60W in gaming loads. For P10 to get 380 levels of performance at 50W, it would need 2.2 times the perf/W of the 750Ti. You don't see that as being a little out of line with what's been stated for the process shrink?
Unfortunatelly I can't.
There was one guy, called Cloudfiresomething. You should have seen how he was hyping the Maxwell on Notebookreview forum. That was good laugh.
Unfortunatelly I can't.
There was one guy, called Cloudfiresomething. You should have seen how he was hyping the Maxwell on Notebookreview forum. That was good laugh.
I think that he was the guy who had the super-connection with all of the inside info on the 300 series, including how it was all going to be using HBM from top to bottom of the stack.
At least he manages to be wrong on both sides of the argument.
Now we know that out of Tonga's 360 sq mm , a 128 bit memory controller was disabled. We also saw Nvidia match R9 380 with GTX 960 using a 128 bit memory controller running GDDR5 at 7 Ghz. So one of the key problem points of the current GCN is bandwidth efficiency. Polaris mentions improved memory compression as a key feature and has a completely new L2 cache and memory controller.
Add to it a significantly improved GCN micro architecture with improved shading efficiency and higher clocks on a new FINFET process which has vastly lower leakage and its not unrealistic to expect R9 380 performance from Polaris 10.
What's with the (3GB DDR5 2GHZ 2048-bit) in the SiSoft benchmark? Can you explain this?
Well, a 380 isn't equal to a 960, it's around 20% faster than it and ~35% faster than the 950 it was demoed against by AMD. That being said I could see versions of P10 hitting performance at launch, though even that would be a stretch depending on how much smaller than GK107 or Cape Verde it is. That wasn't the point of contention though. I would be willing to bet that they won't get there at 50W though. Even AMD aren't claiming anywhere near that kind of increase in perf/W.
LOL. You expect AMD to have some level of control over forum trolls? This may be one of the most absurd things I've read this month
As i have said before a Polaris 10 at 75W TDP could be very close to Tonga R9 380 at 1080p.
Well, a 380 isn't equal to a 960, it's around 20% faster than it and ~35% faster than the 950 it was demoed against by AMD. That being said I could see versions of P10 hitting performance at launch, though even that would be a stretch depending on how much smaller than GK107 or Cape Verde it is. That wasn't the point of contention though. I would be willing to bet that they won't get there at 50W though. Even AMD aren't claiming anywhere near that kind of increase in perf/W.