Again, not my point... guess I should spell it out: Given the performance characteristics MS has claimed, one may reasonably concluded the power consumption of such a device will be higher than RX480. This is not really conducive to producing a cool, compact, quiet box....
AMD: Speculation, Rumors, and Discussion (Archive)
- Thread starter iMacmatician
- Start date
- Status
Why made at TSMC? Given that the CPU and the GPU are manufactured at GF(For AMD that is), it should be a much easier design effort for AMD to design the console chips at GF as well.
The launch of MS Project Scorpio is next fall. Given that..I would expect it to be Vega based (minus HBM. I also expect the CPU to be Zen based). For the PS4 neo its a bit murkier as it is supposedly more of an evolution of the current chip (Jaguar based CPU).
Well MS says 6 Tflops. If we assume this is GPU only I would expect anything between 44-48 CUs at around 1050-1100 Mhz. Should be paired with 64 ROPs given the 4k gaming target.
Love_In_Rio
Veteran
Vega is TSMC and Zen also it seems.TSMC process is more efficient than GF's (Samsung's really) as we already saw with Apple's A9 and now with Polaris and Pascal.I supposse that's one of the main reassons(appart from HBM2) why Vega is announced to be more efficient than Polaris.
Last edited:
D
Deleted member 2197
Guest
Review missing from @Razor1 list
http://www.hardware.fr/articles/951-1/amd-radeon-rx-480-8-go-14nm-polaris-test.html
Contains some geometry tests:
http://www.hardware.fr/articles/951-8/performances-theoriques-geometrie.html
They also have Witcher 3 (hairworks on/off analysis)
http://www.hardware.fr/articles/951-1/amd-radeon-rx-480-8-go-14nm-polaris-test.html
Contains some geometry tests:
http://www.hardware.fr/articles/951-8/performances-theoriques-geometrie.html
They also have Witcher 3 (hairworks on/off analysis)
That would reverse the ROP to channel ratio on a 256-bit bus. At least GCN has either had a 1:1 relationship or in the case of Tahiti a crossbar that allowed ROPs to cover their dedicated channels and share the additional 128 bits. Possibly, they can't without changing the architecture, and the level of contention under load might make things worse. Another downside might depend on whether AMD's compression units could service twice as many ROP cache misses, or if that would need to double as well.
Maybe Polaris has a close cousin in GCN Neo.
Is there a specific slide or reference to real time operation from AMD? I don't recall the context, or mention of a watchdog timer in the command processor.
That kind of runs against Polaris extending into a decent range of real-time industrial work, where controllers can give cycle, microsecond, or nanosecond time frames.
Orion
Regular
Continued benchmark commentary:
Just saw the following bench
http://www.pcgamer.com/radeon-rx-480-review/
Ashes of the singularity 480 above 980 ti at all resolutions.
If the rumored 26% OC AIB provided 26% performance improvement it would put it within 3% of gtx1080 in this application. Even if performance is less, assuming the 26% OC is possible, it seems reasonable to suggest less than 10% performance difference with gtx 1080 might be possible in this application at the tested settings.
The question is, is this a quirk of ashes, or is it indicative of possible performance in future async heavy dx12 titles?
Just saw the following bench
http://www.pcgamer.com/radeon-rx-480-review/
Ashes of the singularity 480 above 980 ti at all resolutions.
If the rumored 26% OC AIB provided 26% performance improvement it would put it within 3% of gtx1080 in this application. Even if performance is less, assuming the 26% OC is possible, it seems reasonable to suggest less than 10% performance difference with gtx 1080 might be possible in this application at the tested settings.
The question is, is this a quirk of ashes, or is it indicative of possible performance in future async heavy dx12 titles?
http://www.guru3d.com/articles_pages/amd_radeon_r9_rx_480_8gb_review,13.html
different results here.
different results here.
While I dont have information..I am skeptical that Vega is at TSMC. Do you have a source for this? And given the WSA with GF..I am reasonably confident that Zen isnt at TSMC.
I have heard different things about TSMC and Samsung's processes..and from what I know they are very close in performance (with tradeoffs in density and leakage). I dont know if TSMC is significantly more efficient than GF though.
What did we see with A9 on Samsung vs TSMC? Nothing significant from what I remember.
Wafer supply?
Maybe it was also cheaper/simpler with their APU lines already targeting 14nmFF?
Sammy A9 was a tad denser, but had more power consumption.
Anarchist4000
Veteran
The architecture actually seems rather good, the card and process look to be having some difficulties though. Given the wide range of overclocking results in reviews it seems likely the process is sloppy. They may improve over time or with a different fab. They ended up with high voltage and low clocks to guarantee everything works, hence we have a card lottery.
As for Vega, I'm not sure how much more they would be changing. The big change was always HBM2 and that would help perf/watt quite a bit. If Vega is launching in the rumored October, that's not a lot of time to make major changes.
AMD had been throwing around the 2.5x number way before Nvidia said anything about the GTX 1080.
Maybe Polaris was meant to come out earlier when they were still planning on 20nm (cancelled mid-2015)? i.e. Polaris is late, Vega is on-schedule? :s
Anarchist4000
Veteran
Distinct possibility, which would make Vega really interesting.
Apparently these things are selling, Amazon(#1 Best Seller) and Newegg are OOS.
A tad denser yes..but I dont remember reading anything conclusive on the power consumption. There were reports of lower battery life on the Samsung A9 but I dont believe anyone conducted a comprehensive analysis (with a large sample size).
Yep..and then it was "up to" 2.8x when RX480 launched. I think the slides did mention that it was RX470 and 480M (Polaris 11 based) that achieved the 2.8x though. But given what we've seen so far with RX480..I'd be very surprised if Rx470 does achieve it.
- Status
