AMD - Piledriver chitchat

[scaryjim]

... It's unlike AMD to implement arbitrary feature lock-outs.

It used to be, but they've been using product features to differentiate lines for a while now - like Kaveri only running dual graphics with R7 240 and R7 250 graphics cards when there's no technical reason it shouldn't run with the 7750/7770 too. I suspect there might have been a few annoyed purchasers of Sapphire's R7 250 low profile and ultimate cards, which were atually 7750s rebadged as R7 250s.

The hexus article about the 300 series cards speculates that they've locked it down to differentiate product lines. If you're looking for a technical "reason", though, it's possible that they've introduced new microcode with the respins that means it's muich easier and more efficient to implement the feature on the new cards. That's a bit of a stretch though, tbh....

[watercooled]

As quoted by CAT, they have apparently changed the microcode WRT the power subsystem, not sure if that would have anything to do with it; perhaps new power granularity works with the FPS lock to extract more power savings than would have previously been possible?

I'm clutching at straws OFC, and like I say I've used software like Rivatuner to limit FPS in certain games for ages now, and it's pretty much a necessity in some weird ports like Dead Space 1, so I don't see why they wouldn't at least allow the FPS control anyway unless there's some additional functionality vs 3rd party software?

WRT dual graphics, 240/250 use a different die (Oland) to 77xx (Cape Verde) and different shader counts. Though having said that, it's weird Kaveri wouldn't work with the 512 shader 7750?

[scaryjim]

... I'm clutching at straws OFC, and like I say I've used software like Rivatuner to limit FPS in certain games for ages now ...

It's possible that the frame rate limiting you were doing there was simply dropping unnecessary rendered frames and didn't provide the power and heat benefits that AMD's implementation does. I doubt rivatuner is capable of changing a card's clockspeed and voltage planes on the fly to any meaningful degree, and certainly not responsively enough to ensure smooth performance....

... WRT dual graphics, 240/250 use a different die (Oland) to 77xx (Cape Verde) and different shader counts. Though having said that, it's weird Kaveri wouldn't work with the 512 shader 7750?

Yeah, it's beyond weird, it's a deliberate lockdown. The VLIW4 Richland APUs could dual graphics with a 7750, and I was pretty excited when it turned out that Kaveri did have 512 shaders in its IGP - I owned a low profile 7750 at the time and the possibility of a tiny little low profile powerhouse based on kaveri and the 7750 was tantalising. Then AMD locked Kaveri down to the Oland cards, and boom, excitement gone.

[watercooled]

Rivatuner did lower GPU utilisation and hence power use/fan speed/coil whine. But yeah that's my point - AMD's solution may be better integrated with the power management.

[watercooled]

Some talk about Fiji in this new TR video: https://www.youtube.com/watch?v=uz-2lO86xfs

They mention Fiji is right up against the reticle size for the interposer, within a few mm.

[CAT-THE-FIFTH]

PCPER podcast on Fury X:

https://www.youtube.com/watch?v=wLiik1-jEc4#t=1703



They go into some of the power saving tech which was used outside HBM.

Heck,they even comment the improvements in performance/watt were close enough to going from Fermi to Kepler.

[CAT-THE-FIFTH]

Looks like Intel will still be using 14NM next year and 10NM is delayed until 2017:

http://www.extremetech.com/computing...-14nm-products

[The Hand]

Looks like Intel will still be using 14NM next year and 10NM is delayed until 2017:

http://www.extremetech.com/computing...-14nm-products

If that's the case, it might be a mistake with Zen coming. All AMD have to do is adjust Zen's price versus the Intel range and AMD will have a foothold again.. Unless Intel are betting that Zen is going to fall on its face or get delayed?

[DanceswithUnix]

Unless Intel are betting that Zen is going to fall on its face or get delayed?

I don't ever see them doing that. Intel just isn't a sit back and wait culture if there is any sort of threat.

I also find this to be one of the hard to believe leaks. I saw it a couple of days ago on Fudzilla linked to http://news.softpedia.com/news/ntel-...e-485110.shtml as a source.

Other foundries are very bullish about 10nm, I can't see why Intel would be stumbling.

[CAT-THE-FIFTH]

The Anandtech review is finally up:

http://www.anandtech.com/show/9390/t...ury-x-review/9

It appears Fury Nano is a full Fiji chip!!

Unlike the R9 Fury, AMD has announced the bulk of the specs for the R9 Nano. This card will feature a fully enabled Fiji GPU, and given AMD’s goals I suspect this is where we’re going to see the lowest leakage bins end up. What separates R9 Nano from R9 Fury X is the power target, and as a result the expected sustained clockspeeds and performance. The R9 Nano will be a 175W card, 100W less than the R9 Fury X, and even with heavy binning it’s a safe bet that it will not be able to hit/sustain R9 Fury X’s 1050MHz clockspeed. However with that said, because clockspeeds, voltages, and power consumption have a non-linear effect, at this point in time it is reasonable to assume that AMD is going to be able to hit and sustain relatively high clockspeeds even at 175W just by backing off on load voltage. AMD is not giving us any expectations for clockspeeds at this time, though on a personal note based on the kind of clockspeed scaling we see on other 28nm GPUs, I would be surprised if a 175W Fiji could not sustain 800MHz or better in games at 175W, assuming the cooler is capable of dissipating that much heat.

[scaryjim]

It appears Fury Nano is a full Fiji chip!!

Good news if it is. Based on my previous calcs for performance relative to Fury X (should be ~ 85%) a clock speed around 850MHz would be about right. Would be interesting to see someone drop the power limit on a Fury X to 175W and see what clock speeds it could sustain....

As to the dual chip card, I know the article says they expect a 500W card, but if they went for a 375W power target (keeping within the PCIe limit on connectors if not the actual specs overall) they should be able to make it perform like two nanos in crossfire. Perfect scaling would make that 70% faster than a Fury X, so around 50% faster should be realistic for a lot of games. That's ... well, very tasty.

[watercooled]

Something else interesting was the overclocked power consumption - it was only marginally higher than stock. I understand there's no overvolting yet, but even so I wonder if the clocks are a bit conservative?

Edit: Oh and another part is on this page: http://anandtech.com/show/9390/the-a...ury-x-review/5

HBM has apparently saved around 20-30W, and I think that number is the whole-system savings including controllers, not just the memory itself, so it should be quite clear that memory is far from the only area AMD have improved efficiency.

[CAT-THE-FIFTH]

I have a feeling that the reason we won't see a fully enabled Tonga on desktop is because it probably would get close to an R9 290/R9 390 at 1080P.

[watercooled]

Hmm maybe you're right - the direct competition for the 380 is the 960 which it comfortably beats in gaming performance. The next one up is the 970 which is what Hawaii competes with so they both have that product gap.

Maybe if Nvidia released something then there would be a reason to compete, but unless they release a further cut-down GM204 that isn't likely as the 960 is already fully enabled AFAIK.

[watercooled]

Some threads have popped up at OCUK/overclock about possible differences in image quality of different graphics cards. This is something we saw in reviews years back as there were often differences, but more recent rendering techniques pretty much removed the differences.

http://www.overclock.net/t/1563118/o...d-kepler-cards
http://forums.overclockers.co.uk/sho...php?t=18679713

I've not read the whole threads yet so haven't drawn any conclusions but there's no harm in checking it. Looking at the BF4 screenshots (post #8 overclock.net), to me it looks like anisotropic filtering is a bit broken on the TX screenshot as the close-up stuff seems OK?

Edit: I've read this far: http://forums.overclockers.co.uk/sho...8679713&page=7
Looks like it could be wonky default settings in the NV control panel lowering IQ while increasing FPS. Not cool if true, because those default screenshots look like crap on the TX.

[kalniel]

Almost certainly a non-event - there are often differences in games based on the optimisations the drivers make - if you have surface optimisation and performance on in CCC then it varies quite a bit whether you notice it.

One of the things I like about AMD cards is the ability to set nice filtering - the negative LOD when applying adaptive AA is really nice.