AMD Polaris Thread: Radeon RX 480, RX 470 & RX 460 launching June 29th

[Maverick177]

For those few Linux enthusiasts: Phoronix apparently got the card (finally!) and we can expect a launch day Linux review https://www.phoronix.com/forums/for...mdgpu-pro-amdgpu-radeonsi?p=879157#post879157

This did not happen for the past few launches afaik. AMD stepping up it's game. This also presumably means day 1 driver support, possibly even open source one, not just catalyst.

Crimson 2 ?

 

[cen1]

Crimson 2 ?

Probably not. Linux still doesn't have Crimson 1.

 

[littleg]

Funny that sweepr reported the obscure Chinese market price but yet didn't feel the need to post the must more relevant British price on the card that's currently on of the top posts on the AMD subreddits forum. That guy loves to post negative news related to AMD, he hides behind the "just posting the news" smokescreen.

Oi! Some of us happen to order all our stuff from that 'obscure' Chinese site.

Jingdong probably do more business than all the UK sites combined.

edit: just checked, their quarterly revenue is about the same as AMD and Nvidias annual revenue combined.

 

[Glo.]

Pretty cool looking

Stock load power consumption does not exceed 110W, according to the leaker. Still waiting for final drivers(!). The GPU is limited to 1400MHz at 6 pin connector.

That may fall in line with information that 6+8 pin models will not have any problem with exceeding 1600 MHz.

 

[sontin]

Why would the GPU limited to 1400MHz with 150W?! That's just 11% higher than the boost clock...

 

[littleg]

That could actually be an interesting product differentiation. The cheaper AMD reference cards running 1266 and limited to 1400 tops with the AIBs providing more expensive factory OC editions further up. Almost like two products with one name.

Good incentive for AIBs to be jumping all over it aswell knowing they're not being undercut by the vendor.

 

[AtenRa]

That could actually be an interesting product differentiation. The cheaper AMD reference cards running 1266 and limited to 1400 tops with the AIBs providing more expensive factory OC editions further up. Almost like two products with one name.

Good incentive for AIBs to be jumping all over it aswell knowing they're not being undercut by the vendor.

Seems to me the Reference "Founder Edition" cards are $199/249, and those with actual craftsmanship and premium materials from AIBs that are going to OC higher and have better/custom PCB designs and Cooling solutions are going for $300+

 

[Glo.]

Why would the GPU limited to 1400MHz with 150W?! That's just 11% higher than the boost clock...

Board circuitry.

One of explanations.
Second is that its artificial lock.

 

[Abwx]

Why would the GPU limited to 1400MHz with 150W?! That's just 11% higher than the boost clock...

Because 11% overclock done right increase power by 1.11^2 = 1.23x, when i say done right it means that voltage margin is preserved..

 

[RussianSensation]

Pretty cool looking

Oh, this is the first time in a decade that it seems AMD is switching fan suppliers to an NV supplier. Look at it.

HD5870->R9 290X all used this style fan (each fin/ridge is bent in an arc shape)

HD5870

R9 290X

On the NV blower fan, the fin/ridge has a distinct inflection point in the arc (or another way to look at it is it's more like 2 separate lines, not even an arc).

GTX580-> GTX 1080

GTX580

GTX 1080

This should make the RX 480 reference quieter than HD5870-> R9 290X blowers as NV's blower fans are superior to AMD's.

Voltage control enabled from the factory via Radeon Global Settings is the completely opposite approach to NV's anti-enthusiast Kepler voltage lock down that continues to this day. Nice to see that at least the option will be there for those who want to overclock with voltage.

NV PR: Incoming perf/watt comparisons of over-volted 1.5-1.6Ghz overclocked RX480 vs. stock voltage GTX1070 OC, crapping all over P10's "poor" perf/watt in overclocked states, while completely ignoring the price/performance ratio. If RX 480 OC can reach a stock GTX980Ti/Fury X at 1440p, this card will be an instant recommendation for 1080p 60Hz gaming!

 

[maddie]

Because 11% overclock done right increase power by 1.11^2 = 1.23x, when i say done right it means that voltage margin is preserved..

I would think that with all of the new power saving circuitry and software controls added to Polaris, AMD is trying to use the bare minimum voltage needed for a given performance level. In other words, they will have no extra power available at any speed, thus minimizing losses.

Can we assume that any substantial over clock [more than 25-50Mhz] will now need more voltage? There might not even be a standard voltage, only that its more than the bare minimum to switch.

If true, we can roughly calculate the power for 1500 and 1600 Mhz operation.

 

[Mopetar]

Why would the GPU limited to 1400MHz with 150W?! That's just 11% higher than the boost clock...

Artificial product differentiation. Doesn't make as much sense when they don't have a 490 to sell you, but eventually they'll have something and it's harder to charge more money for it if you can just crank up the clock on a lower-tier part.

Supposedly the third party cards will be capable of more, so even though they're still going to be 480's they'll be a better performance class.

 

[3DVagabond]

Its more like
470 satisfactory 1080p performance
480 solid 1080p performance
None of these cards are meant for 1440p. 1440p medium probably but not Ultra.
1080p 60fps is the goal here and that too not at max settings. Even Gtx 1080 cannot do 1080p@60 at max settings in a handful of games.

That would be an extremely small handful. Wee girly hands, maybe?

Don't use extreme examples to try and define the rule. Because there will always be exceptions and trying to define the segment with them is an injustice.

You will find that these games that a 1080 can't run 60fps@1080 will have some stupid, poorly optimized, "kill the performance for no visual improvement" setting designed to get people to over spend on hardware at the sponsoring IHV's request.

 

[3DVagabond]

Pretty cool looking

No heat pipes on the cooler. Looks like it could be Vapor chamber, but cheaper aluminum not copper.

 

[3DVagabond]

Why would the GPU limited to 1400MHz with 150W?! That's just 11% higher than the boost clock...

Is that a rhetorical question? Must be. I can't expect that you think someone here can answer that beyond the obvious.

 

[Abwx]

I would think that with all of the new power saving circuitry and software controls added to Polaris, AMD is trying to use the bare minimum voltage needed for a given performance level. In other words, they will have no extra power available at any speed, thus minimizing losses.

Can we assume that any substantial over clock [more than 25-50Mhz] will now need more voltage? There might not even be a standard voltage, only that its more than the bare minimum to switch.

If true, we can roughly calculate the power for 1500 and 1600 Mhz operation.

The standard voltage setting is to measure the lower voltage at wich the card is stable at 100% on all load and then to increase this value by at least 10% wich is the minimal value for voltage margin.

Seen from a frequency POV the voltage is set such that frequency could be theoricaly increased by 21% assuming the process is still in a linear part of its frequency/voltage curve, OEMs eventualy eat in this margin to sell factory overclocked cards without increasing too much the TDP but at the expense of stability.

 

[Silverforce11]

No heat pipes on the cooler. Looks like it could be Vapor chamber, but cheaper aluminum not copper.

100W do not need vapor chambers, just a simple block heatsink will do just fine. However, they put some thought into the cooling design this time, the separation of the VRAM & Mosfet cooling front plate with the heatsink, means the main heat from the GPU will not saturate the front plate causing vram & mosfets to run hotter.

This was AMD's reference design for the 7970 and 290/X where they bonded the heatsink & plate together and it was a bad idea.

At this time, there's way too many leaks that claims 1.5ghz and even 1.6ghz to be purely false hype. I think with FinFet, they can definitely get 1.5ghz with proper overclocking. It's vital that vcore modding is accessible on launch and best in launch reviews.

 

[maddie]

The standard voltage setting is to measure the lower voltage at wich the card is stable at 100% on all load and then to increase this value by at least 10% wich is the minimal value for voltage margin.

Seen from a frequency POV the voltage is set such that frequency could be theoricaly increased by 21% assuming the process is still in a linear part of its frequency/voltage curve, OEMs eventualy eat in this margin to sell factory overclocked cards without increasing too much the TDP but at the expense of stability.

Just speculating here.

Remember the adaptive clocking technology? I am thinking that we will be seeing a much improved latency version here so the need for a voltage margin is much reduced relative to past designs. Maybe the new command processor is tied into signalling the need for the voltage required based on what's being issued.

From David Kantor's article: http://www.realworldtech.com/steamroller-clocking/

Favorite Processor Paper from ISSCC 2014

My favorite paper from the ISSCC processor session (5.6) describes an adaptive clocking technique implemented in AMD’s 28nm Steamroller core that compensates for power supply noise. Most papers in the processor sessions are overviews that emphasize broad feature sets and the scope and scale of the project. In contrast, paper 5.6 was tightly focused on a specific problem (i.e. power supply noise) and clearly articulated a solution that was implemented in the Steamroller core.

...............................................................................................................................
Future Directions

AMD’s adaptive clocking system in Steamroller is quite attractive, offering a significant improvement in power at a minimal cost in terms of area and a negligible impact on performance. However, there are several potential avenues for improvement.
First of all, the latency of the droop detection and clock stretching could be reduced. Currently, there is a minimum 3 cycle lag before the system can begin to compensate. The droop detector is an asynchronous circuit, which creates a slight delay as the output must be synchronized before it is passed to the clock stretcher. This means that Vmin must have enough guardband to tolerate a few cycles (probably <10) of voltage droop. Reducing the response time of the clock stretching would reduce Vmin even further, resulting in greater power savings. Certain dI/dt events may be predicted in the pipeline. For example, the front-end could signal a hint when decoding 256-bit AVX instructions, indicating that there is likely to be a dI/dt event when those instructions are executed.
Second, this technique could be applied to AMD’s discrete and integrated GPUs, although it is hard to say how big the benefits would be for GPUs. The target clock frequency for a GPU is 1GHz rather than 3GHz and the clock domains are bigger and contain more cores. On the other hand, since GPUs are so parallel dI/dt events may be much bigger (e.g., if all the shaders in a GPU simultaneously begin executing a floating point kernel). Even if the benefits are just half of what is possible in a CPU, a 5-10% decrease in power is significant for a 250W GPU.
Third, since adaptive clocking minimizes the impact of voltage droops AMD could remove package decoupling capacitors or package layers to reduce the cost of the overall platform.
Fourth, the adaptive clocking could be used to improve the transition between different voltage/frequency combinations by reducing the latency.
Summary

Overall, AMD’s adaptive clocking paper was enlightening and enjoyable and stood out from the processor section. While it addresses a longstanding problem, the solution is new and an interesting approach to the challenges in power delivery.
The paper also demonstrated one of AMD’s key differentiators, expertise in power management and clocking, that is critical for any computing platform from mobile to servers. The techniques described will first appear in AMD’s Steamroller based platforms, but are expected to roll out across other IP blocks potentially including GPUs, ARM cores, and the Jaguar core.

 

[looncraz]

No heat pipes on the cooler. Looks like it could be Vapor chamber, but cheaper aluminum not copper.

Looks to have a copper core. Aluminum dissipates heat faster, but copper conducts it faster, so using both together usually yields better performance than just one or the other.

A vapor chamber may be present, but would not be needed to dissipate 150W of heat and maintain <80C with decent airflow and ambient temperatures.

 

[Sweepr]

ShopBLT (USA) - Radeon RX 480 8GB

XFX: $275.49
www.shopblt.com/item/xfx-radeon-rx-480-pcie-8gb/pinete_rx480m8bfa6.html

Gigabyte: $254.76
www.shopblt.com/item/gigabyte-gv-rx480d5-8gd-b-hdmi-3xdp/gigmbd_gvrx480d58gdb.html

 

[Abwx]

No heat pipes on the cooler. Looks like it could be Vapor chamber, but cheaper aluminum not copper.

The screw s holes in the card look to be at a much smaller distance than on the alleged cooler....

About 1000% sure that something doesnt add up here...

 

[Silverforce11]

The screw s holes in the card look to be at a much smaller distance than on the alleged cooler....

About 1000% sure that something doesnt add up here...

Photography angle tricking your mind. Heatsink is thick, has depth, don't forget.

 

[DeeJayBump]

The screw s holes in the card look to be at a much smaller distance than on the alleged cooler....

About 1000% sure that something doesnt add up here...

Looks like an issue of perspective more than a difference in size. The heatsink is taller, and therefore closer to the viewer. The GPU/chip are shorter and appear further away than the heatsink. The shaded/less light upon the chip/GPU area also contributes to the perspective issues, IMO.

Which is to say unless one has both the heatsink and card in hand to personally look over, any apparent difference in size of mounting holes cannot be conclusively determined, IMO.

 

[Abwx]

Just speculating here.

Remember the adaptive clocking technology? I am thinking that we will be seeing a much improved latency version here so the need for a voltage margin is much reduced relative to past designs. Maybe the new command processor is tied into signalling the need for the voltage required based on what's being issued.

From David Kantor's article: http://www.realworldtech.com/steamroller-clocking/

It s highly likely that it s used in Polaris, let say that it will allow the system to keep within 10% voltage margin by reducing frequency at a constant voltage rather than increasing voltage to keep the same frequency...

Better effiency is possible thanks to frequency variation being at least 10 000x faster than a voltage regulation loop, this latter is too slow and will keep the device uselessly overvolted for too long duration after dealing with a short peak.

 

[Abwx]

Looks like an issue of perspective more than a difference in size. The heatsink is taller, and therefore closer to the viewer. The GPU/chip are shorter and appear further away than the heatsink. The shaded/less light upon the chip/GPU area also contributes to the perspective issues, IMO.

Which is to say unless one has both the heatsink and card in hand to personally look over, any apparent difference in size of mounting holes cannot be conclusively determined, IMO.

Also, the fins are perpendicular to the air flow, that makes a lot of incongruities, really, and the perspective is not at play if we take the closer parralel holes...

Anyway thank you for getting out of a long sleep to provide us thoses precisions, really the effort is much appreciated..

Photography angle tricking your mind. Heatsink is thick, has depth, don't forget.

As said the orientation of the fins is not logical at all, if the cooler is rotated 90° to make the fins parralel to the air flow then the holes distance is even more unrealistic.