AMD: Speculation, Rumors, and Discussion (Archive)

[CSI PC]

So looks like the Asus GPU Tweak utility allows for 480 cards to go above the 1.15V default limitation.
KitGuru using that utility with the Asus 480 managed to reach 1.23V at 1400MHz, using Wattman they could only reach 1380MHz and I wonder if that meant the 1.15V limit.

Downside, notice the fan speed at 87%, maybe could be a bit lower as they are not showing temps, but most reports mentioning manual OC all talk about needing high fan RPM.
This manual OC put it above the Asus Stix 980, but then that too could be manual OC and would be back on top; caveat is the comparison was only with 3DMark.

Shame they did not show the core voltage when trying to use Wattman, but good news is the GPU Tweak utility does allow for above 1.15V, although not sure if it works universally for all 480 manufacturer/models (needs to be balanced against the possibility some AIB partners implemented cheaper PWM controller and VRMs in some of their more basic models).

http://www.kitguru.net/components/g...us-rx-480-strix-gaming-oc-aura-rgb-8192mb/30/
Maybe some with other 480s will try the Asus GPU Tweak utility.
Cheers

 

[xEx]

But that much more volts for 20MHz...I dont see it worth it.

 

[sir doris]

But that much more volts for 20MHz...I dont see it worth it.

Would that increase in voltage allow the card to stay closer to or at 1400MHz throughout demanding sceens as opposed to dropping back to the 1200MHz range? Might make more difference than the 20MHz target frequency suggests?

 

[Ryan Smith]

Thanks for the link..so I guess AT just got it wrong then. Also I highly doubt that the WX5100 is sub 75W. Its probably in the range of 110-120W.

AMD emailed me in the afternoon to let me know that WX 7100 was 36 CUs and not 32. In my notes I had 32, and VideoCardz had a screencap of the presentation also showing 32. So it looks like AMD messed up the slide. (I don't have a full copy of the deck yet, it hasn't been released)

 

[Kaotik]

AMD emailed me in the afternoon to let me know that WX 7100 was 36 CUs and not 32. In my notes I had 32, and VideoCardz had a screencap of the presentation also showing 32. So it looks like AMD messed up the slide. (I don't have a full copy of the deck yet, it hasn't been released)

Yeah, it was 32 on the presentation, another screencap on my machine confirming that

 

[Ext3h]

Would that increase in voltage allow the card to stay closer to or at 1400MHz throughout demanding sceens as opposed to dropping back to the 1200MHz range? Might make more difference than the 20MHz target frequency suggests?

That is not how it works. The card clocks down during demanding scenes to stay within the power target. It could stay at max clock if the cooling was sufficient and you increased / removed the power target in the firmware. The additional voltage doesn't help with that - on the contrary, the increased voltage means you hit the power target even faster.

 

[Erinyes]

AMD emailed me in the afternoon to let me know that WX 7100 was 36 CUs and not 32. In my notes I had 32, and VideoCardz had a screencap of the presentation also showing 32. So it looks like AMD messed up the slide. (I don't have a full copy of the deck yet, it hasn't been released)

Apologies then..it was AMD who messed up. You would think they'd have someone proofread those slides before a major event. Though it might have been better if you posted an edit or update in the article.

Could you also confirm with them if WX5100 has 28 or 32 CUs?

 

[CSI PC]

But that much more volts for 20MHz...I dont see it worth it.

Technically it is 90MHz overclock not 20MHz as the Strix default is 1310MHz,

The main point though is that you may not need custom BIOS, on the Asus their utility breaks the 1.15V.
To put it into context, the only way to do that to date on Nvidia Pascal is with the crude Asus XOC bios (HoF bios has not been shown too much in public), which does not work well for any other cards and sort of breaks Boost3.

It is a different discussion regarding voltage scaling, which I have been pretty critical about.
But we do not know if Kitguru has a poor/good/run of the mill quality GPU in terms of OC, nor unfortunately do we know if Wattman was locked to 1.15V.

And yes, as I showed in the Nvidia thread, the Pascal do seem to have voltage scaling advantage, but very difficult to break the 1.1V limit compared to this Asus with GPU Tweak.
While the voltage scaling may seem to be poor with Polaris, bear in mind it seems to be more comfortable at the extreme limits around and beyond 1.35V than Pascal, but it is debatable how close you can get to that on water without causing failure long term; this generation for AMD/Nvidia is definitely not like the 28nm in that regard of abuse and the big voltage numbers.

.
Cheers

 

[xEx]

Technically it is 90MHz overclock not 20MHz as the Strix default is 1310MHz,

The main point though is that you may not need custom BIOS, on the Asus their utility breaks the 1.15V.
To put it into context, the only way to do that to date on Nvidia Pascal is with the crude Asus XOC bios (HoF bios has not been shown too much in public), which does not work well for any other cards and sort of breaks Boost3.

It is a different discussion regarding voltage scaling, which I have been pretty critical about.
But we do not know if Kitguru has a poor/good/run of the mill quality GPU in terms of OC, nor unfortunately do we know if Wattman was locked to 1.15V.

And yes, as I showed in the Nvidia thread, the Pascal do seem to have voltage scaling advantage, but very difficult to break the 1.1V limit compared to this Asus with GPU Tweak.
While the voltage scaling may seem to be poor with Polaris, bear in mind it seems to be more comfortable at the extreme limits around and beyond 1.35V than Pascal, but it is debatable how close you can get to that on water without causing failure long term; this generation for AMD/Nvidia is definitely not like the 28nm in that regard of abuse and the big voltage numbers.

.
Cheers

Yes well its the first 14/16 nm so its expected to have some flaws although they have been developing for this process for 5 years? so it is kind of disappointing but I think we are so close to the limit that it is "normal" at this point.

And well maybe the degradation is really high when you increase the voltage and thats why they are so cautious about it but in any case I think the noise/temp and general degradation of the silicon you get for increasing the voltage that much is not worth the 20MHz more that will not be noticeable at all...like you gain what? 2FPS at best case?

 

[xEx]

480 stock issues:

 

[mczak]

Good catch on the PCIE link. 8X for the Polaris 11 shouldn't cause any noticeable performance loss and if it saves a bit of power..its a good move (especially for laptops)

You could, of course, always make a 16x interface and switch off half of the lanes in the mobile versions. I agree though it shouldn't be a big deal.

As a side note, I hadn't noticed before that Polaris 11 only has an 8 lane PCI-Express interface. I doubt there is a noticeable performance loss considering its performance class, and yet I don't recall any other modern desktop-bound GPUs which have done so.

Mars/Oland. Granted it's probably more commonly used in notebooks, but there's a couple cards with it on desktop too. (Though of course with its 6 CUs it's definitely in a lower performance class compared to Polaris 11.)

 

[Ryan Smith]

Apologies then..it was AMD who messed up. You would think they'd have someone proofread those slides before a major event. Though it might have been better if you posted an edit or update in the article.

Could you also confirm with them if WX5100 has 28 or 32 CUs?

Only WX 7100 was corrected. So WX 5100 appears correct.

 

[Erinyes]

Only WX 7100 was corrected. So WX 5100 appears correct.

Thanks Ryan. The clock still appears to be on the higher side for a workstation part.

PS: Still waiting on that 1060 review

 

[Michellstar]

480 stock issues:

This distribution channel it´s more appropriate for stolen goods and/or drug dealers...

Where does it come from??

 

[xEx]

This distribution channel it´s more appropriate for stolen goods and/or drug dealers...

Where does it come from??

Miners

Enviado desde mi HTC One mediante Tapatalk

 

[Deleted member 13524]

So Ethereum miners are replacing Hawaiis with Polaris 10 cards?

The difference in power consumption should give a sizeable profit boost, even if they're probably mining a bit less now.

 

[xEx]

Yes. Also cheaper cards (probly could sells the Hawaii buy Polaris and still gets some money)

Enviado desde mi HTC One mediante Tapatalk

 

[Alexko]

The power efficiency would likely be better on the RX 470. But maybe absolute performance matters, or they underclock their cards.

 

[CarstenS]

So Ethereum miners are replacing Hawaiis with Polaris 10 cards?

The difference in power consumption should give a sizeable profit boost, even if they're probably mining a bit less now.

Weren't the more serious ones already using R9 Nano's by the hundreds? Maybe that's why the prices are not budging... hm...

 

[Deleted member 2197]

Sapphire RX 480 Nitro + OC 4GB & 8GB Review

http://www.kitguru.net/components/graphic-cards/zardon/sapphire-rx-480-nitro-oc-4gb-8gb-review/