AMD Polaris Rumors and Discussion

[CarstenS]

Quick note: while GPU-Z reads the chip revision registers correctly, AMD is not using it correctly. When I asked W1zzard about this, he told me that AMD is using the chip revision as another identifier, in place of a longer device ID. So the chips aren't actually C7 or E7.

I remember that being brought up at the local Polaris (10/11) Techday, because it seemed like a ridiculously high stepping number with what P10 launched, yeah.

But unless AMD is outright lying about simple facts, it is a new stepping of P10. And you generally don't break support for old steppings with new BIOSes.

That leaves open, what exactly you could justify as a new stepping without bending the truth so much that a judge could call it a lie. Did they specify what they changed apart from „bugs“?

 

[xEx]

I remember that being brought up at the local Polaris (10/11) Techday, because it seemed like a ridiculously high stepping number with what P10 launched, yeah.


That leaves open, what exactly you could justify as a new stepping without bending the truth so much that a judge could call it a lie. Did they specify what they changed apart from „bugs“?

The box of course. [emoji23]

The simplest way to find out is to test both cards) 470/570, 480/580 at the same frequency (low and high) and see the power consumption and voltage needed.

Looking at the reviews it seems that both will have the same of almost the same characteristics.

Enviado desde mi HTC One mediante Tapatalk

 

[CarstenS]

The simplest way to find out is to test both cards) 470/570, 480/580 at the same frequency (low and high) and see the power consumption and voltage needed.

You would need a significant number of samples to generalize your findings though.

 

[xEx]

You would need a significant number of samples to generalize your findings though.

Yes but if the number are very close in enters between the variation of silicon lottery and we can assume is just that. If there is only 5W difference then then we can assume is the same chip.

 

[entity279]

Yes but if the number are very close in enters between the variation of silicon lottery and we can assume is just that. If there is only 5W difference then then we can assume is the same chip.

Am... no. With 1 sample you can't asume anything

 

[Deleted member 13524]

Nvidia for the most part sells chips to partners who build the boards sold to consumers--hence why various Nvidia chips have sometimes long chains of rebrands as well.

I wasn't talking in general terms, I was talking about this specific rebrand happening in Q2 2017.
We're not seeing the 1000 series - which released at the same time as the RX 400 sereies - getting a rebrand. There's an incremental step on the GTX1080 and 1060 with higher-clocked memory, without a name change.
nvidia knows a full rebrand wouldn't be well perceived by informed consumers, just like AMD knows the 500 series aren't very well perceived either.

My point still stands. OEMs get to drag AMD around because they're in a position of weakness. OEMs would love to get nvidia to rebrand the 1000 series with 150MHz overclocks while sacrificing power efficiency, yet they're not getting any of it.

 

[Deleted member 2197]

AMD Radeon 520 and 530 Released As Well

The Radeon 530 is a third gen GCN product, based on a 28nm fab and get either 320- or 384- shader processors. So that's 5 or 6 CUs with 1 to 2 GB of DDR3/5 graphics memory based on a 64-bit wide memory bus. A product that would do just under 0.8 TFLOPs.
The Radeon 520 (yes the RX denumerator is lacking), would get 320 Shader processors and is actually a 1st gen GCN based product fabbed at 28nm at 0.66 TFLOPs. AMD (530), AMD (520).

http://www.guru3d.com/news-story/amd-radeon-520-and-530-released-as-well.html

 

[Putas]

AMD Radeon 520 and 530 Released As Well

Is there is third gen GCN chip fiting the specification of 530? And I bet 540 will perform better than 550, oh well, OEM wonderland.

 

[Kaotik]

Is there is third gen GCN chip fiting the specification of 530? And I bet 540 will perform better than 550, oh well, OEM wonderland.

Iceland?

 

[ImSpartacus]

Polaris 20X2 is released in the prosumer space as an updated 250W Radeon Pro Duo.

http://www.amd.com/en-us/products/graphics/workstation/radeon-pro/duo#

Looks like two fully enabled Polaris 20 GPUs with 16GB of 7 Gbps GDDR5 each, peaking at 11.45 TFLOPS.

I'm perplexed at why this didn't come in late 2016 in some kind of consumer part to do battle with GP104 in the $500-600 space. Maybe Polaris 10 would've been slightly hotter and needed 300W, but that's still within the PCIe spec limits. Hell, they could've water cooled the stupid thing and priced it at like $700.

Vega is, what, a month or two away? Surely it'll do better than this at the same 250W TDP.

I'm confused.

 

[Deleted member 13524]

Deferred render engines (namely Unreal Engine 4 games) have been blasting away at multi-GPU's scaling capabilities.
Many of the new DX12 implementations are showing spectacular scaling with AFR on multi-GPU systems, but back in 2016 this card wouldn't be well-received.
It probably wouldn't be well received for gaming today either, unless the reviewers made sure to include Deux Ex: Mankind Divided and Rise of the Tomb Raider with their DX12 runtimes (e.g. anandtech only used DX11 in ROTR when reviewing the RX500 series) and Sniper Elite 4.

Regardless, Vega 10 cards will top at 8GB and this card has a whopping 32GB. For VR development this card could be great.

 

[3dilettante]

I wasn't talking in general terms, I was talking about this specific rebrand happening in Q2 2017.
We're not seeing the 1000 series - which released at the same time as the RX 400 sereies - getting a rebrand. There's an incremental step on the GTX1080 and 1060 with higher-clocked memory, without a name change.

Perhaps there won't be a rebrand this year for Nvidia, but I'm waiting to see if there's any credibility to the rumors that newer products are coming out later in the year.

My point still stands. OEMs get to drag AMD around because they're in a position of weakness. OEMs would love to get nvidia to rebrand the 1000 series with 150MHz overclocks while sacrificing power efficiency, yet they're not getting any of it.

If all Nvidia has to offer its partners is a RAM expansion and one spec bump, then perhaps that's what it estimates will get some movement on the product in proportion to AMD's rebrand.
Another interpretation is that AMD is in a position of needing to do more to please its customers, although I think the board partners wouldn't mind if AMD managed to get a more efficient Polaris chip. It's still on the board makers to shoulder the burden of a hotter chip.

Polaris 20X2 is released in the prosumer space as an updated 250W Radeon Pro Duo.

Looks like two fully enabled Polaris 20 GPUs with 16GB of 7 Gbps GDDR5 each, peaking at 11.45 TFLOPS.

I'm perplexed at why this didn't come in late 2016 in some kind of consumer part to do battle with GP104 in the $500-600 space.

It's crossfire on a stick, and that's been losing ground pretty heavily against any single-GPU competition.

Vega is, what, a month or two away? Surely it'll do better than this at the same 250W TDP.

This is marketed as having support for professional applications. That takes time to spin up, which would explain why Polaris just now got this. Vega's going to take a similar amount of time.

 

[Esrever]

32GB is vram is pretty insane to me. Would it be possible to even use this card if my PC only has 8GB of system ram?

 

[Kaotik]

32GB is vram is pretty insane to me. Would it be possible to even use this card if my PC only has 8GB of system ram?

Of course, why not?

 

[Alexko]

It's more like 16GB effectively, anyway, except maybe in HPC workloads.

Sent from my Archos 50 Helium Plus using Tapatalk

 

[Esrever]

Of course, why not?

Doesn't stuff in VRAM have to come from RAM?

 

[3dilettante]

It's more like 16GB effectively, anyway, except maybe in HPC workloads.

In this case, the product description proposes a situation where the GPUs are running separate applications for different stages of the content creation process.

 

[Kaotik]

Doesn't stuff in VRAM have to come from RAM?

Even if it passes through RAM doesn't mean it has to stay there. DX9 and older had to keep duplicates of everything in RAM but that's history since DX10 and I'd assume same applies to OpenGL too

 

[CSI PC]

......

If all Nvidia has to offer its partners is a RAM expansion and one spec bump, then perhaps that's what it estimates will get some movement on the product in proportion to AMD's rebrand.
Another interpretation is that AMD is in a position of needing to do more to please its customers, although I think the board partners wouldn't mind if AMD managed to get a more efficient Polaris chip. It's still on the board makers to shoulder the burden of a hotter chip.
..
This is marketed as having support for professional applications. That takes time to spin up, which would explain why Polaris just now got this. Vega's going to take a similar amount of time.

I would say what Nvidia did with the memory refresh is pretty comparable to AMD's rebrand.
One involved improved memory equal to or more than 10%, while the other involved more stable higher frequencies by around 6%-10% but with no improvements with regards to the performance power envelope (those sites with scopes show this power behaviour increased by the expected trend amount comparable to 480 if ignoring the worst case examples of 480).

I would prefer if this was not a rebrand myself and kept as a refresh update, but maybe there was pressure from OEMs like you say.
Both had originals that when tweaked performed just as well as either the refresh or the rebrand.

Cheers

 

[xEx]

Polaris 20X2 is released in the prosumer space as an updated 250W Radeon Pro Duo.

http://www.amd.com/en-us/products/graphics/workstation/radeon-pro/duo#

Looks like two fully enabled Polaris 20 GPUs with 16GB of 7 Gbps GDDR5 each, peaking at 11.45 TFLOPS.

I'm perplexed at why this didn't come in late 2016 in some kind of consumer part to do battle with GP104 in the $500-600 space. Maybe Polaris 10 would've been slightly hotter and needed 300W, but that's still within the PCIe spec limits. Hell, they could've water cooled the stupid thing and priced it at like $700.

Vega is, what, a month or two away? Surely it'll do better than this at the same 250W TDP.

I'm confused.

Gamer nexus made a review of Multi GPU and most games show no scaling, negative scaling or minor scaling. Multi GPU is still more of a pain than a bless in this time.