Discussion: Polaris, AMD's 4th Gen GCN Architecture

[17seconds]

Getting a bit toasty in Crossfire.

http://videocardz.com/61396/new-amd-rx-480-crossfire-benchmarks-hit-the-web-exclusive-first-look-at-new-overclocking-tool

 

[TechyInAZ]

At least those temps are better than a reference R9 390X, that's all that matters. 😀

 

[Rogue Leader]

In its defense the reference cooler isn't the best, I'm sure custom coolers will keep it down a lot more.

@Math Geek I love ASUS for motherboards, but I got turned off to their GPUs when I was buying my most recent ones they seemed to have a lot more folks complaining about DOA GPUs or ones with inherent problems than any other brand.

I'm probably gonna go XFX again, I have had a few and they have never done me wrong.

 

[jimmysmitty]

In its defense the reference cooler isn't the best, I'm sure custom coolers will keep it down a lot more.

@Math Geek I love ASUS for motherboards, but I got turned off to their GPUs when I was buying my most recent ones they seemed to have a lot more folks complaining about DOA GPUs or ones with inherent problems than any other brand.

I'm probably gonna go XFX again, I have had a few and they have never done me wrong.

I have seen that before, when Asus was just coming out with GPUs, but man the Strix series is pretty.

 

[TehPenguin]

I wish EVGA did some fancy coolers for the 480. The EVGA 1080 looks amazing.

Can't wait for 4k benchmarks with 2x480 CF!

 

[TechyInAZ]

I wish EVGA did some fancy coolers for the 480. The EVGA 1080 looks amazing.

Can't wait for 4k benchmarks with 2x480 CF!

EVGA doesn't make AMD cards.

 

[TehPenguin]

That's why I said "I wish" 😀

 

[turkey3_scratch]

Case airflow works more wonders than the card's cooler.

 

[17seconds]

Very interesting read, both the summary link and the original article. I've been having fun getting away from the tech sites and venturing into the finance sites for more perspective on the current round of GPUs.

"AMD Gets an 'F' on Pricing"
http://www.hardocp.com/news/2016/06/24/amd_gets_f_on_pricing#.V22b7Z9lDqA
http://www.holdenadvisors.com/holden-resource-center/holden-pricing-newsletter/price-discounting/

 

[turkey3_scratch]

Very interesting read, both the summary link and the original article. I've been having fun getting away from the tech sites and venturing into the finance sites for more perspective on the current round of GPUs.

"AMD Gets an 'F' on Pricing"
http://www.hardocp.com/news/2016/06/24/amd_gets_f_on_pricing#.V22b7Z9lDqA
http://www.holdenadvisors.com/holden-resource-center/holden-pricing-newsletter/price-discounting/

I agree with HardOCP's last two there. However they say:

The real winners of the price war will be the PC makers and ‘maybe’ the consumer

So a consumer can't be a PC maker? Am I reading that properly? And then:

There are limited PC makers in the market who make VR PCs and these PC makers will open the markets by deciding on the value they deliver to the end customer and their end user pricing not by the reduction in the price of the chip

So what is this distinction between PC makers and end users/customers/consumers they are trying to get at here? Also, Adele McLean is just one woman with an opinion vs AMD's entire financial team making these decisions.

 

[Mousemonkey]

Very interesting read, both the summary link and the original article. I've been having fun getting away from the tech sites and venturing into the finance sites for more perspective on the current round of GPUs.

"AMD Gets an 'F' on Pricing"
http://www.hardocp.com/news/2016/06/24/amd_gets_f_on_pricing#.V22b7Z9lDqA
http://www.holdenadvisors.com/holden-resource-center/holden-pricing-newsletter/price-discounting/

I agree with HardOCP's last two there. However they say:

The real winners of the price war will be the PC makers and ‘maybe’ the consumer

So a consumer can't be a PC maker? Am I reading that properly? And then:

There are limited PC makers in the market who make VR PCs and these PC makers will open the markets by deciding on the value they deliver to the end customer and their end user pricing not by the reduction in the price of the chip

So what is this distinction between PC makers and end users/customers/consumers they are trying to get at here? Also, Adele McLean is just one man with an opinion vs AMD's entire financial team making these decisions.

More likely a woman, you might want to do a name check.

 

[cdrkf]

Very interesting read, both the summary link and the original article. I've been having fun getting away from the tech sites and venturing into the finance sites for more perspective on the current round of GPUs.

"AMD Gets an 'F' on Pricing"
http://www.hardocp.com/news/2016/06/24/amd_gets_f_on_pricing#.V22b7Z9lDqA
http://www.holdenadvisors.com/holden-resource-center/holden-pricing-newsletter/price-discounting/

This is quite amusing- at first the price of Polaris sounds shocking- but then lets look at what AMD already sells? Well Tonga can be had for $200 right now (and has been available for that price for quite some time), that uses similar memory tech and offers a similar specification to the cut down Polaris. The thing is though, Tonga (due to it being 28nm) is MUCH bigger (think I heard 40%?). So AMD are selling a 40% smaller GPU for the same cash- ergo they are making MUCH better margins 😛

What is making Polaris such good value isn't in fact AMD pricing it really cheaply, rather nVidia has really inflated the price of Pascal to the point AMD can drop this fairly modest gpu exactly where it's supposed to be, and come out looking like shinning white nights in the process (thanks Nvidia!). I really don't think it's under priced.

 

[turkey3_scratch]

Very interesting read, both the summary link and the original article. I've been having fun getting away from the tech sites and venturing into the finance sites for more perspective on the current round of GPUs.

"AMD Gets an 'F' on Pricing"
http://www.hardocp.com/news/2016/06/24/amd_gets_f_on_pricing#.V22b7Z9lDqA
http://www.holdenadvisors.com/holden-resource-center/holden-pricing-newsletter/price-discounting/

This is quite amusing- at first the price of Polaris sounds shocking- but then lets look at what AMD already sells? Well Tonga can be had for $200 right now (and has been available for that price for quite some time), that uses similar memory tech and offers a similar specification to the cut down Polaris. The thing is though, Tonga (due to it being 28nm) is MUCH bigger (think I heard 40%?). So AMD are selling a 40% smaller GPU for the same cash- ergo they are making MUCH better margins 😛

What is making Polaris such good value isn't in fact AMD pricing it really cheaply, rather nVidia has really inflated the price of Pascal to the point AMD can drop this fairly modest gpu exactly where it's supposed to be, and come out looking like shinning white nights in the process (thanks Nvidia!). I really don't think it's under priced.

Just because the node is smaller doesn't mean the chip is smaller, though.

 

[cdrkf]

Very interesting read, both the summary link and the original article. I've been having fun getting away from the tech sites and venturing into the finance sites for more perspective on the current round of GPUs.

"AMD Gets an 'F' on Pricing"
http://www.hardocp.com/news/2016/06/24/amd_gets_f_on_pricing#.V22b7Z9lDqA
http://www.holdenadvisors.com/holden-resource-center/holden-pricing-newsletter/price-discounting/

This is quite amusing- at first the price of Polaris sounds shocking- but then lets look at what AMD already sells? Well Tonga can be had for $200 right now (and has been available for that price for quite some time), that uses similar memory tech and offers a similar specification to the cut down Polaris. The thing is though, Tonga (due to it being 28nm) is MUCH bigger (think I heard 40%?). So AMD are selling a 40% smaller GPU for the same cash- ergo they are making MUCH better margins 😛

What is making Polaris such good value isn't in fact AMD pricing it really cheaply, rather nVidia has really inflated the price of Pascal to the point AMD can drop this fairly modest gpu exactly where it's supposed to be, and come out looking like shinning white nights in the process (thanks Nvidia!). I really don't think it's under priced.

Just because the node is smaller doesn't mean the chip is smaller, though.

It's already been leaked that Polaris 10 (the bigger one) is about 200mm squared. That happens to be the same size as Pitcarin when it first came out (HD 7870 / R9 270X / R9 370, that was a LONG lived gpu).

Tonga (AMD's current GPU in this price segment) is actually similar in spec / size to the original high end Tahiti at 359mm squared (similar in size to the 1080 also- makes you think twice about nVidias pricing doesn't it?).

A 200mm gpu at $200 isn't stupidly cheap.... It's *cheaper* than the launch price of Pitcairn in 2012, but then Pitcairn was faster than the then fastest gpu the GTX 580 at the time, and once nVidia got the 600 series out those prices dropped quickly.

Don't get me wrong, I'm NOT saying Polaris is a bad deal, not at all. That said, AMD are being a little more cunning than people have given them credit for- they've got analysts ATTACKING the pricing because it's so low, yet have also managed to significantly improve their margins on that gpu segment in the process 😛 I actually think AMD might be onto a winner here 😉

 

[turkey3_scratch]

Ahh okay that's a lot smaller. I wonder how hot the chips will run, as we can see with Pascal the smaller die makes it run hotter since it has less surface area to expel heat through contact with the heat sink.

 

[cdrkf]

Ahh okay that's a lot smaller. I wonder how hot the chips will run, as we can see with Pascal the smaller die makes it run hotter since it has less surface area to expel heat through contact with the heat sink.

Well what we've seen so far doesn't look too bad. Worst I've seen is the crossfire leak above with one card (in a dual card setup) topping out at 87. Fairly high but given AMD designs their gpus with a 94 degree max temp not out of the ordinary- and that is with their (rather useless) stock blower fan. Aftermarket cards will be better.

I think the reason for temps being good though is that it's (supposed to be) pretty power efficient (RX 480 estimated at 100W under normal gaming loads). Much lower than the current 28nm stuff. It will be nice to get the official reviews next week though as all of this is still not certian.

 

[Sakkura]

Probably. Even to this day AMD is still behind nvidia with frame times. Not sure if it is a driver issue or possibly an issue with the uArch (GCN) and how it works because while it has gotten better it still is worse, normally at launch it is the worst, than I think it should be for a uArch that has been around for quite a while.

Doesn't really seem to be the case if you look at benchmarks. The recent GTX 1080 and 1070 reviews saw AMD and Nvidia trade blows on frame time variance for their comparable cards. AMD wins Ashes (duh), BF4, Hitman, and Witcher 3, Nvidia wins GTA V, Project Cars, and The Division; Rise of the Tomb Raider is more or less a draw.

Which site are you getting your benchmarks from? I checked the TH review and most of the results have the 1080 in line for the most part with the Fury X.

The TH reviews of the 1080 and 1070. The AMD cards were trading blows with their Nvidia equivalents.

You can't really compare the 1080 since it's just more powerful than anything AMD has available (aside from a silly dual-GPU card). I was just using the 1080 and 1070 reviews since they're recent and include the high-end cards from AMDs 300 series and Nvidia's 900 series.

 

[Sakkura]

Ahh okay that's a lot smaller. I wonder how hot the chips will run, as we can see with Pascal the smaller die makes it run hotter since it has less surface area to expel heat through contact with the heat sink.

Part of it's Nvidia betting on clock speeds for performance. So far it appears AMD are going for efficiency and thus not pumping clocks up as high, which could help thermals. Speculation is speculation though.

 

[jimmysmitty]

Ahh okay that's a lot smaller. I wonder how hot the chips will run, as we can see with Pascal the smaller die makes it run hotter since it has less surface area to expel heat through contact with the heat sink.

Well what we've seen so far doesn't look too bad. Worst I've seen is the crossfire leak above with one card (in a dual card setup) topping out at 87. Fairly high but given AMD designs their gpus with a 94 degree max temp not out of the ordinary- and that is with their (rather useless) stock blower fan. Aftermarket cards will be better.

I think the reason for temps being good though is that it's (supposed to be) pretty power efficient (RX 480 estimated at 100W under normal gaming loads). Much lower than the current 28nm stuff. It will be nice to get the official reviews next week though as all of this is still not certian.

Do you have any evidence to support that Polaris will also sport a 94c temp ceiling? Because Tahiti, the HD7970, was announced by AMD to sport a max temp of 85c and it was also 28nm.

One thing about smaller nodes, depending on a lot of factors they might not be able to handle 94c as well as 28nm.
I fully expect Asus, Sapphire, MSI and the others to have even their more basic designs keeping these GPUs much happier. The question would be how well can GloFlos (Samsungs) 14nmLPP handle higher temps.

Probably. Even to this day AMD is still behind nvidia with frame times. Not sure if it is a driver issue or possibly an issue with the uArch (GCN) and how it works because while it has gotten better it still is worse, normally at launch it is the worst, than I think it should be for a uArch that has been around for quite a while.

Doesn't really seem to be the case if you look at benchmarks. The recent GTX 1080 and 1070 reviews saw AMD and Nvidia trade blows on frame time variance for their comparable cards. AMD wins Ashes (duh), BF4, Hitman, and Witcher 3, Nvidia wins GTA V, Project Cars, and The Division; Rise of the Tomb Raider is more or less a draw.

Which site are you getting your benchmarks from? I checked the TH review and most of the results have the 1080 in line for the most part with the Fury X.

The TH reviews of the 1080 and 1070. The AMD cards were trading blows with their Nvidia equivalents.

You can't really compare the 1080 since it's just more powerful than anything AMD has available (aside from a silly dual-GPU card). I was just using the 1080 and 1070 reviews since they're recent and include the high-end cards from AMDs 300 series and Nvidia's 900 series.

 

[Sakkura]

Do you have any evidence to support that Polaris will also sport a 94c temp ceiling? Because Tahiti, the HD7970, was announced by AMD to sport a max temp of 85c and it was also 28nm.

One thing about smaller nodes, depending on a lot of factors they might not be able to handle 94c as well as 28nm.
I fully expect Asus, Sapphire, MSI and the others to have even their more basic designs keeping these GPUs much happier. The question would be how well can GloFlos (Samsungs) 14nmLPP handle higher temps.

Let's see if I can get this quote right...

Hawaii, as we all know, was officially "designed to" run at 95C, so I don't really think there's been a major trend towards GPUs being less heat-tolerant up through 28nm.

But it's going to be very interesting to see how the Samsung/GloFo 14LPP turns out in GPUs. First major GPU in a very long time not based on TSMC.

 

[-Fran-]

This is quite amusing- at first the price of Polaris sounds shocking- but then lets look at what AMD already sells? Well Tonga can be had for $200 right now (and has been available for that price for quite some time), that uses similar memory tech and offers a similar specification to the cut down Polaris. The thing is though, Tonga (due to it being 28nm) is MUCH bigger (think I heard 40%?). So AMD are selling a 40% smaller GPU for the same cash- ergo they are making MUCH better margins 😛

What is making Polaris such good value isn't in fact AMD pricing it really cheaply, rather nVidia has really inflated the price of Pascal to the point AMD can drop this fairly modest gpu exactly where it's supposed to be, and come out looking like shinning white nights in the process (thanks Nvidia!). I really don't think it's under priced.

The same thing that I thought when reading the article...

AMD is just not selling "at cost" here. The GPU is cheap *compared* to what nVidia has at those price points. Plus, we don't have any idea how much money AMD is paying Samsung/GloFo (I haven't read anything there TBH) per GPU. And the die area is totally in-line with what they have in the mid segment year after year.

Plus, that group of writers is very vocal about "sell high, don't do discounts"; like an aggressive group of sales people. They give me the shivers, haha.

They do a decent market analysis though.

Cheers!

 

[jtk2515]

Have any of you seen the Benchmarks in doom and witcher 3 for the 480? What do you guys think about the Performance vs the 970? It's very hard for me to get a real FPS advantage number but it looks like 4-6% to me. What are your thoughts?

I'm thinking like 75fps avg in doom, 103 in overwatch and like 40-65 in witcher

ohh btw the videos are back up
http://wccftech.com/amd-rx-480-gaming-benchmarks-leaked/

 

[turkey3_scratch]

Benchmarks won't be released until the 28th.

 

[jtk2515]

I'm looking at the the WCCFTECH ones
http://wccftech.com/amd-rx-480-gaming-benchmarks-leaked/

 

[Martell1977]

I'm looking at the the WCCFTECH ones
http://wccftech.com/amd-rx-480-gaming-benchmarks-leaked/

You have to take into account that official drivers for the 480 haven't been released yet (as far as I've seen), so the performance will likely change. Plus, sometimes "leaked" benchmarks like this are peoples guesstimates as opposed to real numbers.

Until the launch...everything is suspect, even AMD's own demonstration benchmarks.