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

[psolord]

I think now that the wccf benchmarks have been debunked, and it looks like rx470 isn't way slower than rx480, it seems like the real question becomes (at least for me) why not buy two rx470's, overclock them to rx480'ish stock, and have a 300 USD solution with gtx1080ish perf (say up to 10% worse since the aots bench was probably cherry picked) except only 4gb vram?

I entertained this idea as well.

RX470 CFX could come very close to 1080, with half the money. Still, there are big trade offs to consider.

Not all games work well with multi gpu and many of them do not work at all. Unreal Engine 4, which is a big player in game engines, does not like multi gpu one bit. The CPU has a lot more overhead with the presence of two cards. To add insult to injury, if AMD has not done a great deal to improve their DX11 performance, a user with say, a Sandy Bridge, could find his cards underutilized and under performing. Not a nice scenario.

Even the AoTS benchmark that has been leaked, showed the gpus not at full load. Maybe they will fix it later, dunno.

I think that a user like me, with a non hyper threaded sandy, maybe should wait for Vega 11 or something.

But let's see how big Polaris performs. We may be surprised. At least for 1080p which most users care.

 

[Newbian]

With Hitman I can agree, but not with Ashes.

I can only assume this is a joke as aots is probably the most known modern game for this issue right now.

 

[Glo.]

I can only assume this is a joke as aots is probably the most known modern game for this issue right now.

What I meant is that games are optimized for specific brand. Ashes is not optimized for specific brand, Hitman - well, we do not know that for sure, but it can very well be optimized for AMD, more than for Nvidia.

 

[zinfamous]

Glo. you picked 2 of the few games AMD cards are known to perform much better at than in the majority of games... Biased.

Isn't that generally known to be a DX12 thing, though? So if that is the case, then are the Pascal chips for the next 3-4+ years only allowed to be benched against years-old DX11 games because there is no "bias" there, and AMD can only fairly be benched against those same old games so as not to show..."bias"?

I could be wrong, but it seems it isn't so much a game bias here with AMD but a DX12 bias that has been known for some time now. If Pascal really is a failure with DX12, then what is this going to mean for benchmarks for top games for the rest of this year and going forward?

how does one define bias appropriately?

 

[Newbian]

Isn't that generally known to be a DX12 thing, though? So if that is the case, then are the Pascal chips for the next 3-4+ years only allowed to be benched against years-old DX11 games because there is no "bias" there, and AMD can only fairly be benched against those same old games so as not to show..."bias"?

I could be wrong, but it seems it isn't so much a game bias here with AMD but a DX12 bias that has been known for some time now. If Pascal really is a failure with DX12, then what is this going to mean for benchmarks for top games for the rest of this year and going forward?

how does one define bias appropriately?

What about the new tomb raider game clearly being on the side of nvidia?

The problem is Async utilization and aots has that issue clearly and games with it will favor amd for awhile it seems.

 

[DeeJayBump]

how does one define bias appropriately?

For this "crew" in general, bias is defined as:

1. NOT doing open-air testing without warm-up to hide throttling of nvidia cards ala TPU...

2. NOT using overclocked nVidia cards compared to stock-clocked AMD offerings to put nVidia cards in the best (manipulated) light..

3. If DX12 games and non-GameWorks games/tests are predominantly used in the testing...

THEN there's Bias.

 

[Shivansps]

AoTS is not even a good RTS, Stellaris is waaaaaay better, and its done on DX11, and it does not have render problems, it has IA programming issues, very CPU heavy, common on RTS.
AoTS is little else than a tech demo for Async shaders, if not for Async Shaders no one whould know what that is, and even less use it on benchmarks.

 

[Bacon1]

Why do amd need that reference stuff anyway?

Look how successfull 390 launch was vs practicall all other stuff and thats even with something more rebranded than properly facelifted. I dont understand why they dont keep their dirty engineering hands off and let the aib do the work. Might even help on kyles temper

390(x) just reused the same coolers the companies made for 290(x).

 

[Glo.]

For this "crew" in general, bias is defined as:

1. NOT doing open-air testing without warm-up to hide throttling of nvidia cards ala TPU...

2. NOT using overclocked nVidia cards compared to stock-clocked AMD offerings to put nVidia cards in the best (manipulated) light..

3. If DX12 games and non-GameWorks games/tests are predominantly used in the testing...

THEN there's Bias.

TPU also benches Hitman in DX11, but also benches Rise of the Tomb Rider using DX12.

 

[Riek]

I think now that the wccf benchmarks have been debunked, and it looks like rx470 isn't way slower than rx480, it seems like the real question becomes (at least for me) why not buy two rx470's, overclock them to rx480'ish stock, and have a 300 USD solution with gtx1080ish perf (say up to 10% worse since the aots bench was probably cherry picked) except only 4gb vram?

But that doesn't make sense. We know how the 470 would perform from AMD. We know that if the 470 is ~15% away of 480 that it has a huge deficit in flops -> performance

Either we believe 470 is close to 480 and thus AMD has a big regression in CU performance that only exists with 470 (480 doesn't have that as performance per FLOP, according to the leaks, it is more or less the same as hawaii).
Or 470-480 is a bigger difference and the 3dmark leaks are either incorrect or are worst case scenario of 4x0.. series

 

[DeeJayBump]

TPU also benches Hitman in DX11, but also benches Rise of the Tomb Rider using DX12.

Add those to the list as well, as those tactics also benefit nVidia in reviews/testing.

 

[Shivansps]

Hitman has issues under DX12, not sure about Tomb Rider.

 

[Bacon1]

What about the new tomb raider game clearly being on the side of nvidia?

The problem is Async utilization and aots has that issue clearly and games with it will favor amd for awhile it seems.

The same DX12 support in Rise of the Tomb Raider that even the developers suggest not using? Its not done well at all, all cards lose performance (which should never happen) and they even removed the async compute support that existed in the XB1 version.

AoTS is not even a good RTS, Stellaris is waaaaaay better, and its done on DX11, and it does not have render problems, it has IA programming issues, very CPU heavy, common on RTS.
AoTS is little else than a tech demo for Async shaders, if not for Async Shaders no one whould know what that is, and even less use it on benchmarks.

So Ashes of the Singularity which is known to have great AI support because the CPU isn't busy making GPU calls, is a worse RTS than a game that has AI issues?

Stellaris isn't even RTS because its not REALTIME which is what the RT stands for

They aren't similar games at all, no idea why you are comparing them.

 

[Armsdealer]

But that doesn't make sense. We know how the 470 would perform from AMD. We know that if the 470 is ~15% away of 480 that it has a huge deficit in flops -> performance

Either we believe 470 is close to 480 and thus AMD has a big regression in CU performance that only exists with 470 (480 doesn't have that as performance per FLOP, according to the leaks, it is more or less the same as hawaii).
Or 470-480 is a bigger difference and the 3dmark leaks are either incorrect or are worst case scenario of 4x0.. series

...or the simpler answer that 470 stock clock is set lower and should have headroom

We still don't know how many cu 470 is, but if it's 32 vs 36, then it becomes an attractive alternative in my opinion. If it's 28 vs 36, then it's a no go.

 

[Shivansps]

So Ashes of the Singularity which is known to have great AI support because the CPU isn't busy making GPU calls, is a worse RTS than a game that has AI issues?

Stellaris isn't even RTS because its not REALTIME which is what the R stands for

They aren't similar games at all, no idea why you are comparing them.

AoTS uses dirty AI tricks, like making 1 unit = 3 visible units, on Stellaris each unit is individual on combat, the problem is you can have really huge fleets, of several hundred units, even thousands, on confederation wars is not rare to see thousand of units vs thousand of units, AoTS cant even come closer to that, that will destroy any cpu, and its the main problem of Stellaris, CPU perf, rendering is not, you not gona free a lot of cpu power by using async or dx12.

And its not RTS? WTH you are talking about? You dont even know what Stellaris is... i played both games.

https://www.youtube.com/watch?v=CMuoQkxClIc

Thats not even a large confederation battle.

 

[Stormflux]

Stellaris is a 4X game that features Real Time with Pause. No, it's not a RTS.

 

[William Gaatjes]

Tons of pictures of the 480 here:

http://imgur.com/a/Kpdx2

There are the mosfets packages .
I wondered where they were.
I wonder which brand they are . Vishay ?
Or Onsemi ?
Onsemi has some nice mosfets.

 

[Bacon1]

AoTS uses dirty AI tricks, like making 1 unit = 3 visible units, on Stellaris each unit is individual on combat, the problem is you can have really huge fleets, of several hundred units, even thousands, on confederation wars is not rare to see thousand of units vs thousand of units, AoTS cant even come closer to that, that will destroy any cpu, and its the main problem of Stellaris, CPU perf, rendering is not, you not gona free a lot of cpu power by using async or dx12.

And its not RTS? WTH you are talking about? You dont even know what Stellaris is... i played both games.

https://www.youtube.com/watch?v=CMuoQkxClIc

Thats not even a large confederation battle.

It is NOT an RTS. It is a 4x strategy game.

It doesn't handle CPU usage very well - http://www.pic-upload.de/view-30634364/StellarisCPUusage.jpg.html

I have no idea why you continue to bring it up though, it has nothing to do with GPU performance or Polaris (except being space related I guess )

 

[krumme]

390(x) just reused the same coolers the companies made for 290(x).

Yes 390x = 290 plus some memory timing optimizations, pcb rev, slight oc and a good cooler. Aib at launch day and they carried the card forward. Hardly rocket science and dirt cheap to develop and market but with good results and reviewer reception. Shareholders wet dream product and brilliant low cost launch.

Now look at fiji. Fancy tech. Expensive to develop and produce. No sales of any importance. Failure of a product. Its was a risky product and the reward risk profile was crap from the inception.

 

[dahorns]

Stellaris is a 4X game that features Real Time with Pause. No, it's not a RTS.

There are no turns, therefore rts. What does pause have to do with anything? You could pause command and conquer and warcraft, and they were definitely real time strategy games.

 

[krumme]

There are the mosfets packages .
I wondered where they were.
I wonder which brand they are . Vishay ?
Or Onsemi ?
Onsemi has some nice mosfets.

Some tight regulation going on?
Perhaps 14nm process is in need of a boss?

 

[Stormflux]

There are no turns, therefore rts. What does pause have to do with anything? You could pause command and conquer and warcraft, and they were definitely real time strategy games.

You could pause those RTS by going to the system menu, it was not part of the action to action design flow of the game. In Stellaris you can also change the speed of the entire game to be slower or faster. This is just splitting hairs. There are fundamental differences between the two that go beyond the ability to pause that place Stellaris squarely in the 4X genre and not the RTS genre.

 

[William Gaatjes]

Some tight regulation going on?
Perhaps 14nm process is in need of a boss?

Somebody asked about the 6+1 smps vrm on the rx 480 while in some review site it was mentioned that the gtx1080 only had a 5+1 smps vrm while the rx 480 has an 6+1 smps vrm. I answered the post but thought the mosfets where on the backside. It was not clear to see in the picture. But the mosfets are there hiding behind the inductors.

If you wonder, 6 phases for the core voltage for the gpu and 1 phase to power the logic of the GDDR5 memory and the memory controller io driver (to the gddr5 dram)on polaris. At least , that is what i think.

 

[Kenmitch]

It is NOT an RTS. It is a 4x strategy game.

It doesn't handle CPU usage very well - http://www.pic-upload.de/view-30634364/StellarisCPUusage.jpg.html

I have no idea why you continue to bring it up though, it has nothing to do with GPU performance or Polaris (except being space related I guess )

There are no turns, therefore rts. What does pause have to do with anything? You could pause command and conquer and warcraft, and they were definitely real time strategy games.

You could pause those RTS by going to the system menu, it was not part of the action to action design flow of the game. This is just splitting hairs. There are fundamental differences between the two that go beyond the ability to pause that place Stellaris squarely in the 4X/Grand Strategy genre and not the RTS genre.

The above doesn't really fit the thread topic. On another note it gets kind of annoying to view the thread for possible new leaks and see this kind of stuff.

 

[Stormflux]

The above doesn't really fit the thread topic. On another note it gets kind of annoying to view the thread for possible new leaks and see this kind of stuff.

I agree, but the only other news to really talk about is a possible NDA drop for the 24th that has been debunked. Everything else is just reaffirming what everyone seems to know or expect. $229 8GB ~110Watt 390X performance.

I feel bad for reviewers with all the leaks happening and the long NDA... that is still 7 days away.