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

[Headfoot]

I think we may see an ~80% cut die (e.g. 28 CU) with low clocks / TDP target, a ~90% (e.g. 32 CU) die with slightly lowered clocks / TDP target vis a vis rx 480 and the full die with hot clocks for rx 480. I have a feeling they will lower clocks and TDP targets / powertune limits (reaping better perf/w in the process) before cutting the die to ridiculous levels

 

[Ansau]

I think the future will go towards a more simplified lineup of gpus with bigger differences between tiers.

Nvidia has already made wider the difference between 1080 and 1070 (75% of full chip) than the one in Maxwell. 1080 has 2560 CUDA, 1070 has 1920 and it is speculated that 1060 will have 1280.
We didn't see a 960ti, and it doesn't seem to be space for a 1060ti with the RX 480 around without eating either 1060 or 1070.

AMD will probably also go the same way. They have pointed all three products in very specific and delimited uses. RX 460 will have 1024 SP while rx 480 2304. Quite a huge gap between both, but 1536 SP is the perfect spot to be in the middle.
Making it closer to the RX 480 would wide the gap between RX 460 and 470, making the first less valuable, because for just a bit more cash you have much more performance. Or it would lead to a third cut of the Polaris 10 that would push AMD in a situation of having too many product in a very tiny price gap.
Also, with the 300 series AMD started to separate more the tiers in the entry level by making only 2 products: r7 360 and r7 370, with a wider gap between entry level and mid level (r9 380).

With all these I don't see a cut of 67% anything rare...

 

[Headfoot]

I think the future will go towards a more simplified lineup of gpus with bigger differences between tiers.

Nvidia has already made wider the difference between 1080 and 1070 (75% of full chip) than the one in Maxwell. 1080 has 2560 CUDA, 1070 has 1920 and it is speculated that 1060 will have 1280.
We didn't see a 960ti, and it doesn't seem to be space for a 1060ti with the RX 480 around without eating either 1060 or 1070.

AMD will probably also go the same way. They have pointed all three products in very specific and delimited uses. RX 460 will have 1024 SP while rx 480 2304. Quite a huge gap between both, but 1536 SP is the perfect spot to be in the middle.
Making it closer to the RX 480 would wide the gap between RX 460 and 470, making the first less valuable, because for just a bit more cash you have much more performance. Or it would lead to a third cut of the Polaris 10 that would push AMD in a situation of having too many product in a very tiny price gap.
Also, with the 300 series AMD started to separate more the tiers in the entry level by making only 2 products: r7 360 and r7 370, with a wider gap between entry level and mid level (r9 380).

With all these I don't see a cut of 67% anything rare...

That very well could be true. I think you're on to something with the simplified lineup idea. But you are assuming the only way to get to the midpoint between p10 and p11 is cutting the die. That is not true. As we saw with Fury Nano and AMD's mobile laptop APUs, they are very willing to place firmer TDP/power limits and decrease clocks.

I maintain my guess of a historically normal die cut + strict power and clock limiting to hit the middle point, and I think this is more likely than a cut to 67% die. It lets them brag about perf/watt and fill a hole in the lineup at the same time while using cut dies to save money

 

[A_Skywalker]

Do you expect RX 480 to perform so much better than it should be in DX 12? I expect this cards to be very good for the buck, unlike Nvidia which doesnt have async compute and sell cards for 700$.

 

[Ansau]

That very well could be true. I think you're on to something with the simplified lineup idea. But you are assuming the only way to get to the midpoint between p10 and p11 is cutting the die. That is not true. As we saw with Fury Nano and AMD's mobile laptop APUs, they are very willing to place firmer TDP/power limits and decrease clocks.

I maintain my guess of a historically normal die cut + strict power and clock limiting to hit the middle point, and I think this is more likely than a cut to 67% die. It lets them brag about perf/watt and fill a hole in the lineup at the same time while using cut dies to save money

Well, the Fury Nano is more a product made for small form factor pc rather than a midpoint, which Fury non-X is aimed for. In fact, in the AMD web both the Fury X and Fury specifications are packaged in the same tab, while Nano has its own.

 

[zinfamous]

Do you expect RX 480 to perform so much better than it should be in DX 12? I expect this cards to be very good for the buck, unlike Nvidia which doesnt have async compute and sell cards for 700$.

Based on known, quantified recent history, and the little we do know from the trustable benchmarks here:

--Polaris/Vega will outperform nVidia in DX12, this means it will have a much longer lifespan than nVidia's offerings.

--Driver updates from AMD throughout the next 2 or 3+ years will start to bring these cards into performance ahead of their competitors at the same time period--it is certainly highly probable that 480X will edge ahead of 1070 a few years out.

We've seen this happen.

Depending on how you look at that, either AMD makes much better cards (for their customers), or is just terrible at business (doesn't inspire the need to update every cycle with up-marked cards for the "slightly better-than-last-year" premium with nVidia, that never seems to have any desire to optimize their cards for long life). I think both perspectives are correct.

 

[Mopetar]

I don't think driver support matters much on the high-end for Nvidia, at least not for people here. Most of the people who have a Pascal card were upgrading from a 980 or 980 Ti and will be upgrading to Volta as soon as that comes out.

Why do they care if NV drops driver support sooner than AMD if they're always hopping to the newest card?

 

[Head1985]

Do you remembter GTX970M?That was also unreleased GTX960TI.
It is pretty hard cutdown gm204 with 1280SP.I really dont see problem with RX 470 and 1536SP
https://www.techpowerup.com/gpudb/2623/geforce-gtx-970m

Btw 380x is 40% faster than 370x.If rx 470 will have 2048SP the gap will be max 10% in rx 470 vs rx 480.
If rx470 will have 1536SP the gap will be 40% just like 370x vs 380x

 

[Glo.]

If rx470 will have 1536SP the gap will be 40% just like 370x vs 380x

And if the RX470 really has 1536 Cores, it means that guy from reddit who said that there will be 3 SKUs based on Polaris 10 with Configs: 24CU's, 36, and 48 was right in 2 places. So he may be also right in the third one.

Also it would mean perfect closing of the gap between P10 and P11. Remember that we are looking this:

RX 460 - 1024 GCN cores.
RX 470 - 1536/2048 GCN cores.
RX 480 - 2304 cores.

 

[Sweepr]

Btw 380x is 40% faster than 370x.If rx 470 will have 2048SP the gap will be max 10% in rx 470 vs rx 480.
If rx470 will have 1536SP the gap will be 40% just like 370x vs 380x

Radeon R9 370X is based on 4-year old Pitcairn. Cut down Tonga is Radeon R9 380, which has 87.5% the SPs and delivers ~90% the performance of the full chip for a similar price gap (at launch) as the rumoured $50 for the new cards. TechPowerUp's specs aren't more credible/valid than PCGamesHardware's 1920-2048 SPs for Radeon RX 470, it's basically all speculation at this point. If past releases are any indication, MrTeal is correct and you're wrong.

So he may be also right in the third one.

Let it go, please.

 

[dark zero]

Hoping that this product crashes hard in order to retain the minimun prices. Only that would assure a good marketshare and not a paper launch like Pascal (which is impossible to get one in South America)

 

[DisarmedDespot]

Hoping that this product crashes hard in order to retain the minimun prices. Only that would assure a good marketshare and not a paper launch like Pascal (which is impossible to get one in South America)

wut

 

[dark zero]

wut

I mean. The hype of this product is so high that will increase prices hard. So if the hype fails, very few people will buy them. And since the cost is more attractive in lower tier markets like South America or Asia (where the cost is more important than quality), those products will end on those places and people are likely to buy them, increasing the marketshare on those places.

PS: And since I live on South America, I could get one for 240 dollars and without any issue while people in the US and Europe won't get it for a long time... just as planned.

 

[zinfamous]

I don't think driver support matters much on the high-end for Nvidia, at least not for people here. Most of the people who have a Pascal card were upgrading from a 980 or 980 Ti and will be upgrading to Volta as soon as that comes out.

Why do they care if NV drops driver support sooner than AMD if they're always hopping to the newest card?

They don't care. That's my point. NVidia's model is to sell the absolute high end on very short cycles with little need to extend life on these cards because their market will pay whatever nVidia wants them to pay as soon as they have new chips available.

It's just a different business model from where AMD seems to have placed themselves. AMD seems to be owning the console and long-refresh laptop/Apple cycle where their cards need to have extended, competitive lifespans.

Some users value the price/performance/duration of AMD cards vs paying premium for the top kitty every year with nVidia.

 

[dark zero]

They don't care. That's my point. NVidia's model is to sell the absolute high end on very short cycles with little need to extend life on these cards because their market will pay whatever nVidia wants them to pay as soon as they have new chips available.

Similar to Apple. At least there is no pay to win in those games... they would pay it.

 

[zinfamous]

I mean. The hype of this product is so high that will increase prices hard. So if the hype fails, very few people will buy them. And since the cost is more attractive in lower tier markets like South America or Asia (where the cost is more important than quality), those products will end on those places and people are likely to buy them, increasing the marketshare on those places.

PS: And since I live on South America, I could get one for 240 dollars and without any issue while people in the US and Europe won't get it for a long time... just as planned.

yeah, good luck there.

 

[Silverforce11]

are these specs real? 1536 SPs, 48ROPs and so on!?
https://www.techpowerup.com/gpudb/2861/radeon-rx-470

or did he get the specs from a 470M or something?

No they are not real. It's placeholder, same as the RX 460.

I don't know why you guys give credence to such nonsense.

There's only been 2 SP counts that are reliable for C4 and C7, that's 2048 SP and 2304 SP. Since RX 480 is already confirmed to be 36 CU or 2304 SP, it stands to reason the same leak will be true for RX 470 at 2048 SP.

As for people being so negative about AMD's slides. Drivers 16.1 and 16.2 are from January and Feb. Do you think AMD have optimized for Polaris back then already? Remember what they said in CES (Jan), demo was running on ES sample with non-optimized drivers.

The interesting thing here is why AMD deliberately send out such misleading info that's using very out-dated drivers, and they included a very specific disclaimer "results may vary due to use of drivers"... why do that?

We have other leaks, claimed review samples that have different bioses, and recently, claimed retail samples, which show much better results using more current drivers.

This looks like a misinformation campaign all the way to the launch reviews.

 

[DDH]

No they are not real. It's placeholder, same as the RX 460.

I don't know why you guys give credence to such nonsense.

There's only been 2 SP counts that are reliable for C4 and C7, that's 2048 SP and 2304 SP. Since RX 480 is already confirmed to be 36 CU or 2304 SP, it stands to reason the same leak will be true for RX 470 at 2048 SP.

As for people being so negative about AMD's slides. Drivers 16.1 and 16.2 are from January and Feb. Do you think AMD have optimized for Polaris back then already? Remember what they said in CES (Jan), demo was running on ES sample with non-optimized drivers.

The interesting thing here is why AMD deliberately send out such misleading info that's using very out-dated drivers, and they included a very specific disclaimer "results may vary due to use of drivers"... why do that?

We have other leaks, claimed review samples that have different bioses, and recently, claimed retail samples, which show much better results using more current drivers.

This looks like a misinformation campaign all the way to the launch reviews.

Im not sure I believe it was deliberate. That type of campaign to spread incorrect information about a product would hinder not help imo. And the only plausible reason for doing such a thing would be to obfuscate their competition, but I cant imagine AMD providing a slide like that for such an end.

To me, the more believable option is they were the results at the time, the marketing team put together the slide based on that information. Perhaps at the AMD conference, the speaker made note of early drivers used in the test, indicating that better results would be made with proper drivers, we wont know.


But I also dont put much validity in the SteamVR test with such a wide range of variations with the same GPU but different other hardware configurations. In my mind 3dmark is more of an indicator(official benchs, not leaks) as it isolates the graphics score from other the cpu and overall system score.

 

[sirmo]

No they are not real. It's placeholder, same as the RX 460.

I don't know why you guys give credence to such nonsense.

There's only been 2 SP counts that are reliable for C4 and C7, that's 2048 SP and 2304 SP. Since RX 480 is already confirmed to be 36 CU or 2304 SP, it stands to reason the same leak will be true for RX 470 at 2048 SP.

As for people being so negative about AMD's slides. Drivers 16.1 and 16.2 are from January and Feb. Do you think AMD have optimized for Polaris back then already? Remember what they said in CES (Jan), demo was running on ES sample with non-optimized drivers.

The interesting thing here is why AMD deliberately send out such misleading info that's using very out-dated drivers, and they included a very specific disclaimer "results may vary due to use of drivers"... why do that?

We have other leaks, claimed review samples that have different bioses, and recently, claimed retail samples, which show much better results using more current drivers.

This looks like a misinformation campaign all the way to the launch reviews.

Yup, everything AMD has released I see as an attempt to hide the true performance of Polaris as much as possible. They seem to be purposely picking the most vague and underwhelming specs

If I was in charge of AMD, and I was about to launch one of the most market disruptive GPUs in recent years, I would try to hide the true performance numbers until the cards are on the shelves.

Releasing those numbers early, or overhyping a product only hurts the launch.

 

[Armsdealer]

Yup, everything AMD has released I see as an attempt to hide the true performance of Polaris as much as possible. They seem to be purposely picking the most vague and underwhelming specs

If I was in charge of AMD, and I was about to launch one of the most market disruptive GPUs in recent years, I would try to hide the true performance numbers until the cards are on the shelves.

Releasing those numbers early, or overhyping a product only hurts the launch.

As absolutely retarded as it sounds, I think this is the only possible conclusion. They're taking a page from the Porsche playbook. Porsche always downplays their cars performance stats so when the consumer gets them they're like - "Wow! This car is way better than the company said. Porsche is pure quality!" Obviously Porsche knows there are plenty of third parties that will accurately test and assess the car, so if you can seem like a class act by downplaying the performance, why not?

I don't think they're doing it for the sake of NVidia. If P10 is as good as wccf is making it sound, NVidia probably rushed to market to fleece the enthusiasts before AMD's launch knowing full well the strength of Polaris.

 

[sirmo]

As absolutely retarded as it sounds, I think this is the only possible conclusion. They're taking a page from the Porsche playbook. Porsche always downplays their cars performance stats so when the consumer gets them they're like - "Wow! This car is way better than the company said. Porsche is pure quality!" Obviously Porsche knows there are plenty of third parties that will accurately test and assess the car, so if you can seem like a class act by downplaying the performance, why not?

I don't think they're doing it for the sake of NVidia. If P10 is as good as wccf is making it sound, NVidia probably rushed to market to fleece the enthusiasts before AMD's launch knowing full well the strength of Polaris.

You also have to remember, some of the top brass at AMD today came straight from Apple. If they learned anything about product launches, this would be the one thing to learn, as this is one thing Apple does really well.

 

[Silverforce11]

Well I am not impressed at all. I think it's very poor form to have official slides this close to launch with what amounts to wrong info.

This is like Raja at Computex, 2x RX 480 vs 1080, "but wait.. that's not all... 51% GPU usage"... like what? You do not advertise the fact you have terrible CF scaling. No, that's just stupid. I was watching that stream and literally /facepalm when I saw he was hyping up it's scaling as if it's a good thing.

 

[sirmo]

Well I am not impressed at all. I think it's very poor form to have official slides this close to launch with what amounts to wrong info.

This is like Raja in Computex, 2x RX 480 vs 1080, "but wait.. that's not all... 51% GPU usage"... like what? You do not advertise the fact you have terrible CF scaling. No, that's just stupid.

The 51% GPU usage thing was pretty stupid I agree. I understand what he was trying to do, but it did come out wrong.

 

[Armsdealer]

You also have to remember, some of the top brass at AMD today came straight from Apple. If they learned anything about product launches, this would be the one thing to learn, as this is one thing Apple does really well.

great point

 

[96Firebird]

Time to add to the board "AMD faking marketing slides"...