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

[Head1985]

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.

Then AMD is pretty dumb
RX 470 with same performance as rx 480.Its like GTX1060 with almost same performance as GTX 1070.The gap should be 30-40% if one card have 470 name and other 480 name and not 10%.
Also 2048Sp dont make any sense.rx 470 is 60% faster than 270x.
270x have 1280SP at 1Ghz
rx470 have 2048SP at 1.2Ghz + new architecture + much faster memory.
The gap should be more than 100%.

SP difference is 60%
Clock difference is 20%
Pitcairn is more efficient than rx 470?

If rx 470 have 1536SP:
270x 1280SP at 1Ghz
rx470 1536SP at 1.2Ghz

rx470 have 20% more SP and 20% more clock and 20% from new architekture + faster memory vs old tahity.

 

[Glo.]

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.

67DF:C4 according to SiSoft has 2304 GCN cores. http://ranker.sisoftware.net/show_d...a09dac8ae2dfeaccb489b89efb9ea393b5c6fbcb&l=en

 

[flopper]

Just like "Pascal is 10x faster than Maxwell"... later we find out only in very unique compute application. Then JHH hyped it, "GTX 1080 is 2x as fast as Titan X".. only in their imaginary VR game that does not exist.

So gamers always will read the launch reviews from the tech press for validation. Even if AMD provided a full benchmark suite, people will still give clicks to tech sites to confirm.

My thought on this, they are holding back because they want to wait til the final launch, giving their driver team more time to *optimize* as much as possible. When they want the tech press to start reviews on it, they'll send a link to download the final bios & driver so these guys can run benchmarks. Prior to this, samples have different bioses and running on non-ready drivers, so any leak will be misleading.

* We all know how AMD improves performance of a new architecture over time via drivers.

Amd's marketing always failes so hard, huang's presentation of he performance of nvidia cards are miles better.

One cant say the Nvidia CEO did lie in front of an audience saying 10x faster but its missleading as it dont apply for what people would expect which then cause a huge backlash for anyone buying for that 10x improvement and find their games dont run any 10x faster.

AMD had more time to finalize, driver and bios, seems the thirdparty solutions with 6+8 power connections will OC well and then amd have done something really interesting,
they locked down 1080p to the 470 card and then the 1440p with the 480 card creating a vaccum as the majority will understand the value of that offering and buy 480 today

So a perfect offering with the 470/480 and then hype builds for Vega to be like the best things since Mantle and Dx12. An awesome plan as Vega will rock with 4k.

all resolutions covered and Nvidia has no answer then.
To expensive with to big die.
and the number of cards shown they have yield issues.
and besides do anyone enjoy to pay a 100$ tax really for no benefit?

 

[3DVagabond]

One cant say the Nvidia CEO did lie in front of an audience saying 10x faster but its missleading as it dont apply for what people would expect which then cause a huge backlash for anyone buying for that 10x improvement and find their games dont run any 10x faster.

AMD had more time to finalize, driver and bios, seems the thirdparty solutions with 6+8 power connections will OC well and then amd have done something really interesting,
they locked down 1080p to the 470 card and then the 1440p with the 480 card creating a vaccum as the majority will understand the value of that offering and buy 480 today

So a perfect offering with the 470/480 and then hype builds for Vega to be like the best things since Mantle and Dx12. An awesome plan as Vega will rock with 4k.

all resolutions covered and Nvidia has no answer then.
To expensive with to big die.
and the number of cards shown they have yield issues.
and besides do anyone enjoy to pay a 100$ tax really for no benefit?

He lies all of the time. When he said he had Fermi in his hand and it was a wood screw mock up? When he said he had GP100 in his hand and it was a Maxwell mobile GPU? When he said Pascal Founder editions had premium cooling solutions tp help justify the early adopter tax?

 

[DeathReborn]

VR ready for $199 and 2.8X perf/W are the claims AMD has made. You obviously don't want to talk about what they've actually said. Instead you want to talk about what they haven't said. Why is that?

Didn't they call it VR Premium & up to 2.8x Perf/W?

 

[Silverforce11]

67DF:C4 according to SiSoft has 2304 GCN cores. http://ranker.sisoftware.net/show_d...a09dac8ae2dfeaccb489b89efb9ea393b5c6fbcb&l=en

I've seen two other entries for C4 with 2048 SP, I made a note of it a few months ago with the shader counts for C4 and C7. I've been trying to find it again, it's impossible to dig out from the database there. They have such useless search features.

 

[zinfamous]

Didn't they call it VR Premium & up to 2.8x Perf/W?

No, the second part I don't recall.

I think they said something like "VR ready...for premium graphics experience."

There's a difference between defining VR as the premium tier of graphics, and something being VR ready, and then further a premium VR experience.

This brings you into the premium tier at half the cost of current intro products, but probably not the premium level within the premium tier.

 

[Glo.]

I've seen two other entries for C4 with 2048 SP, I made a note of it a few months ago with the shader counts for C4 and C7. I've been trying to find it again, it's impossible to dig out from the database there. They have such useless search features.

It isn't the case. If it would be like you are describing it, the specs for 67DF:C4 would be on the bottom of the SiSoft page with other entries for 67DF:C4.

And only GPU with 67DF:C4 Device ID nas 2304 GCN cores. Not even Google finds anything about 2048 GCN core chip with mentioned Device ID.

 

[cytg111]

...Some poster says he heard an unconfirmed report that Asus is releasing a 1600MHz part and to take it with a grain of salt. Now it's supposedly everyone's expectations and somehow it's AMD's fault.

It is how the Spetsnaz operate. Illuminati confirmed.
Seriously though, yea, alot of resonance over nothing.

 

[sm625]

Some poster says he heard an unconfirmed report that Asus is releasing a 1600MHz part and to take it with a grain of salt. Now it's supposedly everyone's expectations and somehow it's AMD's fault.

Meh. I'm looking for 1800.

 

[casiofx]

Then AMD is pretty dumb
RX 470 with same performance as rx 480.Its like GTX1060 with almost same performance as GTX 1070.The gap should be 30-40% if one card have 470 name and other 480 name and not 10%.
Also 2048Sp dont make any sense.rx 470 is 60% faster than 270x.
270x have 1280SP at 1Ghz
rx470 have 2048SP at 1.2Ghz + new architecture + much faster memory.
The gap should be more than 100%.

SP difference is 60%
Clock difference is 20%
Pitcairn is more efficient than rx 470?

If rx 470 have 1536SP:
270x 1280SP at 1Ghz
rx470 1536SP at 1.2Ghz

rx470 have 20% more SP and 20% more clock and 20% from new architekture + faster memory vs old tahity.

IPC doesn't mean anything, end result is the thing to look for.
Besides, did you forgot R9 290 vs 290X? $150 difference but performance difference was around 10%

 

[Sweepr]

Possible final specs for the new lineup?

74% Radeon RX 480 processing power and 87.5% memory bandwidth (base clocks). That's pretty good actually. Geforce GTX 1070 delivers 82.6% Geforce GTX 1080 performance at 1440p with 70% processing power and 80% the bandwidth (base clocks).

 

[antihelten]

Possible final specs for the new lineup?

74% Radeon RX 480 processing power and 87.5% memory bandwidth (base clocks). That's pretty good actually. Geforce GTX 1070 delivers 82.6% Geforce GTX 1080 performance at 1440p with 70% processing power and 80% the bandwidth (base clocks).

Specs really seem to be all over the place for RX 470, I've seen reports of 1536, 1792, 2048 and 2304 shaders for it.

Either way the benches AMD released, showed the 470 as 78% faster than a 270X in the 3 games tested, which would make it roughly equal to a 390 (a 390 is 77% faster than a 270X). If we go by the specs above and assume linear scaling (which of course are never really the case), then that would make the RX 480 35% faster than the RX 470 (unless bandwidth is a bottleneck*).

The 480 being 35% faster than the 470 would make it roughly equal to a 980 Ti. That level of performance would be well above the majority of other rumours out there, so I would take those specs with a grain of salt.

*If bandwidth is the main bottleneck, then 480 should be about the level of R9 Fury.

 

[Sweepr]

Latest from WCCFTech:

AMD Radeon RX 470 Single and Crossfire 3DMark 11 Benchmarks Spotted – Targeting the ‘VR Minimum Spec’ Under $200

http://wccftech.com/radeon-rx-470-crossfire-3dmark-11-benchmarks/#ixzz4BkmkYUKK

I think '67DF:C4' is not Radeon RX 470. According to all leaks I saw that's a 2304 SPs 800 MHz part. Doesn't make sense to have clockspeed as the only difference between it and the Radeon RX 480. Possibly a mobile SKU?

 

[Sweepr]

If we go by the specs above and assume linear scaling (which of course are never really the case), then that would make the RX 480 35% faster than the RX 470 (unless bandwidth is a bottleneck*).

As you said that's usually not the case. Just look at GP104, specs suggest (linear scaling) the Geforce GTX 1080 is 42% faster than the cut down model, but it's actually only 21%, with a bigger processing & bandwidth 'gap' than Radeon RX 480 vs (rumoured) Radeon RX 470.

 

[Glo.]

Latest from WCCFTech:

AMD Radeon RX 470 Single and Crossfire 3DMark 11 Benchmarks Spotted – Targeting the ‘VR Minimum Spec’ Under $200

http://wccftech.com/radeon-rx-470-crossfire-3dmark-11-benchmarks/#ixzz4BkmkYUKK

I think '67DF:C4' is not Radeon RX 470. According to all leaks I saw that's a 2304 SPs 800 MHz part. Doesn't make sense to have clockspeed as the only difference between it and the Radeon RX 480. Possibly a mobile SKU?

It is a possibility. The problem with that is that Mobile parts have slightly different deviceID's.

I also don't think that RX 470 is 67DF:C4 part.

AMD played very well to keep the GPUs completely secret, and at this point, so close to official launch we still do not know everything for sure.

Edit:

The Radeon RX 470 is based on the cut version of the Polaris 10 GPU, specifically the variant we know as 67DF:C4.

This part directly contradicts to what we know from the SiSoft benchmark.

On the other hand. Difference in core clocks is enough to pay the difference in 3dMark 11 performance(12k vs 16.5k).

RX 480 has 1266 MHz core clock, and 67DF:C4 has around 800 MHz core clock.

 

[antihelten]

As you said that's usually not the case. Just look at GP104, specs suggest (linear scaling) the Geforce GTX 1080 is 42% faster than the cut down model, but it's actually only 21%, with a bigger processing & bandwidth 'gap' than Radeon RX 480 vs (rumoured) Radeon RX 470.

Even if the performance of the 480 relative to the 470 would primarily be dictated by the difference in memory bandwidth (as appears to be the case with the 1080 vs. the 1070), the 480 would still end up somewhere around the R9 Fury, which still seems a tad high, but not entirely unrealistic I suppose.

Edit: This part directly contradicts to what we know from the SiSoft benchmark.

On the other hand. Difference in core clocks is enough to pay the difference in 3dMark 11 performance(12k vs 16.5k).

RX 480 has 1266 MHz core clock, and 67DF:C4 has around 800 MHz core clock.

Some of that could be down to base clock versus boost clock and engineering sample versus retail sample, so the numbers aren't necessarily contradictory (just measuring different things/samples).

 

[krumme]

Imo the success depends on gf ability to produce the stuff.

If yield is low and capacity is constrained Apple Sone amd Ms is first in line. Second row is mobile. Desktop gets what is left. Ps4 neo and next xbox will take up very much capacity at gf - like what 40M usnits over 3 years? A few M for apple and and few M for mobile. Man it can easily be constrained if some bug is difficult to find. Lol.

 

[Armsdealer]

Possible final specs for the new lineup?

What's the source on that? Those seem to be the most believable numbers we've seen yet...

 

[Sweepr]

Here comes the price cuts:

- Radeon R9 390X down to 299 Euro
- Radeon R9 390 down to 254 Euro
- Radeon R9 380X down to 179 Euro

www.computerbase.de/2016-06/radeon-r9-390-x-380-x-preissenkung

What's the source on that? Those seem to be the most believable numbers we've seen yet...

Chiphell.

 

[MrTeal]

Really, the 460 might end up being by far the best upgrade of the whole bunch of these cards. Right now the 360 (1GHz/768SP) is available for right around $100AR, which is the listed launch price in Sweepr's chart. Even assuming a GCN1.1 SP = GCN4 SP, that's 33% more SPs and TMUs and twice the ROPs running 30% faster, or at least 73% more processing power. That, plus you get all the new codec/interface goodies and drop the 6-pin connector on the card. That's a massive upgrade for the same price as the currently available outgoing card, when you consider that overclocked 980Ti's are available for the same $380 that the 1070 is supposed to eventually hit, and you can buy a $249 R9 390 for only a few more dollars that the 8GB 480 is expected to come in at while both pairs of cards might offer similar performance.

The $100 GPU segment has been cesspool of crap value for a long time. We'll see what GP107/8 brings, but this card could be a godsend for the budget 1080p HTPC/Gaming build.

 

[antihelten]

Really, the 460 might end up being by far the best upgrade of the whole bunch of these cards. Right now the 360 (1GHz/768SP) is available for right around $100AR, which is the listed launch price in Sweepr's chart. Even assuming a GCN1.1 SP = GCN4 SP, that's 33% more SPs and TMUs and twice the ROPs running 30% faster, or at least 73% more processing power. That, plus you get all the new codec/interface goodies and drop the 6-pin connector on the card. That's a massive upgrade for the same price as the currently available outgoing card, when you consider that overclocked 980Ti's are available for the same $380 that the 1070 is supposed to eventually hit, and you can buy a $249 R9 390 for only a few more dollars that the 8GB 480 is expected to come in at while both pairs of cards might offer similar performance.

The $100 GPU segment has been cesspool of crap value for a long time. We'll see what GP107/8 brings, but this card could be a godsend for the budget 1080p HTPC/Gaming build.

Basically if the specs Sweepr posted are correct, then the RX 460 should be about equal to a GTX 960, which seems like a fairly decent deal at $100

 

[dazelord]

Basically if the specs Sweepr posted are correct, then the RX 460 should be about equal to a GTX 960, which seems like a fairly decent deal at $100

I remember Sweepr posted some days ago leaked benchmarks for the 460 that put it next to the 270X. So a little bit slower than the 960.

 

[dark zero]

Really, the 460 might end up being by far the best upgrade of the whole bunch of these cards.

The $100 GPU segment has been cesspool of crap value for a long time. We'll see what GP107/8 brings, but this card could be a godsend for the budget 1080p HTPC/Gaming build.

The RX 460 might end at 99 dollars being the savior of that área.

Also expecting the RX 450 for LATAM and African markets.

The RX 460 is the competitor of the GT 1040 and the RX 450 of the GT 1030.

Expecting to see the RX 440 and the GT 1020.

Anyways. All those chips will humiliate Intel iGPU to new levels.

 

[DeathReborn]

No, the second part I don't recall.

I think they said something like "VR ready...for premium graphics experience."

There's a difference between defining VR as the premium tier of graphics, and something being VR ready, and then further a premium VR experience.

This brings you into the premium tier at half the cost of current intro products, but probably not the premium level within the premium tier.

http://www.fudzilla.com/news/graphi...most-polaris-11-and-polaris-10-specifications

VR Premium on the RX 480 & up to 2.8x perf/w on both P11 & P10. Now please excuse me while I shower to get rid of the grime of linking that site.