AMD Radeon RX 480 specifications leaked?

[HTC]

PowerColor PCS+ R9 380X 4GB $199.99

Double the memory and 350 MHz faster clocks. Not really even worth waiting for...



Polaris 11 is a card you use in HTPCs and Servers to compensate for lack of graphics card. It'll only be able to handle very light gaming.

I have an APU with no dedicated video card: only onboard video. Polaris 11 should be a HUGE upgrade, no?

 

[Caring1]

PowerColor PCS+ R9 380X 4GB $199.99

Double the memory and 350 MHz faster clocks. Not really even worth waiting for...

Both have 4Gb and the 380X has two 6 pin connectors, so power usage will be much higher for the same performance.

 

[FordGT90Concept]

I have an APU with no dedicated video card: only onboard video. Polaris 11 should be a HUGE upgrade, no?

Your APU has 8 GPU cores. RX 480 has 36. Even if Polaris 11 has half that, it should still be at least twice as fast. DX12 games should be able to combine the power of both.

You won't get a 8GB 480 for $199 though. That price is reserved for the 4GB version.

Both have 4Gb and the 380X has two 6 pin connectors, so power usage will be much higher for the same performance.

Makes sense. Truth be told, 8 GiB isn't a selling point for me compared to 4 GiB so, 216 MHz higher clock and a little lower power consumption...

I guess I'm not excited because I usually buy graphics cards around the $300 price point. RX 480 looks like a side grade more than upgrade to me.

 

[NdMk2o1o]

Both have 4Gb and the 380X has two 6 pin connectors, so power usage will be much higher for the same performance.

The same performance? It should be a lot better than a 380x...

Minimum of 390/390x performance if going off all the speculation

 

[FordGT90Concept]

I suppose it is theoretically possible from architectural improvements but I doubt it. We'll have to wait until June 29 for the benchmarks.

 

[Kanan]

I suppose it is theoretically possible from architectural improvements but I doubt it. We'll have to wait until June 29 for the benchmarks.

It's easily faster than 380X and 390(X). But yeah for a 390(X) user it's more like a light upgrade with lower power consumption, but let's wait how good it overclocks, maybe there's a lot of untapped power in it. Still great value for that price.

~29:00 if you want to see the actual card (Raja holds it in his hands).

 

[NdMk2o1o]

R9 Fury performance for $200 is pretty darn sweet if this holds up to be true

Hoping zen will be a success as well so I can have my first amd rig in about 7 years....

 

[Caring1]

We'll have to wait until June 29 for the benchmarks.

I still question this chart as figures I have seen put the 480 around the GTX970 range.
If the 480X is the C7, that would make more sense.

 

[Kanan]

I still question this chart as figures I have seen put the 480 around the GTX970 range.
If the 480X is the C7, that would make more sense.

Wouldn't change much. The 480X has probably 2 additional CU's activated, that's only a tad more at 2560 shaders (same as with 380 vs 380X or 390 vs 390X). I really think this is the RX 480 it's faster than 980 and comparable with Fury which is great. I'm sure because AMD posted benches with AotS and Tflop numbers that go hand in hand with it. Just the "CF" chart seems off, probably not working properly pre optimized drivers, it should be easily faster than 1080 in CF.

 

[Caring1]

Just the "CF" chart seems off, probably not working properly pre optimized drivers, it should be easily faster than 1080 in CF.

True, it barely makes a 50% improvement over a single card when it should be at least 70%.

 

[Kanan]

~38:04 ... that's more realistic numbers for RX 480 in CF. But I dont want to hype it, let's wait for proper reviews to be sure.

 

[ivan375]

Just the "CF" chart seems off, probably not working properly pre optimized drivers, it should be easily faster than 1080 in CF.

Could have been alluding to that with the 51% utilization. He says "we have much more headroom as the developer tunes the APIs and as we tune the drivers for the new gpu."

They also used a vague "> 5 tflops" but 36 CUs at a 1266 clock would be above 5.8 tflops. Seems like they were holding back a little, perhaps waiting on a better driver to do a proper demo. Not sure, but the floor for this seems like something slightly above 390/970, which is not bad.

 

[medi01]

WSJ article mentiones 199$ card with 500$ performance.
This perfectly matches C7 perf in this chart, which is between 980 and Fury (non X).
In the worst case, if it is C4-ish (for whatever reason) we get perf between 970 and 980/390x.
Still quite good.
(Ignore top white bar, it's crossfire)

 

[HD64G]

Wouldn't change much. The 480X has probably 2 additional CU's activated, that's only a tad more at 2560 shaders (same as with 380 vs 380X or 390 vs 390X). I really think this is the RX 480 it's faster than 980 and comparable with Fury which is great. I'm sure because AMD posted benches with AotS and Tflop numbers that go hand in hand with it. Just the "CF" chart seems off, probably not working properly pre optimized drivers, it should be easily faster than 1080 in CF.

For AotS comparison they showed that Polaris duo was utilised at 51%. Practical, just ne of them was used. And it beat 1080 but in lower settings as we can see from the video. But for the diff in price, that's surely a very good GPU there.

 

[medi01]

it beat 1080 but in lower settings

It was said having more or less snow in game is "normal" for AoT, quite a bit of it is random so not repeatable.

 

[HD64G]

It was said having more or less snow in game is "normal" for AoT, quite a bit of it is random so not repeatable.

If so, and we all should hope that's true, we are talking about GREAT performance in DX12. But let's wait for reviews before making that assumption definite eh?

 

[medi01]

But let's wait for reviews before making that assumption definite eh?

Yep.

PS
By some random guy on the internet, to be taken with a pound of salt (480 is assumed to perform between 970 and 980):

 

[Totally]

Yep.

PS
By some random guy on the internet, to be taken with a pound of salt (480 is assumed to perform between 970 and 980):

isn't that info we already know? 480X is C7 and 480 is C4 and that lands between the 980 ans 970

 

[medi01]

isn't that info we already know? 480X is C7 and 480 is C4 and that lands between the 980 ans 970

We don't know what C7 and C4 really are.
We know that one of the 480's that was demoed was C7 (the Doom demo "leak")
But that 3D mark can be... anything. Including OCed cards.
Based on AotS results C7 isn't that different from 390 perf wise, which is lower than expected from 3dmark leaks.

 

[Eroticus]

AMD claims that Polaris architecture has 2.8 times the performance per Watt compared to older architecture (Radeon R9 270X to be specific).

Footnotes from new press deck revealed some interesting information. First off, the mobile RX series will use different naming schema than already released R? M4?? series. For instance, AMD confirmed that new card based on Polaris 11 is called Radeon RX 480M (so the M is now a postfix). This card is also confirmed with TDP at 35W, and speaking of watts, Radeon RX 470 is confirmed to be 110W TDP card. Later in this post you will find some actual official benchmarks.



AMD Radeon RX 480 8GB scores 6.3 points in Stream VR benchmark. As noted by ComputerBase, R9 390 has similar/indentical score, which explains why AMD used R9 380 in this comparison..



It’s worth nothing that this slide is titled ‘first in a family’, does it mean there are more cards coming later?



We also get to see the first official pictures of Polaris silicons. Polaris 10 will deliver 5+ FLOPS, while Polaris 11 can offer 2+ TFLOPs single-precision computing power.





UPDATE: Tweakers.net posted some benchmarks based on footnotes in the press deck:

Radeon RX 480M vs R9 280M


Radeon RX 470 vs R9 270X


Source: ComputerBase
Originally posted by WhyCry.