AMD: Speculation, Rumors, and Discussion (Archive)

[Silent_Buddha]

That's going to be a big die... and yeah you can save some power with lower clocks, but...

Well Rx480 min clock is apparently 910MHz... so not that much...

I mean one can wear whatever shade of glasses he or she pleases, but it is not going to change reality....

Yes, but the PS4 Neo is not likely going to feature a boost clock. IE - it's never going to hit 1.266 GHz when gaming. Ever. They won't need it to hit the rumored 4.4 TFLOPs.

If the Rx480 stayed around 910 GHz when gaming at all times, it's power consumption is likely to be quite significantly lower.

That's a rather significant 256 MHz. Or a 20% lower clockspeed. And if it puts it below the knee of the power curve, a far FAR greater reduction in power consumption than 20%.

Regards,
SB

 

[Kaotik]

Seriously though, wth is up with european pricing? It seems that only UK has the pricing right (USD > GBP + VAT), while rest of the europe is paying 10's of euros extra.

 

[Silent_Buddha]

Reduced voltage too..but of course you meant that

Slight correction on the Jaguar cores..console parts are clocked at 1.6 Ghz but there are 8 of them. The console GPUs run at ~800 mhz vs ~1 ghz for non-console.

Whoops, I totally wiffed on that one. Thanks for the correction.

Regards,
SB

 

[ninelven]

Yes, but the PS4 Neo is not likely going to feature a boost clock. IE - it's never going to hit 1.266 GHz when gaming. Ever. They won't need it to hit the rumored 4.4 TFLOPs.

Why are you telling me things I already know? And why even this topic why my OP specifically referenced MS? Donezo.

 

[Orion]

Yeah with you on this.
Also KitGuru, hardware.fr, Techspot, TomsHardware has the 480 below or close to the 980, and not challenging the 980ti - strange results from pcgamer.
In all those reviews the 980 is close to but just below a 390, and for some reason the 480 is slower than the 390.
I could also mention HardwareCanuck but they use PresentMon not the internal benchmark, although still similar positions.
This is at 1080p rather than 1440p.

Cheers

Did most of those use crazy settings? I'm reading high and extreme settings. The one's with high are obviously not crazy settings. What about extreme settings is that the same as crazy settings(don't have ashes to test)?

http://www.hardwarecanucks.com/foru...9-radeon-rx480-8gb-performance-review-20.html
Canuck had 480 with 30% faster performance than gtx 980 in hitman at 1440p, iirc. And amd performance difference with nvidia cards in hitman is less than amd performance difference in crazy settings ashes.

Something weird is definitely going on. The 480 is at most within 10-15~% of 390x in many titles, in some cases the difference is under 5%. The 390x reviewed at 30% faster than a 980ti in ashes in crazy. So unless something weird's going on with the benches or mad optimization has taken place something's off.

 

[3dilettante]

*image snipped*

I know it's an artistic rendition, but if it's any indication as to how pad-limited a chip with a 256-bit bus can be it might explain why the PS4's getting an APU with a larger GPU. Not sure it could avoid a swath of blank silicon.

Can this be reconciled with way the GPU is organized in the diagram? The CUs might be grouped differently.

 

[Jawed]

https://www.reddit.com/r/Amd/comments/4qfwd4/rx480_fails_pcie_specification/d4szyz6

We're shipping 4GB to Germany, UK and USA for now. Later in the summer we will spread it to more regions.

So, not EOL.

That thread looks like a right old mess, by the way.

 

[bdmosky]

Those are high settings
In feb, at high settings on ashes the 980 ti > 390x. But going to crazy settings drastically reversed the situation in AMDs favor with 390x > 980 ti, by 30%.


It is likely pcgamer used crazy settings, in which case we could see how 480>980ti.

Very odd that crazy benches seemed missing in many reviews.

Well, they said themselves they reused old scores with older drivers and game versions. I wouldn't put much stock in any of their comparisons with the lack of clarity in their descriptions.

 

[CSI PC]

Did most of those use crazy settings? I'm reading high and extreme settings. The one's with high are obviously not crazy settings. What about extreme settings is that the same as crazy settings(don't have ashes to test)?

http://www.hardwarecanucks.com/foru...9-radeon-rx480-8gb-performance-review-20.html
Canuck had 480 with 30% faster performance than gtx 980 in hitman at 1440p, iirc. And amd performance difference with nvidia cards in hitman is less than amd performance difference in crazy settings ashes.

Something weird is definitely going on. The 480 is at most within 10-15~% of 390x in many titles, in some cases the difference is under 5%. The 390x reviewed at 30% faster than a 980ti in ashes in crazy. So unless something weird's going on with the benches or mad optimization has taken place something's off.

Hitman is an anomaly that works for either Nvidia or AMD depending upon which episode one uses and also section,
As an example here is an example showing the results skewing one way or the other:
http://www.pcgameshardware.de/Hitma...Episode-2-Test-Benchmarks-DirectX-12-1193618/
http://www.pcgameshardware.de/AMD-Radeon-Grafikkarte-255597/Specials/RX-480-Test-1199839/2/
Look at DX12 or DX11 for both and the performance of the 980ti relative to a FuryX - point is Hitman is too erratic as a consistent benchmark.

HardwareCanucks is crazy setting but uses PresentMon.
The others used Extreme and High because Crazy is an impractical setting for most when playing the game with a 480 (if not going by the PCGamer fps but other sites)
I have just checked pcgamer and they state:

we run all three resolutions using the default Extreme preset.

So they are using Extreme like some of those I mentioned earlier, something is strange IMO about PCGamer in this specific game.
But still, the 480 is the definite card of choice if it was a decision between it, 970 and 390 even with their reduced prices.
Cheers

 

[Malo]

https://www.reddit.com/r/Amd/comments/4qfwd4/rx480_fails_pcie_specification/d4szyz6
That thread looks like a right old mess, by the way.

Does indeed look like a mess. Isn't the point to get your best cards to reviewers, not your worst cards?

 

[pharma]

I've been swayed by AMD marketing so many times before. I am not convinced that out of the hundreds of reviewers, by some crazy coincident, the ones that do these power consumption tests all received faulty cards.

https://np.reddit.com/r/Amd/comments/4qhez3/this_seems_important/

 

[msia2k75]

Seems there are 2880sp in that shot?

 

[Kaotik]

Seems there are 2880sp in that shot?

Huh? How do you get 2880 from that?

 

[entity279]

OMG AMD has just released a slightly cheaper GTX 970, with the entire RAM pool available this time? *runaway*

 

[Orion]

Well, they said themselves they reused old scores with older drivers and game versions. I wouldn't put much stock in any of their comparisons with the lack of clarity in their descriptions.

Still the performance at 1440p crazy seems to have dropped from 45fps for 390x to 39fps. Going from guru3d to Canucks. One would've expected more recent drivers and updates to improve not decrease performance. Could it be that presentmon frames are lower than whatever method guru3d used for getting fps?

Without that drop the 390x would be within about 10~% of an overclocked gtx1080 comparing between reviews scores, as pcworld shows gtx 1080 at 51 fps at 1440p crazy.

Hitman is an anomaly that works for either Nvidia or AMD depending upon which episode one uses and also section,
As an example here is an example showing the results skewing one way or the other:
http://www.pcgameshardware.de/Hitma...Episode-2-Test-Benchmarks-DirectX-12-1193618/
http://www.pcgameshardware.de/AMD-Radeon-Grafikkarte-255597/Specials/RX-480-Test-1199839/2/
Look at DX12 or DX11 for both and the performance of the 980ti relative to a FuryX - point is Hitman is too erratic as a consistent benchmark.

HardwareCanucks is crazy setting but uses PresentMon.
The others used Extreme and High because Crazy is an impractical setting for most when playing the game with a 480 (if not going by the PCGamer fps but other sites)
I have just checked pcgamer and they state:

So they are using Extreme like some of those I mentioned earlier, something is strange IMO about PCGamer in this specific game.
Cheers

Still even in the 1080 gtx, where it would be practical, iirc, crazy was absent from many reviewers, even ashes itself seems to be missing from quite a few.

The PCgamer score at extreme setting is interesting, because it is within 1fps of a 980ti, the same level of difference as was seen between the 390x vs 980ti at high settings in guru3d's february bench.

But it is strange as you say.

Here's another site's Ashes benchmarks
http://www.pcworld.com/article/3071...badass-graphics-card-ever-created.html?page=8

The 390x has about 26-32% advantage over basic 980 in pcworld's bench, which is why it surprised me the 480 < 980 in canucks site. The 390x isn't 26-32% faster than the 480, at least judging from the benches in today's reviews.

 

[AlNom]

Huh? How do you get 2880 from that?

He's counting the "tex" blocks as the "shaders".

 

[Rys]

It's also a complete work of fiction.

 

[el etro]

Maybe Polaris has a close cousin in GCN Neo.

Really looks. Matches with another rumors/infos present on the webs. Especially with the one Damien(hardware.fr) saying RX480 not having all the power saving features the Polaris family haves.

 

[Love_In_Rio]

One thing, this chip was designed at the Shangai team?. If so, what happened to the old Santa Clara and Marlborough architect teams?. Were they disbanded ?.

 

[Frenetic Pony]

Welp, the RX 480 is just as disappointing in terms of power and mm^2 scaling as I feared, and for sales it doesn't matter for shit. Even with numerous reports of vast stock Newegg is entirely sold out on day 1. To be fair to AMD, Nvidia has been harnessing the power of PR for the last several years straight, it's about time AMD has caught up.

But I wonder what the hell went wrong? The scaling, in terms of TDP and frequency for the move to Finfet is incredibly poor with this first release for AMD. Not too mention their "15% efficiency gain" in architecture is over GCN 1.1 instead of 1.2 which is just... what? Ah well, at least competition will survive and thrive.