[Rumor] RX480 AIB Card Leaked and Tested!

[mrthanhnguyen]

Why project car, doom are not a good benchmark?

I call it head on head. Nothing is better than the other. Just pick your side.

 

[Eroticus]

That makes my head hurt, lol. Coming from the nv world, it doesn't make sense to me.

16.15 and 16.20 = Driver version

16.3.2 and 16.6.2 = Software Version

Why project car, doom are not a good benchmark?

Project Cars has GameNotWorks

In Doom amd had problems with OpenGL 4.3

 

[medi01]

Custom ones are nearby now and will solve any of your problems with RX480 (except for the consumption I guess )

I hope so, from the looks of it they should.
Higher temperature also leads to higher power consumption, by the way, e.g. Fury X vs Fury perf/watt, so it could address even that to a point.

 

[mrthanhnguyen]

390 is head on head with 970 on DOOM 2016. AMD took 3-5 days to release a hotfix.

 

[Eroticus]

390 is head on head with 970 on DOOM 2016. AMD took 3-5 days to release a hotfix.

Thanks, I finished the game by then and I'm still waiting for Vulcan API.

"Shortly After Launch"

 

[okidna]

Why project car, doom are not a good benchmark?

Some say that Project Cars is GameWorks title, but apparently it isn't : http://forum.projectcarsgame.com/sh...1f64b4cedd283b&p=923733&viewfull=1#post923733

Even the PhysX is running on CPU (not GPU PhysX) so AMD cards shouldn't have any problem with it. Reddit user already proved that PhysX is running on CPU : https://www.reddit.com/r/pcgaming/comments/36aoir/project_cars_is_physx_really_running_on_gpu_a/

 

[Eroticus]

RX 480 > 1350

 

[Ungari]

So reference 4GB like this one can easily be "upgraded" to 8GB by a simple BIOS flash

http://www.newegg.com/Product/Product.aspx?Item=N82E16814202222&ignorebbr=1

I think the article is talking about memory speeds.

 

[Tatty_Two]

In order for those benches to have a fair playing field you need to have every card shown there at max overclock if we are to get a true picture against a 480 at max overclock, personally as 90% of consumers never overclock their cards it's probably best to keep all at stock though.

 

[xorbe]

I seriously doubt they are selling a dormant 4GB on retail cards ... though that would be like xmas!

 

[Eroticus]

I seriously doubt they are selling a dormant 4GB on retail cards ... though that would be like xmas!

4GB worth just 40$ - this could be happen and could not =P.

390 > 390X
290 > 290x

 

[the54thvoid]

One would expect techsite forum user would recognize which side shows Guru3Ds benches. And if knowing that you can't figure who is biased, you probably need to think a bit more.

Yes, of course i can recognise that. It still doesn't explain what you mean by bias. Who was biased? Guru3D or TOT. If you cant answer a question, don't be such a fucking smart arse.

EDIT: especially considering how balls out cryptic you can be.

 

[EarthDog]

Since it is just all rumor, here is another one:

View attachment 75429

This graph looks a bit off now, doesn't it?

 

[Eroticus]

This graph looks a bit off now, doesn't it?

Heh, Yep...

 

[the54thvoid]

I'm out of this thread..

Seems pointless now it's no longer rumour. Now we need a Vega rumour and leaks thread instead.

 

[HumanSmoke]

No idea, @HumanSmoke is usually good with that sort of info. Perhaps TSMC were a known quantity and moving to a new process with Samsung was too risky? Could even be cost as you mention to maximise profits.

I think the RX 480 review might provide the answers. The 16nmFF+ is offering 70+% better per efficiency (perf/watt) over the 14nmLPP RX 480, and significantly higher clocks (a clock for clock power requirement would be quite illuminating). I think that the Vega hype train might need to take stock of the capabilities of the process node.

 

[Eroticus]

I think the RX 480 review might provide the answers. The 16nmFF+ is offering 70+% better per efficiency (perf/watt) over the 14nmLPP RX 480, and significantly higher clocks (a clock for clock power requirement would be quite illuminating). I think that the Vega hype train might need to take stock of the capabilities of the process node.

Comparing GPU clocks is useless like comparing CPU clocks.

28nm vs 28nm
Same story like every year, AMD is offering Performance per dollar and Nvidia Performance per watt.

 

[HumanSmoke]

Comparing GPU clocks is useless like comparing CPU clocks.

28nm vs 28nm
Same story like every year, AMD is offering Performance per dollar and Nvidia Performance per watt.

Well done on completely missing the point. I'm talking about process node comparison, not some perf/$ metric that doesn't have any direct correlation with process node...unlike clock speed and efficiency.

 

[medi01]

Has this been posted? OCing with... "custom cooler" (25$):

http://oc.jagatreview.com/2016/06/t...on-rx480-ke-1-4ghz-dengan-cooler-3rd-party/2/

Out of 4 cards, 1 achieved 1450Mhz, others merely went over 13xx.

Elsewhere there are also bizarre results such as undervolting allowing to OC more.
Memory looks like the only bit that reliably OCes well.

The 16nmFF+ is offering 70+% better per efficiency (perf/watt) over the 14nmLPP RX 480, and significantly higher clocks (a clock for clock power requirement would be quite illuminating).

What facts is this conclusion based on?

I think the article is talking about memory speeds.

No, it's not.
But carefully, that seems to be true ("all cards have 8Gb, even those sold as 4Gb") only for review samples.

 

[xorbe]

Elsewhere there are also bizarre results such as undervolting allowing to OC more

This is the crap we put up with on the nVidia side when og Titan came out that required a custom flash to add power headroom. "Oh you raised the voltage? I'm gonna cut MHz to keep the power level the same!!!" Which lead to people lowering vram speed to gain gpu speed.

 

[ikeke]

Could be as simple as all cards being overvolted for safety margin. Same thing as with Hawaii, where, if you reduced power limit you got -75W off the power consumption, -15C off temperatures and card not throttling again.

See this for reference
http://techreport.com/forums/viewtopic.php?f=3&t=109936

 

[TRWOV]

Elsewhere there are also bizarre results such as undervolting allowing to OC more.
Memory looks like the only bit that reliably OCes well.

I experienced that with my HD3850 AGP. Undervolted to 0.95v and managed to clock it to 824Mhz. At stock voltages I could OC to 798Mhz tops. Temps went down too.

It seems that AMD prefers to play safe with voltages. I've also run FX 83xx CPUs undervolted with no issue, I managed to get an 8320 running on a 95w board that way.

 

[RejZoR]

Well, that's not "bizarre". Cards have pre-set parameters for thermals and power consumption.

Increase voltage and you'll be hitting thermal limits. -> Cool it better to avoid thermal throttling. -> You start hitting power limit. -> Increase power limit. -> You'll start hitting thermal limit again. Or a noise limit that you can still bear with.

This is how modern overclocking of graphic cards goes. The thing is, today, graphic card protects itself with all these safeguards. In the past, you had to figure out which part of it you're stressing too much. Which kinda makes it easier to work with today, if we exclude the fact you usually have to fiddle with BIOS to really properly tweak it.

 

[Dethroy]

This graph looks a bit off now, doesn't it?

Hmm, not seeing anything awfully wrong with it... Would you care to elaborate?

 

[medi01]

480 performance analysis, based on Guru3D benchmark results, probably somewhat biased in the set of games chosen (I'd also wish "within 5fps" to rather be a % metric, but anyhow all numbers are visible):