AMD RX 480 - First 3DMark results

There is no mistake, they wouldn't make a mistake such as that in a presentation slide, it would have been checked and rechecked multiple times.

Just keep this in mind, what could the potential bottleneck be for P10 compared to the r390 in VR, and then you will have your answer.
 
Wait really... They spent years designing it but come launch they forget a key aspect of its design and omit it from all marketing material. A mistake is a single employee saying differently than the rest. This was no mistake.

That's not marketing material, that's a leaked slide without the rest of the presentation to go along with it.
 
There is no mistake, they wouldn't make a mistake such as that in a presentation slide, it would have been checked and rechecked multiple times.

Just keep this in mind, what could the potential bottleneck be for P10 compared to the r390 in VR, and then you will have your answer.

Yeah but the 980 does better as well, at least when overclocked to comparable tflops, it has ~2/3 the bw(of 390x), it seems like a silly oversight, or a bad decision not to invest xtors in compression, imc efficiency if it will be a bottleneck. Then again, if it simply performs like a 980, even in games in which Hawaii performed in badly, it will be a success, if not then I will also be disappointed
 
Desktop 470 is p10 according to the link I posted earlier. 110w

Right, and Polaris 11 is also x480m (which is the chip that's in desktop 470, and the new ps4 and xbone from what I've read) - this is at least what the article states. Mobile GPU chips often are a step down from what their name would suggest.

AMD is going to be churning out a shitton of Polaris 11s it seems.
 
Right, and Polaris 11 is also x480m (which is the chip that's in desktop 470, and the new ps4 and xbone from what I've read) - this is at least what the article states. Mobile GPU chips often are a step down from what their name would suggest.

AMD is going to be churning out a shitton of Polaris 11s it seems.
Unless they are doing HBM style stacking of dies Polaris 11 isn't in consoles. Those are all custom SoCs with iGPUs.

rx480 and rx470 are both Polaris 10 with a 256-bit memory bus. rx460 and smaller mobile chips are Polaris 11.
 
Of course they shouldn't lie. But are you saying paper specs matter than actual measured performance?

Why would someone be pissed if the specs were wrong but actual performance is accurate?
Of course paper specs matter. If they didn't matter then they wouldn't bother publishing them in the first place.

A lot of us look at the architecture of a GPU in order to estimate performance compared to other cards and of upcoming gaming titles. If the paper specs are wrong then so are the estimates.
 
Of course paper specs matter. If they didn't matter then they wouldn't bother publishing them in the first place.

A lot of us look at the architecture of a GPU in order to estimate performance compared to other cards and of upcoming gaming titles. If the paper specs are wrong then so are the estimates.

Specs matter most during pre-launch silly season when we don't have actual performance numbers. Once we have those numbers specs are useless for estimating anything.

How accurate are paper specs in predicting the performance of Maxwell vs Kepler or GCN?

I think everyone here likes talking about specs but I'm not sure why you think they're so important once you can actually measure performance.
 
You guys must've never seen typos on packaging... Oh wait there are lots.

For fuck's sake, there's a typo on commemorative money bills.

MAS 'sorry' for typo on SG50 notes set packaging

n6j868l.jpg


So yeah...NVIDIA making a cock up in communicating between departments doesn't seem so unlikely now, does it?
 
Here's a leak.
The guy says in the thread that Ultra was done on old drivers and Extreme is done on the latest drivers. It's also at stock speeds.


For reference:
index.php
 
Last edited:
Seems to be about right and pretty much what I though. Much more efficient R9 390. The GPU is likely >70c, to be expected from ref cooler. Custom cards probably have rather nice temps.
 
The guy says in the thread that Ultra was done on old drivers and Extreme is done on the latest drivers. It's also at stock speeds.

Thanks, added this to my post. What thread btw ?
 
Last edited:
Thanks, added this to my post. What thread btw ?

I was looking for the thread link, but people say it was removed by the moderators right before he was going to start overclocking.

I should also add a correction, Extreme was done on both older and new drivers.
 
I'm not sure what to make of the reference versions that AMD are going to be releasing when they have pretty much given the AIB carte blanche to rag it up to full whack which could be anything upto 20%+ performance for not a lot more money.

Maybe the reference versions are targeting small factor system builders who want to keep the power in check, and the AIB releases to go for the enthusiast?

Will make the 10's of board variation reviews worth while, at least!
 
Yes and probably more in the future.

Reason - DX12 allows more cpu threads to be used, more assets, more draw calls - basically more of everything. Monitor resolution, VR etc. is increasing. All of this needs effective fast memory to work well. I see the biggest issue with AMD Fiji line is the 4gb amount so maybe AMD's first video card that will be very limited a couple of years after launch. Usually AMD video cards have a great life span in performance and usability. For example my HTPC has a 7970 (bios moded to ghz edition without issue) it still tears up most games with high settings at 1080p. My 290x will migrate to that machine on my next major upgrade and will probably last another 2 years before really upgrading the TV to a high dynamic range, higher resolution one. The the computer and everything in it will probably be tossed

Anyways if you are going to keepa video card for a few years or more - 8gb will probably work out well but will be limited probably in that time.
Fair point, but would be hitting the performance buffer before the larger vram will make any appreciable difference? We had this with the 390X with 8gb, but really in reality it would have been just fine with 4gb because you start hitting gpu limitations at the higher res.

Maybe with DX12 this may change, but for the target market for this card, and the lifespan (i expect this to be refreshed in a year with much better performing cards), 8gb seems overkill IMO.
 
I'm not sure what to make of the reference versions that AMD are going to be releasing when they have pretty much given the AIB carte blanche to rag it up to full whack which could be anything upto 20%+ performance for not a lot more money.

Maybe the reference versions are targeting small factor system builders who want to keep the power in check, and the AIB releases to go for the enthusiast?

Will make the 10's of board variation reviews worth while, at least!

Sounds like they are copying Maxwell. The GTX 970 reference clock is 1050/1178 (base/boost). My EVGA SSC is 1190/1342, a gain of 13.3%/13.9% respectively, and there are cards with higher overclocks.

The RX 480 is supposedly 1080/1266. Applying the same percentage as my SSC yields 1220/1441. So we'd have to see 1400+mhz boost clocks to see the 480 offer Maxwell type factory OCs.
 
I believe I scored just under 14K on my GTX 980 / 6700K Machine. This looks promising. I might have to make the switch to team Red.
 
Back
Top