Some RX 480 4GB stuff leaking out.1080Mhz Clock

^ Read his resume. Think about what he wrote. Was he limited by NDA?

And where he mentions Polaris? Sorry, he does not. But you can put it there as you wish. I do not expect anything less from You.

 

I know logic isn't your forte, but if Greenland aka Vega is the 'leading project of graphic ip V9.0', it necessarily implies that Polaris is not using 9.0

If I am in the lead, then there is nobody ahead of me.

 

Use bold and underlined in conjunction with very table you posted as undeniable proof.

Apply logic between Changes in GPU and numbering between 7.0 and 8.0. You'll find there some magical numbers like "7.1".
Wow? There are numbers between 7.0 and 8.0...

Now apply what you do not know to scheme. You do not have 1st party information on revision Polaris uses, and from prior knowledge it can as well be 8.1.
Because AMD with their lovely slide showing "New" stickers on building blocks did put "New" on "4th generation GCN CU" as well as other things.

And it still stands, that: "You Do not Know." And "You push your unconfirmable beliefs as facts."

 

It could be 8.1, it could be 8.25. It could be a transcendental number lol, Vega is the leading chip with gfx ip 9, so Polaris is necessarily using 8.0 or a minor revision of it. Happy ? Either way,in this respect Polaris is more similar to Fiji than Vega will be.

Edit:

The gfx IP block contains more than just the CUs though, I forget how amd calls their stuff, but the gcn equivalent of gpc function blocks would have to be included under gfx ip, so improved raster engine, primitive discard, all these things could contribute to changing ipblock revision, depending of course whether they just to change the revision or not. Depends on where they consider these functions in terms of ipblock naming.

I'm not pushing anything as fact, I just posted this table and the LinkedIn stuff and you harangue me as usual. Your saying that the ip block table is wrong is actually you pushing your beliefs as fact xD

 

Stop fighting about a thing you wont be agreeing upon. Just agree to disagree. Spamming the thread "but I am right" and slowly escalating out of hand. Such a fruitless fight.

 

Its annoying! I just said I would post the table later and all of a sudden 'you said Fiji and Tonga are the same! You said x! You said Y!'

I was just gonna post that table, to point out that Vega could be very different from Polaris. I wasn't trying to convince anyone of it, just the info available to me, that I had been looking for for a while

. The actual thing I wanted to get to before fox derailed me is that this is compatible with the rumor that amd was employing custom cell libraries for Vega. Even if implementing identical function blocks, changing cell libraries would warrant changing ipblock revision entirely. This means we could see Vega silicon displaying very different characteristics comparing to Polaris. It could clock higher it could be more efficient etc

 

How can you agree or disagree on something we do not know?
I do not know what AMD changed. And I am not making stories one way or another.

And that's exactly what should everyone else do. Theory crafting? All the way, but based on what we know. And where we have knowledge holes we should not cling to things which suit our ideas.
Instead acceptance of unknown is right way.
- - - -
Take situation. I'll present you with 3 doors.
Door on left have sweet lady behind.
Door on right have hungry and horny bear.
What's behind middle door? Not one person ever heard from me and not one person ever opened that door.

Then comes the 'Hero' and presents you with detailed information on what's behind middle door even without ever getting anywhere near them.
But that information is just construct of Hero's imagination. And should be taken as such.

 

Right. Now we're talking about bear rape.

 

Yes, fitting analogy for your posting since you are our J.K. Rowling of technology.

 

Are you fighting about what exactly? That Polaris won't be using the "packed libraries" or whatever? It's not only thing that matters, keep it in mind. AMD keeps claiming up to 2.8x performance/watt, and they actually give a figure of 1.7x because of the 28nm to 14nm transition. That leaves a ton of space for a lot of innovation. One could argue that architecturally Polaris is more different than Fiji, than Pascal is to Polaris.

That doesn't mean that Aelders is wrong about the packed libraries, that also doesn't make him right when he sounds like the damn thing is the same in general.

 

One could argue anything really, I'm really not saying they're the same. I'm just saying they're the same architecture revision, which they are. My point was that this is compatible with previous rumor (can't find source now either) that they're using custom cell libraries. It's a positive thing, potentially.

I'm offended by that fox. Jk Rowling sucks ass, but what does it have to do with your hungry, horny bear xD

Btw AMD's 2.8x claim is specific to some scenarios, much like nvidia ''s 3x claim was specific to VR. I imagine this one is specific to rasterizer limited situations, compared to gcn 1. 1/1.2

 

So it possibly can do 1.5Ghz like I suspected, Im pretty sure this will reach near/around stock 980Ti - GTX1070.

 

1.5ghz from 1266 is around 20% oc. If we assume linear performance scaling that's 20% improved performance.

If the stock 480 is 85% the perf of a stock 980ti then yes, assuming linear scaling.

Power is then the question stock 480 is expected to consume around 120w, +25% and we're at 150w,thats assuming no voltage increase. Perhaps custom boards will actually present an improvement in this case

 

According to that 3dmark score it's short of 980 performance. 20% oc won't get you 980ti performance.

 

Videocardz don't redo their tests on new drivers btw those results are based on whenever they last tested all those cards

 

Fair, but 13.xxx score for the stock 980 is about spot on and 10.xxx for stock 970 too. So the 480 sitting in between the two means it'll perform somewhere between the 970 and 980. Seems to be sitting right in the middle of these two.

No matter if you hit 1500 on the core of the 480, it won't go from 11,893 to stock 980ti graphics score (around 17k). No chance.

If by any magic it did, AMD would suddenly have to start selling their entire range (including FuryX, Fury, 390/390x) for sub $200. A 390 would probably have to cost $150. No chance.

 

I don't believe that any custom boards will come with the paltry 6-pin. If I get one of those (my pocket doesn't allow for much higher, really), I'll make sure it's at least an 8-pin. On the other hand, the card might actually perform around 100W (which is the rumor I've seen somewhere, can't remember where), and the single 6-pin might prove to be aplenty.

 

To match a stock 980ti it needs to hit 1466mhz, this is just considering raw shader throughput, it could very well hit that in OC for all we know, but what I'm interesting in is seeing firstly how it performs in games/scenes in which GCN typically underperformed due to geo limits, and second if it is bandwidth starved or not.

I believe we can agree that 1266mhz is the boost clock on the 480, but let's assume it isn't, let's assume it's 1080mhz.

The performance disparity between the 480 and 390X, by AMD's own 480 number's in that convoluted mgpu benchmark at the event, does not correlate linearly with the shader throughput disparity;
1080mhz; that's 5 tflops.

The 390x would have 18% higher throughput.

The performance disparity is 24.7%.

You can go see for yourself on AotS benchmark browsers, 390X scores start around the 41fps mark. It's odd

Naturally, since I was bashed to hell when I said this, and being the immense ******* that I am, I asked 390X users to test 1080p crazy 8xmsaa for me on a system similar to the one used in the AMD demo to boot. Hehe. Ieldra, the eternal *******.

Stock clocks! Naturally

http://www.ashesofthesingularity.co...-details/95a7ccd7-73a1-4749-969e-a8d3058e7df0

So someone has some splainin' to do...

At 1266mhz it should be basically on par with a 390x, it ain't, it's 20% slower

Edit: nope not the same system, that was another guy, but he had cpu at stock as well, if I find the 5820K guy's benchmark I'll post it here as well.

 

Wait, are you saying that a 200mhz overclock will take it from 11.8k score to 17k score?

 

That's a good one.

You're saying that the company that hyped FuryX as "overclockers dream", then benched it as faster than 980Ti, and mislead everyone by "several months ahead of competition", is "is under microscope"?
And is suddenly so modest and conservative to purposely report lower performance on 480 than they're getting.

Meanwhile the fact that 1080 is not overclocking to 2.1GHz in every single game/workload, is getting away with murder.
Nevermind that NV had the official slide saying we should expect "over 2GHz overclocking".
NVM that "throttling" is still several hundred MHz above the base clock and that "throttling" is GPU Boost doing exactly what it's supposed to in order to give maximum performance/clocks.

:bang: