Discussion in 'Videocards - AMD Radeon' started by gerardfraser, Jun 11, 2016.
Similar scores here. 10156 fire strike and 12521 graphics score.
So it does look like it will be pretty much way more efficient R9 390. Overclocking potential is still a mystery though. Those could fakes though.
For comparison my GTX 970 @ 1,6 GHz.
I can reach 10k graphics score on 280X, so what? Does that mean im close to rx480 12k performance, no. It will probably murder my card in every possible way.
Let's not compare performance just yet.
I can also get 10k in 3D Mark, I'm, not sure what's the value of that from a point on. Usually it's a good indicator, don't get me wrong. The only thing said about Polaris CFX from official lips was that it's bad at the moment, keep that in mind too.
I could get 10k+ on a single 7950 too. I find 3dmark graphics score to be pretty accurate when it comes to judging performance. A stock 480 sitting between a stock 970 and 980 seems on the money.
Nobody is expecting more from it, surely?
Wccftech is now claiming 1.5ghz+
Btw I finally found that table detailing ipblock revisions of various gcn gpus
Polaris and Fiji share graphics block revision. Vega will use a revised one, I'll post it when I'm home
Did any of u guys saying bad things about the rx 480 6.3 steam vr result know that the driver used for that benchmark was not launch driver, and we all know that a gpu company would not be able to optimize a gpu for a benchmark thats not so important, also do u guys know that the steam vr benchmark is highly dependent upon gpu clock speed and i think this benchmark is performed on lower clock speeds.
I read somewhere that this benchmark increases points substantially relative to the clock speed increase. So maybe by a 100 mhz clock overclock this benchmark gets to 8.0 scores then? But all is to be seen when gpu launches, then drivers will come and well see the scores.
AMD published the SteamVR results themselves. I would expect it is in their best interest to cast their product in the best light possible. They could have been using outdated drivers from 1995 for all I care, it's official PR.
3Dmark score for single GPU lower than i thought honestly.
We all know about AMD PR, and the blunders they make, so i would not believe any of this unless i get a card in my own hands or i would just get to see a reviewer review it honestly with stable drivers.
Yeah, I'm sure AMD were trying to make the gpu look bad.
Last thing AMD wants is to have official score 0.1 point higher than what someone out there will benchmark on launch date.
You know, some companies may get with claim of 2.1 standard Air OC with 80°C while in reality GPU throttles and heats up much higher.
But AMD is under microscope of everyone who loves to throw...
I think it is in their best interest to say that they are bringing great, magnificent, wonderful, ... product. And let reviewers to show numbers which matter.
And that's what they are basically doing. All numbers they gave are not telling anything about final performance. They did not mention OC at all.
- - - -
Ieldra, btw. It was you who claimed that Tonga is exactly same as Fiji even through it is known that they have different revision of UVD and power measurement, management, gating.
And that Polaris and Vega is again one and same.
So, Now it is Fiji and Polaris one and same? And Vega is new thing? If you even remotely knew how AMD develops those parts of GPU and puts them into final design, you would not try to put out random stuff like that in front of anyone.
Bring your picture of unreleased Polaris where practically everything is under NDA.
Have your moment of: "I found a fake image. But it is true to me since I like it."
But they had no qualms misrepresenting the 1080's performance in AotS Go look at AotS benchmark browser, their claimed 58 fps on the 1080 is low. Results start at 61+
They had no qualms bull****ting about 50% utilization.
They had no qualms claiming 2 480s are more efficient when a single 1080 us 180W, two 480s probably 250w range.
This is very inconsistent Señor Fox
I claimed Tonga was exactly the same as Fiji ? Quote me on that !
I'm pretty sure I said they are using the same architecture; same GCN revision.
I still stand by that claim. Tonga and Fiji are using the same architecture.
Polaris and Vega are using the same architecture.
Power measurement and gating aren't ipblocks afaik, no idea what you're on about. Management isn't either unless you're referring to what AMD calls SMC
Just because ipblock revision changed does not make it a new uarch. Otherwise you would consider most dies to be an independent architecture which they are not.
Changing ipblocks does not necessarily mean they are functionally different.
Are you saying they are one and the same ? I don't understand. You think because they share ipblock they are one and the same? Your problem not mine !
Seriously. I "put out random stuff like that in front of anyone" ? The charcoal is calling the kettle black.
If AMD were to change their cell libraries, they would have to essentially rewrite their IP using the new libraries, despite the ip blocks being functionally identical, the implementation at RTL may differ, at netlist level they may use different cell libraries.
It's not as simplistic as you make it out to be Señor Fox
What is same and what differs... What's reliable source of those claims of yours?
Heh Can't quote me can you, cause I never said that.
Which claims do you want a source for exactly. I've made no claims in the post you are quoting.
Unless you mean the claim that Fiji and Tonga are the same architecture.
Technically when I say that gravitational force is proportional to mass I am making a claim, you can hardly expect me to source that, because at that point I would have to source two million other things.
However, I will humor you this one time, for science.
Here's the thing btw
GMC - graphics mem controller
ih- interrupt handling
smc- sys management controller
dce - display controller
gfx - guess!
sdma - DMA controller
vce/uvd - video encoder/decoder
By your logic every single one of the dies mentioned in this table is a different architecture. Or maybe you only consider GPUs with differing graphics block revisions as different uarchs, so Polaris and Fiji are the same uarch ? Your words! Not mine !
That's fine, you can say that, but then the word 'architecture' no longer means what it used to.
So let's pick a new word to replace it ! How about 'functional framework' or 'superstructure ' ?
I have better idea, let's get to source. You downloaded and reuploded image from WCCF.
Original is Here: http://pctforum.tyden.cz/viewtopic.php?f=9&t=233219&start=357
Do you get what he wrote there and what follows? I do not think you do.
His source is Carrizo Linux AMDGPU (+93k lines of code) driver update from February which did contain some parts for Polaris (+63k lines of code iirc). That code is widely considered as Work In Progress, even mentioned there (WiP).
AMD pushed what they had to have it at least somewhat ready environment.
Yes, it was freaking Carrizo driver update for linux just half a year after Carrizo have been widely available.
Was it final code which supports all there is to Carrizo? Unlikely.
Was it final code which supports all there is to Polaris? Definitely not.
And yes, I did it intentionally, because I knew exactly what picture you are going to post.
So ? Of course I don't understand what's written in the actual forum, I don't speak czech, you do.
You want to contend that this table of ipblock revision is total trash! Sure, it could be 100% wrong. I have no idea why it would be wrong, it's not like this is in WIP code, this is a table of f***cking ipblock revisions.
Either way, this was not your initial contention, you're now arguing a different point from the previous post, which itself was a different point from the one before. This is just arguing for the sake of arguing imo.
Anyway, I posted what I wanted to post, and I replied to your questions fox.
I had been looking for this table for a few months, so satisfying that I found it again
You did it intentionally ? Did what ? You did nothing!
That table is made from information someone read from driver source code. Be it comments or functions. Does not matter. It is not 1st party information. It is not final code. And there is not guarantee that even freaking comments were updated properly.
Have you ever seen source code for some large project? Or specifically linux? It is lovely, full of sh*t & F* and outdated code & comments.
And good that you are happy to find your lost buddy after months. It took me like 30 seconds to get to source.
It is 1st party information, unless AMD outsources it's drivers.
You also conveniently ignored these
Let me guess, outdated linkedin profiles ?
It's also full of updated code and updated comments. I really don't see a valid argument for why this would be wrong in AMD's own driver, it sounds like you're saying "it could be wrong". Sure. It could also be right, but I digress. This wasn't what we were discussing a few minutes ago.
Heh, what did you search ? I kept trying 'gcn ipblock table' and variations on that and got nothing
Maybe you searched in czech, or used reverse image search. Anyway, no need to be testy Señor Fox. You are the undisputed GCN ipblock table search champion, nobody can find it faster than you! You deserve a medal.