AMD Polaris Thread: Radeon RX 480, RX 470 & RX 460 launching June 29th

[Stuka87]

Hi all,

My name is Jan Stach, and I am editor in chief - www.DDWorld.cz.

Charts with the performance of the RX 480 are 100% FAKE! Yes photographs RX 480 card are our (I photographed them). But the performance graphs are modified by someone else not from DDWorld.cz. It's not our charts or our data because we have not yet completed all the tests RX 480!

And most importantly: DDWorld.cz respects the NDA.

Someone abused our charts and modified them = 100% fake ... and with bad data .

Our tests and reviews will be done with the end of NDA. And all of you are welcome to find out how good (or bad) RX 480 is.

Sincerely, Jan "DD" Stach (editor in chief www.DDWorld.cz)

Thank you for joining up to let us know!

 

[zinfamous]

That was clear from the first second, who whould be dumb enoght to underclock a 480 for benchs?

I was wondering about that, but as mentioned before, the leaks were showing differently-clocked reference cards and it had been suggested that it was part of AMDs NDA-tracking to distribute variously-clocked cards...so that's what I assumed was going on.

EDIT: wow, nice of EiC to catch these and discredit that bogus stuff from being attached to his content. :thumbsup:

Must be a headache, though.

 

[AnandThenMan]

For crying out loud, they managed to give a $650 780 a Gold Award and that was one of the worst videocards made in the last 5 years.

Author: Brent Justice

Isn't he the one that will be doing the RX480 review? ...oh and this -

Compared to the XFX Radeon HD 7950 Black Edition the ASUS GeForce GTX 660 Ti DirectCU II TOP provided a better gameplay experience at 1080p in every game except for The Witcher 2.

Imagine pitting the 7950 against 660 Ti now.

 

[tonyfreak215]

Author: Brent Justice

Isn't he the one that will be doing the RX480 review? ...oh and this -

Imagine pitting the 7950 against 660 Ti now.

That would be rather fun to see.

 

[dooon]

I want XFX Black.

 

[maddie]

Just because a card can draw up to 150W doesn't mean that it will always do so. AMD has said that they have 2x+ perf/watt with Polaris so if it has about the same performance as a 390/X, it makes perfect sense that it should draw at most half the power.

What I'm saying is that the 480 could already be using up almost all of its headroom in terms of clock speed for a given voltage as well as how much voltage the card can handle. For example, someone could build a board with 2x 8-pin power connectors, but it's highly unlikely that you could supply 375W of power to Polaris without damaging the chip.

If both of those are true (or close enough to it) it would explain why there are rumors that are stating that it only draws about 110W but it isn't overclocking very well. Obviously either or both of those could be false, but if you assume that both are true (or have some truth to them) then you need an explanation for why both of those rumors would be true and the only way that works is if you assume that most of the Polaris chips (at least the ones being used in reference card designs at any rate) are at their limits or are being artificially limited through bios settings that amount to the same thing.

My stated reason [several posts now] for the low stock overclock is simply that AMD has reduced the traditional needed overvolt margin [to ensure no low voltage dips when a jump in power is needed, eg many shaders starting at the same time] to much less than previous designs by using their adaptive clocking technique on a much larger scale. This reduces any spare voltage and ensures lower power consumed.

It's almost certain that this is one of the major power saving techniques in Polaris, and if true, would mean there is little voltage margin for overclocking without adjusting core voltage.

RWT has a good read on it. http://www.realworldtech.com/steamroller-clocking/

 

[Digitaldoom]

EDIT: wow, nice of EiC to catch these and discredit that bogus stuff from being attached to his content. :thumbsup:

Must be a headache, though.

Yes, in my country (Czech Rep) is very bad (read: disaster) situations on the field of IT journalism and magazines. Most high-quality editorial ended, the majority of high-quality magazines as well. What remains is a joke .... (not all, unfortunately, some major HWmagazines yes). Many "jurnalism" are just "Flamers".

DDWorld.cz is the largest independent (non publisher in the back, or giant paid editors) PC technology site in CZ, just a bunch of enthusiasts for PC and technologies . At present, we are almost the last magazine in the Czech Republic, where we are still testing hardware in the Czech Republic and we give space to all. We do the same amount work as a big professional magazine but with the budget and the people of a small blog. We have the support of many leading manufacturers and companies in the IT, but we have also "enemies", undermining our feet past 10 years.

This is probably another way. Small DDWorld.cz is getting new things to test, often the first in the Czech Republic. And that bothers some.

I already informed representative of AMD about this situation.... if these FAKE graphs appear somewhere, AMD knows that they are not from us and it is FAKE.

Sincerely, John "DD" Stach.

 

[Thala]

It's almost certain that this is one of the major power saving techniques in Polaris, and if true, would mean there is little voltage margin for overclocking without adjusting core voltage.

Indeed, adaptive clocking allows you chose a smaller distance between Vdd and Vmin. It gets interesting, when you consider that Vmin is different for fast and slow corner chips.

 

[Mopetar]

It's almost certain that this is one of the major power saving techniques in Polaris, and if true, would mean there is little voltage margin for overclocking without adjusting core voltage.

RWT has a good read on it. http://www.realworldtech.com/steamroller-clocking/

Thanks for the link.

 

[Elixer]

Radeon RX 480 Live Stream with Raja Koduri and Ryan Shrout
10:30am PT / 1:30pm ET - June 29th
http://pcper.com/live

No AT live stream?

 

[selni]

I am honestly shocked how can someone who spent a decade(s) on building his site's reputation online, building a solid community, and gaining trust of consumers/gamers throw it all away in such a short period of time. Even from a business point of view, unless he was delusional that NV will forever have 80%+ market share, he should have thought ahead that should AMD actually return to historical market share, he will forever lose the trust of objective/brand agnostic and AMD-loyal gamers.

How can someone be considered an objective professional after giving a 960 2GB a Gold Award in April of 2015 (!) against a 280X, but then go on to bash and downplay P10 and AMD's new launch strategy?

They guy should literally recommend every AMD GPU for the next 2-3 years after the horrible advice he provided in recent years. Today a 280X is 23% faster than a 960 2GB at 1080p.



Let's not forget, he also gave Gold Awards to the garbage GTX660/660Ti series when HD7870/7950 by far outlasted them and beat them in modern games anyway.

If you start looking back at it, HardOCP has not been objective at all since Kepler generation in 2012.

For crying out loud, they managed to give a $650 780 a Gold Award and that was one of the worst videocards made in the last 5 years.

Yeah... I actually like HardOCP and think they do a good job most of the time but have no idea what they're doing with awards, never made much sense to me. Even recently it's clear that they weren't happy with the FE mess but those cards got gold awards... sort of justifiable due to how far they are ahead, but when your advice is "don't buy this, wait for custom versions"...?

 

[Silverforce11]

For the people criticizing [H] for giving a gold award for some of the Kepler cards...

They do it because it's performance was outstanding at the time compared to AMD.

The 660 vs 7950 for example, was on-par back then.

780 had no competition, it was 20-30% faster than a 7970.

Hardware reviewers can only go by what the results say. Sure, they can speculate about future proofing, but who knew at the time how fast and hard Kepler would fall off the cliff? 280X (7970 rebadge) often ~= 780 in games since late 2014.

The onus is on AMD to provide competitive performance when it matters, at the launch. Not everyone understands the importance of all GCN consoles and what it means for the gaming industry, especially the masses who buys GPUs, so AMD cannot rely on the "future proofed" architecture to sell. They've showed signs of learning that lesson, the new features in Crimson are an excellent start. AMD needs to keep pushing, newly released games should run great on day 1, not after a week or 2 weeks.

Actions speak louder than their lame marketing #BetterRed or #RadeonRebellion campaign.

 

[AnandThenMan]

For the people criticizing [H] for giving a gold award for some of the Kepler cards...

They do it because it's performance was outstanding at the time compared to AMD..

And then completely ignored how much better AMD's architecture aged going forward. Never hear about it from them even to this day.

 

[Silverforce11]

And then completely ignored how much better AMD's architecture aged going forward. Never hear about it from them even to this day.

When you get given new and latest gen hardware for free... let's just say they won't appreciate future proofing or respecting a gamer's investment, it's not something on their radar.

I've sold my 290X awhile ago for a good price, been gaming on an old 7950 from a spare rig. It still delivers a great gaming experience at 1080p in the latest titles on high settings & maxed texture quality.

 

[daxzy]

The onus is on AMD to provide competitive performance when it matters, at the launch. Not everyone understands the importance of all GCN consoles and what it means for the gaming industry, especially the masses who buys GPUs, so AMD cannot rely on the "future proofed" architecture to sell.

I think AMD is doing an awesome job in terms of providing competitive performance. My Radeon 7950 was near top tier when it came out in early 2012. And through subsequent driver improvements, it still comes within 90% of a GeForce GTX 960.

I can't say the same about people who bought Kepler's or Fermi's. So as it stands, its not unreasonable for me to assume AMD will continue to have better value for folks like me (and most my friends) that upgrade every few years. I still recommend Nvidia to people that like to burn money on computer hardware yearly or for current generation laptops (I own a GTX 960M laptop).

 

[maddie]

Indeed, adaptive clocking allows you chose a smaller distance between Vdd and Vmin. It gets interesting, when you consider that Vmin is different for fast and slow corner chips.

Interesting and more complicated? Yes, but I'm not seeing the problem.

Traditional values of Vdd & Vmin ratios would still have had to contend with fast & slow corner chips. All this does it lower the spread of the voltages and at the same time you make sure that you don't go below the Vmin of the slow corner case. I suppose for a very variable process, you might not be able to make it work.

Another possible solution might be to break the chip into many, lesser variable domains, each with its own adaptive clock circuitry. Just thinking out loud.

If you combine it with a boost tech like what nvidia has, then it gets very interesting.

I really hope someone has a deep look at the design on the 29th or soon after.

 

[Silverforce11]

I think AMD is doing an awesome job in terms of providing competitive performance. My Radeon 7950 was near top tier when it came out in early 2012. And through subsequent driver improvements, it still comes within 90% of a GeForce GTX 960.

I can't say the same about people who bought Kepler's or Fermi's. So as it stands, its not unreasonable for me to assume AMD will continue to have better value for folks like me (and most my friends) that upgrade every few years. I still recommend Nvidia to people that like to burn money on computer hardware yearly or for current generation laptops (I own a GTX 960M laptop).

If you can get 1.1ghz on your 7950 (it should do it no sweat), it's a 7970Ghz or 280X, which is plenty faster than a 960. It's got a lot of legs left actually, I am pleasantly surprise, because I'm playing TW: Warhammer on it now and it's just delivering excellent performance.

My point though is AMD needs to work harder at day 1 game optimizations. The crap performance in Doom for a week was unacceptable for such a AAA title. Actions speak louder than words, so I don't buy their PR hype campaign.

 

[guskline]

Silverforce11, are you getting 1 or 2 RX 480s?

 

[Thala]

Traditional values of Vdd & Vmin ratios would still have had to contend with fast & slow corner chips. All this does it lower the spread of the voltages and at the same time you make sure that you don't go below the Vmin of the slow corner case. I suppose for a very variable process, you might not be able to make it work.

Right. I was saying that for a variable process you will gain additional power benefits if you not only making the frequency adaptive but also Vdd and with it the threshold at which your adaptive frequency scaling sets in. For instance you can bake this into the Fw/BIOS. This is particularly important for the fast corner chips, which not only have lower Vmin but also leaking like crazy (in particular at 14/16nm).

 

[Silverforce11]

Silverforce11, are you getting 1 or 2 RX 480s?

Maybe 20. I'm still deciding.

 

[maddie]

Maybe 20. I'm still deciding.

Thankfully I'm not in the Australian market.

Polaris only has to outsell Pascal by roughly 2.5:1 to generate equal revenue. Interesting times.

 

[maddie]

Right. I was saying that for a variable process you will gain additional power benefits if you not only making the frequency adaptive but also Vdd and with it the threshold at which your adaptive frequency scaling sets in. For instance you can bake this into the Fw/BIOS. This is particularly important for the fast corner chips, which not only have lower Vmin but also leaking like crazy (in particular at 14/16nm).

Agreed. I'm thinking this is one of the main reasons for the long period between working silicon and release. The complexity of the power saving and performance enhancing algorithms.

 

[YBS1]

Silverforce11, are you getting 1 or 2 RX 480s?

Providing the thing doesn't tank in actual reviews versus the hype, I'll be getting at least one 8GB model for my 6700K. I might scrap the 780s for one temporarily as well while I wait on Vega/Pascal.

 

[Triloby]

My assumption on the 480 is that, performance-wise, it'll be on the same level of the 390X and 980, or even slightly better than that. I don't think the 480 will come anywhere near close to reaching the 1070 in terms of performance, but I've seen crazier things happen before in the past.

 

[Mopetar]

Polaris only has to outsell Pascal by roughly 2.5:1 to generate equal revenue. Interesting times.

Revenue is kind of a useless metric considering there are many companies that have a huge revenue but barely any profit to show for it. Even though AMD is going to sell more cards over the next few months, Nvidia is going to make more money. Even without the FE markup, they'd take in more per card but with the FE pricing it's not even going to be close.