RX 480 post mortem - the good, the bad, and the ugly.

[Stereodude]

I suspect we'll see on July 7. Kyle's been making noises that are consistent with [H]ardOCP already having a GTX 1060 in the lab for testing already (e.g. "not a phantom product"). And NVidia may allow GTX 1060 reviews to be posted a week ahead of availability, as they did for the GTX 1080, just to stick the knife into AMD and drink their milkshake that much sooner.

I think it does exist. Kyle possibly very subtly hinted that they have one, if I read the shades of the fonts a certain way on one of his posts).

EDIT: Damn it! Beaten to the punch by half a minute.

FWIW, I took Kyle's comment to be a reference to the Infinium Labs Phantom Console and the colorful history he and the [H] had with it, not a statement that [H] had a 1060 already.

 

[Algrim]

He most assuredly was reminding us of that. And I think, based on that statement and nVidia purportedly paper launching this in seven days with a hard launch in fourteen days, that it wasn't inconceivable that [H] already has a card or one is incoming very soon. Yes, it's speculation on my part but I'm not the only one (maybe only one of two) with that point of view.

 

[Pyroja]

The fan is purportedly extremely loud, approaching the infamous 290x, at 100%. Much louder than the 1070's blower at 100%. Also people saying it's a "grindy" noise.

How relevant is that if the fan never hits 100% unless you force it there? Every review states that under load the fan never gets loud enough to be a nuisance. Yes, you can force the fan to scream like a banshee, but so what? Should AMD relabel 50% as 100% so it's no longer an issue?

 

[Araxie]

Uglier story, the GTX 1060 have been teased since a month of so, but it remained quiet.. the fact that 3rd party manufacturers had already since a month ago in their labs for testing, just prove that card have been in production and testing since a even longer period of time. this is a card hinted to perform at GTX 980 levels and given the history of pascal with GTX 1070 and 1080 It would not be nothing strange that they are capable of reach that kind of performance at a much lower power consumption than AMD, so we are talking about a card that probably will be able to perform at GTX 980 OC'd levels which it's way way above the level of the RX 480 which barely reached GTX 970 levels of performance, what's worse it will have the Nvidia name so it will sell order of magnitude more than the RX 480 and will eat any possible market for AMD. want to believe or not?. just wait 2 weeks and see it, more than one will be surprised as the same way GTX 1070 and GTX 1080 surprised everyone.

 

[kalston]

I don't understand why this is a problem. Who runs their fans @ 100%? I have a 980 Ti with a very quiet EVGA ACX 2.0 cooler and I still max it out at 70%. And I game with headphones.

100% might make sense for benchmarking but you're talking about a significant noise increase on any cooler for what is typical a very marginal performance increase.

Well, that's you. But there are some like me who find that the 980 ti EVGA ACX 2.0 @50% is too loud even when wearing headphones (now I'm not saying it's horrible but I've heard better). This matters for video playback too - for example I use high quality uspcaling algorithms with madVR and my GPU usage is 90-100% (can get as hot as the most demanding games) for a lot of the content I watch. Headphones or not, noise becomes critical then since films always have some quiet parts (games too of course!).

The RX 480 actually could have been a pretty interesting HTPC card.

And TPU says the rx 480 is actually loud under load relative to other cards (as loud as much faster cards like the ref 980 ti and much lower than say, a gtx 970).

 

[Rizen]

Well, that's you. But there are some like me who find that the 980 ti EVGA ACX 2.0 @50% is too loud even when wearing headphones (now I'm not saying it's horrible but I've heard better). This matters for video playback too - for example I use high quality uspcaling algorithms with madVR and my GPU usage is 90-100% (can get as hot as the most demanding games) for a lot of the content I watch. Headphones or not, noise becomes critical then since films always have some quiet parts (games too of course!).

The RX 480 actually could have been a pretty interesting HTPC card.

And TPU says the rx 480 is actually loud under load relative to other cards (as loud as much faster cards like the ref 980 ti and much lower than say, a gtx 970).

There must be something wrong with your card. The ACX 2.0 is basically inaudible at 50% fan speed. I have to strain my ears to hear it from 3-4 feet away.

Realistically, if the noise level of an ACX 2.0 bothers you, you're definitely a niche case and not the norm.

 

[yourgrandma]

I have a acx 2.0 980 and no joke it's the quietest card I've ever owned.

 

[Pyroja]

There must be something wrong with your card. The ACX 2.0 is basically inaudible at 50% fan speed. I have to strain my ears to hear it from 3-4 feet away.

Realistically, if the noise level of an ACX 2.0 bothers you, you're definitely a niche case and not the norm.

I have an ACX 2.0 GTX 970. It's a damned quiet card.

 

[kalston]

Different standards that's all. Very quiet case, very quiet room etc. I'm the kind of guy lurking on silentpcreview.com I'm well aware that I'm not in the majority (never said that), but it does matter to some people more than others. And the reviews agree on the rx 480 : it's very much audible under load & hot, despite being a ~150w card.

 

[quiktake2009]

my 480 is getting returned. The card freuency keeps dropping to 300hz during fallout 4. The drama about power usage also has me a bit spooked

 

[Tup3x]

Muropaketti.com got probably the worst sample: it run at 90c without overclocking. They managed to get ~6 % OC when fan was running at 5000 RPM (77 dB) and the power usage was higher than GTX 1080. It look like GF's 14nm process is a real mess.

 

[lolfail9001]

Muropaketti.com got probably the worst sample: it run at 90c without overclocking. They managed to get ~6 % OC when fan was running at 5000 RPM (77 dB) and the power usage was higher than GTX 1080. It look like GF's 14nm process is a real mess.

According to The Stilt, Muropaketti actually got best GPU ASIC-score wise. Leakage concerns are real on this card.

 

[Semantics]

There's no compelling reason to want >4GB of RAM on a card with this level of performance. The 8GB model has its memory clocked higher, so that may matter; we don't know yet.

As for the rest, backplates are cosmetic fluff, the fan is only loud when you overclock (and you should buy an AIB card for that), and the power consumption stuff isn't yet confirmed as a real problem.

Next thing you're going to tell me is that RAM heatsinks actually heat up the chips not cool them!

Historically, power draw doesn't matter. Power draw causing a flop is just what fans of the brand with the better efficiency always spout. Whenever AMD had the power draw advantage, they'd spout this same line, and yet NV still maintained >50% market share. The majority of users are simply going to look at their budget and get the fastest card within that budget. Well, actually, no, that would be the majority of enlightened users. The vast majority of users will succumb to marketing

Fermi did sell, what is going to matter is price performance of 1060. That being said this performance and power doesn't look good for AMD competing at the enthusiast high end levels with nvidia. Fury x always tooked me as a card forced to competing with what Nvidia offered, doesn't OC it was pushed to it's limits just to meet what Nvidia already had, was already drawing all the power it could had, all the clocks it could.

 

[Crosshairs]

It is a shame that cards used to cost less at the higher end so that more folks could enjoy them but now, it is just lets make all we can off the PC gamer well not bothering to make the games fully optimized for the PC platform.

High end cards have always been in the $600.00 to $800.00 range... even 10 years ago the high end cards were $600.00 to $800.00. ..the only thing that has changed is you now get a whole lot more performance for your money....at least with Nvidia you do...I cant speak for AMD as I haven't used their products in a few years

 

[linkgoron]

Talking about the ugly:

AMD Radeon RX 480 im Test (Seite 12)

Looks like they undervolted the card using wattman. The undervolt reduced watt usage by 33, reduced noise by 3.5db and they also gained 5% in performance.

If every card can do such an undervolt (i.e. stay stable) AMD really dropped the ball.

 

[Chimpee]

Next thing you're going to tell me is that RAM heatsinks actually heat up the chips not cool them!


Fermi did sell, what is going to matter is price performance of 1060. That being said this performance and power doesn't look good for AMD competing at the enthusiast high end levels with nvidia. Fury x always tooked me as a card forced to competing with what Nvidia offered, doesn't OC it was pushed to it's limits just to meet what Nvidia already had, was already drawing all the power it could had, all the clocks it could.

I really think if 980Ti didn't show up, the Fury line up would have been a compelling buy for a lot of us.

 

[Presbytier]

I really think if 980Ti didn't show up, the Fury line up would have been a compelling buy for a lot of us.

Except it did come out, so it seems like an odd amount of speculation. Not picking on you, just not sure what point you are trying to make.

 

[Chimpee]

Except it did come out, so it seems like an odd amount of speculation. Not picking on you, just not sure what point you are trying to make.

Oh just replying to Semantics regarding the Fury X is a card forced to compete nVidia offering. Nothing 480 related, sorry if I went off topic.

 

[ChosenUno]

Talking about the ugly:

AMD Radeon RX 480 im Test (Seite 12)

Looks like they undervolted the card using wattman. The undervolt reduced watt usage by 33, reduced noise by 3.5db and they also gained 5% in performance.

If every card can do such an undervolt (i.e. stay stable) AMD really dropped the ball.

I don't think every card can do this. It's dependent on overclocking ability of the chip, which according to Kyle is hugely variable.

 

[mesyn191]

Rx 480 is basically running out of its optimum clock range and the main reason is the GF 14LPP implementation. the GF 14LPP process sucks and has significant variation among even review samples. If Vega is built at 14LPP then AMD is dead.

I don't think its clear you can say GF's 14nm process is crap yet.

GCN has never been able to overclock well, has always had fairly high power usage, and this is a launch product on a new process too. Polaris is just the latest iteration of GCN (1.4) remember so there is no reason to expect why the first 2 issues would go away. The 3rd issue I brought up + the relatively high idle power usage + significant variation among samples strongly suggests a process issue though. That could all be process teething issues. For all we know AMD was willing to accept all those issues to get really good wafer yields at launch to improve profits and supply. If so those problems could either be mitigated over time or possibly eliminated altogether as the process matures.

Given that GF's 14nm process is essentially a adapted version of Samsung's, which as far as anyone knows works just fine, there is some reason to be optimistic here about improvements over time.

As for Vega, since its still GCN, its a given that it'll use lots more power than the 1080 and that it'll overclock poorly while also having a relatively ho hum clock speed. The interesting part will be to see just how willing AMD is compete on price with their high end part AND if they can still make a profit while having good sales with such a part. If they get near 1080's performance for say $500 I think they could have a good, but low profit, seller on their hands even if it uses around 100-150W more than the 1080. We'll have to wait and see but hopefully not for much longer if the rumors of Vega being pulled forward into Oct. are true. Worst case is probably Jan 2017.

 

[Algrim]

I think someone's already pointed out that the RX 480 silicon is showing it being a seventh stepping. That implies a few modifications to the original design to get it to 'spec'.

 

[Chimpee]

I think someone's already pointed out that the RX 480 silicon is showing it being a seventh stepping. That implies a few modifications to the original design to get it to 'spec'.

C7 revision if I am not mistaken.

 

[NKD]

I don't think its clear you can say GF's 14nm process is crap yet.

GCN has never been able to overclock well, has always had fairly high power usage, and this is a launch product on a new process too. Polaris is just the latest iteration of GCN (1.4) remember so there is no reason to expect why the first 2 issues would go away. The 3rd issue I brought up + the relatively high idle power usage + significant variation among samples strongly suggests a process issue though. That could all be process teething issues. For all we know AMD was willing to accept all those issues to get really good wafer yields at launch to improve profits and supply. If so those problems could either be mitigated over time or possibly eliminated altogether as the process matures.

Given that GF's 14nm process is essentially a adapted version of Samsung's, which as far as anyone knows works just fine, there is some reason to be optimistic here about improvements over time.

As for Vega, since its still GCN, its a given that it'll use lots more power than the 1080 and that it'll overclock poorly while also having a relatively ho hum clock speed. The interesting part will be to see just how willing AMD is compete on price with their high end part AND if they can still make a profit while having good sales with such a part. If they get near 1080's performance for say $500 I think they could have a good, but low profit, seller on their hands even if it uses around 100-150W more than the 1080. We'll have to wait and see but hopefully not for much longer if the rumors of Vega being pulled forward into Oct. are true. Worst case is probably Jan 2017.

Vega is not exactly the same. It's using newer version of Graphics IP 9.0. I don't know exactly what that means but people say it could mean possibly higher clocks, cuz they might be using updated libraries. Rest is chinese to me.

 

[primetime]

my 480 is getting returned. The card freuency keeps dropping to 300hz during fallout 4. The drama about power usage also has me a bit spooked

Arnt these new cards using some kind of new overclocking software? I just wonder if its got new power saving glitch? Its like AMD expects the driver teams to make the cards power efficient....which is fine but at what cost lol

 

[Algrim]

There's a thread on OCN about 1070/1080s dropping to 300 MHz as well. Are you running an overclocked monitor or double monitors?

 

[NKD]

Arnt these new cards using some kind of new overclocking software? I just wonder if its got new power saving glitch? Its like AMD expects the driver teams to make the cards power efficient....which is fine but at what cost lol

I think they got this new tech and shit with boost clocks and adaptive control or something. They probably haven't gotten full hang of it. I am thinking card getting too much voltage at times.

May be thats why undervolting it is a big difference someone linked a site. It might not need that shit load of voltage that the software might be throwing at it all the time.

 

[razor1]

I think someone's already pointed out that the RX 480 silicon is showing it being a seventh stepping. That implies a few modifications to the original design to get it to 'spec'.

Yeah its 3 metal layer changes (which is what the C stands for, A, B, C) and 7 stepping (these 7 spins can be just minor tweaks)

 

[mesyn191]

I think someone's already pointed out that the RX 480 silicon is showing it being a seventh stepping. That implies a few modifications to the original design to get it to 'spec'.

That is normal though. Every chip ever goes through at least a few revisions before they have a shippable product. Even at Intel this is true. No one gets it right the first time.

Vega is not exactly the same. It's using newer version of Graphics IP 9.0.

No one knows exactly what Graphics IP 9 refers to. Its likely just a internal development name for Vega (remember its just listed in some guys public LinkedIn profile) and doesn't suggest any architectural improvements. In the end even if they did make some changes its still to what is a GCN design. Not much is going to change. Maybe it'll support some more advanced features of DX12 better.

 

[Stereodude]

There letters are likely all layer changes. The number being metal changes.

 

[razor1]

That is normal though. Every chip ever goes through at least a few revisions before they have a shippable product. Even at Intel this is true. No one gets it right the first time.


No one knows exactly what Graphics IP 9 refers to. Its likely just a internal development name for Vega (remember its just listed in some guys public LinkedIn profile) and doesn't suggest any architectural improvements. In the end even if they did make some changes its still to what is a GCN design. Not much is going to change. Maybe it'll support some more advanced features of DX12 better.

Graphics IP is the individual silicon blocks that make the GPU is my understanding. And it is different from Graphics IP 8, which is used in Fiji, Tonga, Hawaii.

Vega might be using TSMC, as there has been rumors that AMD might be using custom libraries or semi custom, this is the first time I have heard AMD using custom libs from their end, and not using 3rd party libs is conjunction with standard libs from TSMC.......

 

[razor1]

There letters are likely all layer changes. The number being metal changes.

Other way around , mask layer changes are numbers and the letters are metal

 

[heatlesssun]

There's a thread on OCN about 1070/1080s dropping to 300 MHz as well. Are you running an overclocked monitor or double monitors?

Do you have a link to that? I've been running my Gigabyte 1080 FE for a month. It will down clock well below 300 Mhz on the desktop when idle, I've noticed going to as low as 141 Mhz. But not while gaming.

 

[Palladium@SG]

What happens with the 1060 comes out with pretty much the same performance and has less power consumption, (better perf/watt)?

Will you say the same thing?

What did nV say, perf/watt is the same thing as performance. It actually is, because the more you can do with less, the more you have

If AMD just matched Maxwell 2 from last gen which is 1 and half nodes back (28nm vs 14nm), that doesn't give you a chill up your back what the 1060 can do, a smaller chip than the rx 480 and the same 1 and half node jump from Maxwell 2?

If everything is equal on the performance and price fronts, then other factors come into play, like perf/watt, like power consumption, like how hot the chip gets. If those other fronts are better, then price can be set higher, if they are lower then price will be set lower.

Even if the 1060 didn't exist, did AMD really believe nobody would want to pony up from a RX480 to a 1070 where perf/price is similar when OCing is concerned, and Pascal is 80% more power efficient?

Then we have the RX480 PCIE power overdraw shenanigans that is going to kill a ton of appeal for that alone.

Heck if anything the RX480 would have made even more people spamming F5 on the $400 AIB 1070s. Because they can be reasonably sure that there is no way Vega is going to compete with GP104 at this rate.

 

[Trimlock]

Vega is spinning up to be another Fury, very late to the game and only barely able to compete.

Although I have to admit I wanted the Fury only for the included water cooler, it would of worked in my HTPC if my PSU was up to handle it or if I needed that much GPU.

 

[Smashed Ixnay]

I was really wanting to get a 470, but I dunno if I should bother now. I'll probably just try to snag a gtx 970 when I can find one for cheap.

 

[Meaker]

Even if the 1060 didn't exist, did AMD really believe nobody would want to pony up from a RX480 to a 1070 where perf/price is similar when OCing is concerned, and Pascal is 80% more power efficient?

Then we have the RX480 PCIE power overdraw shenanigans that is going to kill a ton of appeal for that alone.

Heck if anything the RX480 would have made even more people spamming F5 on the $400 AIB 1070s. Because they can be reasonably sure that there is no way Vega is going to compete with GP104 at this rate.

Sure if people can spend more they can for extra performance but most people can't

 

[Trimlock]

Sure if people can spend more they can for extra performance but most people can't

The $200 price area is a big area to own. Nvidia knows this too, which is why I think the rumors of them releasing the 1060 earlier is why.

I was considering the RX480 when I was going to buy a 1070, so I held off, now I get to hold off to see if the 1060 will be any good. I don't need much, excellent 1080p gaming and marginal 1440 is good enough for me.

 

[Rvenger]

Anyone remember the old reference GTX 670 fan and how it sounded mechanical at idle and a little chattery/noisy at load? Yeah, this fan has the same attributes. It must be an acoustical thing with the extended shroud.

 

[Sodapopjones]

Sure if people can spend more they can for extra performance but most people can't

I was going to grab a 480 and put away the space heater till Vega or Pascal Ti hits, but really the 1070 looks more and more the way to go.

 

[razor1]

Even if the 1060 didn't exist, did AMD really believe nobody would want to pony up from a RX480 to a 1070 where perf/price is similar when OCing is concerned, and Pascal is 80% more power efficient?

Then we have the RX480 PCIE power overdraw shenanigans that is going to kill a ton of appeal for that alone.

From 240 bucks to 400 bucks that's a big difference, yeah people that are interested in the AIB cards from 300 to 400 that's a much less price bump lol (and if its $379 if those cards ever show up), but still its sizable its a 33% increase.

Heck if anything the RX480 would have made even more people spamming F5 on the $400 AIB 1070s. Because they can be reasonably sure that there is no way Vega is going to compete with GP104 at this rate.

I really wouldn't make assumptions on Vega at this moment cause we don't know much about it and added to that with the new graphics IP and also if there are node issues with 14nm that might be fixed by that time too or they go to TSMC...... I would say yeah its a tough challenge for AMD and Vega, but lets wait on it.