AMD Readies Radeon RX 490 for December?

[cdawall]

It's sad to see how blinded people are thinking Polaris is more efficient than Pascal.

Blinded by what? There are already a couple consumer bins in the 80w range for the RX480. Not every card is the 150-200W cards we saw at initial release. If you look closely at the cooler design you can tell the initial plan was for a GPU in the 80-100w power envelope. Same goes for the power delivery circuit that was on the cards. Literally every single thing points towards a much lower wattage card design. It has been assumed on multiple fronts that GloFo couldn't give AMD enough good dies and that led to what we saw. AMD bottom of the barrel grabbed dies to have a product on the market. We will continue to see lower and lower wattage polaris parts as time goes on and yields improve.

Another example of AMD's preferential binning. The macbook pro can get a full fledged RX460 that is sitting at a 30w TDP.

Hell my cards with a massive overclock are still only hitting the 150-200w range with a voltage bump and air cooling.

 

[efikkan]

Hold on there a minute, mobile GPUs are not only binned, they are also heavily throttled. Both AMD and Nvidia do this, and it should not be confused with energy efficiency. AMD also tried this on Fury to make it more "energy efficient", pushing the consumption from 275W to 175W, while claiming to give up to the same level of performance. The key here is up to, in reality these GPUs will operate most of the time at much lower clock, e.g. ~850 MHz, while a typical desktop counterpart runs at ~1200 MHz (AMD), >1600 MHz (Nvidia). Binning does of course matter, but it's not going to give GPUs that are twice as energy efficient.

And exactly how are you measuring the power consumption of your GPU?

 

[cdawall]

Hold on there a minute, mobile GPUs are not only binned, they are also heavily throttled. Both AMD and Nvidia do this, and it should not be confused with energy efficiency. AMD also tried this on Fury to make it more "energy efficient", pushing the consumption from 275W to 175W, while claiming to give up to the same level of performance. The key here is up to, in reality these GPUs will operate most of the time at much lower clock, e.g. ~850 MHz, while a typical desktop counterpart runs at ~1200 MHz (AMD), >1600 MHz (Nvidia). Binning does of course matter, but it's not going to give GPUs that are twice as energy efficient.

And exactly how are you measuring the power consumption of your GPU?

Again this is where you have done no research. The rx460M is clocked to 1180 boost vs 1200 boost for a desktop version, but that card is junk it isn't the low wattage version found in the MacBook pro. Those actually have a rumor mill "900mhz" 460 PRO, with all of the Shaders enabled. So 1024 vs 890 and is actually faster than the desktop variant while only using a 35W TDP. The desktop card is 75W for reference. AMD has also already released multiple embedded versions of the RX480 that run at full 1288mhz boost speeds and have a 95W TDP for the whole board memory included.

Take off your blinders and look around a bit. GloFo released some junk early chips. They were not efficient at all. Now that those have been burned through and the process is improving we are seeing cards that don't suck.

My GPU's I am measuring with a combination of a kill-a-watt and MSI afterburner oddly enough software actually seems pretty accurate for a GPU alone measurement on these cards.

 

[Ungari]

For those who keep parroting the whole "Power Consumption, the Power Consumption!":

What will you object to when you see Polaris chips with a 95W TDP?

 

[efikkan]

AMD has also already released multiple embedded versions of the RX480 that run at full 1288mhz boost speeds and have a 95W TDP for the whole board memory included.

I'm sorry, but you clearly don't know how throttling works on modern GPUs. The "base clock", and "(average) boost clock" are simply targets, not actual clocks. If the power target is low, the GPU might clock much lower than the "base clock". Just look at the R9 Nano with a "base clock" of 1000 MHz, but the power target forced it way lower than that under heavy load. So whenever you see a mobile or embedded part with the "same specs" but much lower TDP, the real world performance will be much lower than the desktop counterpart with higher TDP.

Take off your blinders and look around a bit. GloFo released some junk early chips. They were not efficient at all. Now that those have been burned through and the process is improving we are seeing cards that don't suck.

Yes, I know the rumors. There is supposed to be some "mythical" cards out there. And it's basically all based on some PR claims from AMD. It's funny, isn't it? Every time AMD don't live up to the expectations, fans always claims they are just holding back? Or it's the wrong drivers? There's always an excuse.

There is no evidence that there are RX 480s out there with twice the energy efficiency of the reviews here at Techpowerup. I've checked >80 reviews now, both the initial ones and more recent ones. There are just the expected small variations between tests, and there are a lot of tests where RX 480 consume more than GTX 1080. The newer custom versions are not more efficient, actually in most cases they are worse. Even with yields improving, that's only going to give a few percent gains, nothing like a factor of two. You might be able to find a couple of random guys claiming they have a RX 480 with such capabilities, but any one with common sense knows it's one of two things; either they are lying or they have measured incorrectly. When the evidence is so massive, you've got to stop this nonsense.

My GPU's I am measuring with a combination of a kill-a-watt and MSI afterburner oddly enough software actually seems pretty accurate for a GPU alone measurement on these cards.

Any software measurement is not going to give you an accurate board power.

 

[Ungari]

I'm sorry, but you clearly don't know how throttling works on modern GPUs. The "base clock", and "(average) boost clock" are simply targets, not actual clocks. If the power target is low, the GPU might clock much lower than the "base clock". Just look at the R9 Nano with a "base clock" of 1000 MHz, but the power target forced it way lower than that under heavy load. So whenever you see a mobile or embedded part with the "same specs" but much lower TDP, the real world performance will be much lower than the desktop counterpart with higher TDP.

Yes, I know the rumors. There is supposed to be some "mythical" cards out there. And it's basically all based on some PR claims from AMD. It's funny, isn't it? Every time AMD don't live up to the expectations, fans always claims they are just holding back? Or it's the wrong drivers? There's always an excuse.

There is no evidence that there are RX 480s out there with twice the energy efficiency of the reviews here at Techpowerup. I've checked >80 reviews now, both the initial ones and more recent ones. There are just the expected small variations between tests, and there are a lot of tests where RX 480 consume more than GTX 1080. The newer custom versions are not more efficient, actually in most cases they are worse. Even with yields improving, that's only going to give a few percent gains, nothing like a factor of two. You might be able to find a couple of random guys claiming they have a RX 480 with such capabilities, but any one with common sense knows it's one of two things; either they are lying or they have measured incorrectly. When the evidence is so massive, you've got to stop this nonsense.

Any software measurement is not going to give you an accurate board power.

These AMD commercial Polaris chips down-clocked to 1100Mhz for long service life, shows the RX 480 variant featuring 95W Power Consumption, which would make this capable of higher clock-speeds than the current consumer offerings at 150W @ 1266Mhz.
http://www.amd.com/Documents/high-performance-gpu-product-brief.pdf
http://www.amd.com/Documents/high-performance-gpu-product-brief.pdf

 

[cdawall]

I'm sorry, but you clearly don't know how throttling works on modern GPUs. The "base clock", and "(average) boost clock" are simply targets, not actual clocks. If the power target is low, the GPU might clock much lower than the "base clock". Just look at the R9 Nano with a "base clock" of 1000 MHz, but the power target forced it way lower than that under heavy load. So whenever you see a mobile or embedded part with the "same specs" but much lower TDP, the real world performance will be much lower than the desktop counterpart with higher TDP.


Yes, I know the rumors. There is supposed to be some "mythical" cards out there. And it's basically all based on some PR claims from AMD. It's funny, isn't it? Every time AMD don't live up to the expectations, fans always claims they are just holding back? Or it's the wrong drivers? There's always an excuse.

There is no evidence that there are RX 480s out there with twice the energy efficiency of the reviews here at Techpowerup. I've checked >80 reviews now, both the initial ones and more recent ones. There are just the expected small variations between tests, and there are a lot of tests where RX 480 consume more than GTX 1080. The newer custom versions are not more efficient, actually in most cases they are worse. Even with yields improving, that's only going to give a few percent gains, nothing like a factor of two. You might be able to find a couple of random guys claiming they have a RX 480 with such capabilities, but any one with common sense knows it's one of two things; either they are lying or they have measured incorrectly. When the evidence is so massive, you've got to stop this nonsense.


Any software measurement is not going to give you an accurate board power.

Dude what are you arguing at this point? I know how boost clocks work.

The rumors of these mythical cards have been confirmed by multiple reviewers, google the XFX GTR. Oh wait he doesn't support your beliefs and must be a liar.

I also never said I had the most accurate measuring equipment I said rough figures and did so for a reason. There are plenty of cards out there with similar numbers.

 

[Ungari]

As yields of the highest binned chips improve, the yields of middle-binned chip may also improve giving a bump to the current consumer AIB RX 480s, as could be the case with these latest reports of XFX GTRs.

 

[cdawall]

As yields of the highest binned chips improve, the yields of middle-binned chip may also improve giving a bump to the current consumer AIB RX 480s, as could be the case with these latest reports of XFX GTRs.

Nope bins apparently can only change minute percents.

 

[ratirt]

Not sure what you guys are arguing about. RX 480 is a very good card and it is getting better and better. Just to clarify if the card wasn't good enough and not getting better and energy efficient(yeah!! right it is getting better in that matter too for those who believe it ain't) AMD would never try to release 490. Even if it's a dual gpu or a single it would never happen but it's happening. AMD's going to release it so it does mean something. And don't get cocky about 480 is not matching 1070 or 1080 since it was never its purpose. if 490 will match even 1070 or maybe 1080 that would be a huge success. The question is if it is meant to be a dual or a single GPU how will it handle power consumption and heat?

Just to add.
On the other hand AMD's never said about the 490 to be released. Does this mean VEGA is way more stronger than AMD anticipated? If 490 is a single GPU based on Polaris and it would surpass 1070 in performance almost reaching 1080 performance then it may be that VEGA might be extremely strong.

 

[Ungari]

Nope bins apparently can only change minute percents.

My understanding is that refinement to the process will produce more viable chips increasing the number of quality chips, with those chips increasing in performance.
Is this not correct?

 

[ratirt]

My understanding is that refinement to the process will produce more viable chips increasing the number of quality chips, with those chips increasing in performance.
Is this not correct?

Yeah!!!!
that's pretty what I mean Thanks Ungari.

 

[cdawall]

My understanding is that refinement to the process will produce more viable chips increasing the number of quality chips, with those chips increasing in performance.
Is this not correct?

Oh no you are completely correct my post was annoyed sarcasm based off of one users inability to accept things changing if they aren't green.