Radeon released RX 480. (Card is Released)

[Bokelka]

Mining review:

http://lab501.ro/placi-video/review-amd-radeon-rx-480/18

24 Mh/s on Claymore,not bad

It depends on the power consumption. If that is 24 MH or 90 to 100 W, then it is quite good. If it is 120W, not so.

"So taking into account the power conversion efficiency of an 80 Plus Gold PSU that was used the actual power usage of the card is apparently 150W at the default settings. This is further confirmed by the fact that while mining Ethereum with the default settings the GPU frequency hovers up to about 1240 MHz and does not go all the way up to 1266 MHz. Essentially hitting the power limit of 150W that the card has"

I wish they would have tried -100 mv on the card to see what it did to hash rate or lowered the TDP to 90%.

I have 4 coming and will be testing them in a few days.

It seems to achieve the highest hash rate at 9000MHz memory, it does not need the core to be 1266. So it is possible to reduce the core voltage.

[PovertyByte]

Mining review:

http://lab501.ro/placi-video/review-amd-radeon-rx-480/18

24 Mh/s on Claymore,not bad

It depends on the power consumption. If that is 24 MH or 90 to 100 W, then it is quite good. If it is 120W, not so.

"So taking into account the power conversion efficiency of an 80 Plus Gold PSU that was used the actual power usage of the card is apparently 150W at the default settings. This is further confirmed by the fact that while mining Ethereum with the default settings the GPU frequency hovers up to about 1240 MHz and does not go all the way up to 1266 MHz. Essentially hitting the power limit of 150W that the card has"

I wish they would have tried -100 mv on the card to see what it did to hash rate or lowered the TDP to 90%.

I have 4 coming and will be testing them in a few days.

It seems to achieve the highest hash rate at 9000MHz memory, it does not need the core to be 1266. So it is possible to reduce the core voltage.

That 9ghz memory was also the limit of the software, which may change as newer versions of the card come while also including better cooling. Memory clock alone only increased 2 watts per MH which is tremendously efficient. We may see variants of this card pushing over 30 later.

https://www.youtube.com/watch?v=5CJBaZ9V2Eo

However, this video for gaming claims that the temps rise very fast from small overclocks and gets instable fast. Most likely vcore. It also tops the 390 in some games but the 390 seems to top the 480 more often. There is no denying the 480 is significantly more power efficient.

I feel like AMD shot themselves in the foot with the reference cooler on a major hype card. Months from now when some people are thinking to upgrade and see what is new they will find lots of videos and threads mentioning these temps, even if we see non reference models finally rise.

[generalt]

It seems newegg sold all of their  $199 cards out.

Very disappointing.

I bought 2 x sapphire 480 - 8 gb and 2 x powercolor 480 - 8gb

From what I understand the 4gb cards memory clock is 7 ghz stock and the 8gb cards are 8 ghz stock so you might want to stick with the 8gb cards.

[greaterninja]

It seems newegg sold all of their  $199 cards out.

Very disappointing.

I bought 2 x sapphire 480 - 8 gb and 2 x powercolor 480 - 8gb

From what I understand the 4gb cards memory clock is 7 ghz stock and the 8gb cards are 8 ghz stock so you might want to stick with the 8gb cards.

The 4gb is 8ghz on Sapphire and XFX models @ 199+ tax.

I've cancelled my 8gb orders as I cannot justify paying $40 more + tax  for mining purposes.

18 units inbound @ $199+tax.  I've gone with the brands I believe will have the best thermal dissipation on reference .

If the cards suck ass I'll be sending a few back for refund or selling them.

I'm pretty confident I can achieve 25-26 mh/s.  Even 24 mh/s will be acceptable for the price and power usage.

EDIT:  OOh actually you might be right.  XFX website shows 7000mhz for the 4gb XFX and the other is 8000 mhz for the 4gb.   So newegg could be listing the products with the wrong specs.

[adaseb]

This card isn't really worth it if it does 24MH/s using 150Watts. My 7950 uses 150Watts and hashes at 20MH/s.

[greaterninja]

This card isn't really worth it if it does 24MH/s using 150Watts. My 7950 uses 150Watts and hashes at 20MH/s.

The 7950 is Tahiti based @ 28nm, ~3 generations behind the rx 480 and about 4 years old (released in 2012).
Tahiti uses a lot of power at max load unless you undervolt or downclock.  


The rx 480 is @ 14nm and power utilization benchmarks already indicate it uses significantly less power at max load.
Comparing the 480x to the Tonga (which is newer and more efficient than the Tahiti), it is at least 17% more power efficient in watts and hashes Ethereum ~20% faster.
And well the Tonga is 20-35% more power efficient than the latest Tahiti chips.

By this data, the rx 480 will likely be at least 20-55% more power efficient than the 7950 hashing Ethereum and hash around 20% faster than a 7950.

[adaseb]

This card isn't really worth it if it does 24MH/s using 150Watts. My 7950 uses 150Watts and hashes at 20MH/s.

The 7950 is Tahiti based @ 28nm, ~3 generations behind the rx 480 and about 4 years old (released in 2012).
Tahiti uses a lot of power at max load unless you undervolt or downclock.  


The rx 480 is @ 14nm and power utilization benchmarks already indicate it uses significantly less power at max load.
Comparing the 480x to the Tonga (which is newer and more efficient than the Tahiti), it is at least 17% more power efficient in watts and hashes Ethereum ~20% faster.
And well the Tonga is 20-35% more power efficient than the latest Tahiti chips.

By this data, the rx 480 will likely be at least 20-55% more power efficient than the 7950 hashing Ethereum and hash around 20% faster than a 7950.

Yeah old card but its only  25% slower and uses the same amount of power as an 480.

Maybe for games its much much faster but not for ETH mining.

[reb0rn21]

What even worse 7950 & 280x etc can undervolt on top of that NANO with undervolt can draw as low as <100W for same speed as 480x

No one expected that AMD new GPU will use that match power, also reference GPU cooler is disaster with the heat it produce, keeping it in rig of 4-6 cards you must have AC that directly blow cold air at it!

So far miners should wait for 470 but I don`t know how it can be match better with disastrous TDP, so if the 470 with custom cooler and good OC can have some decent speed and low price, it might be OK
But 480 deffo NOT, maybe for 200$ no more!, with the diff and price drop its hard to make a profit at the end!

[Marvell1]

What even worse 7950 & 280x etc can undervolt on top of that NANO with undervolt can draw as low as <100W for same speed as 480x

No one expected that AMD new GPU will use that match power, also reference GPU cooler is disaster with the heat it produce, keeping it in rig of 4-6 cards you must have AC that directly blow cold air at it!

So far miners should wait for 470 but I don`t know how it can be match better with disastrous TDP, so if the 470 with custom cooler and good OC can have some decent speed and low price, it might be OK
But 480 deffo NOT, maybe for 200$ no more!, with the diff and price drop its hard to make a profit at the end!

got two vsiontek 480s , one is stable a 28mh/s with a 1000 or so mem OC, the other one won't go over 24 mhs, i think its a dud

power ussage does seems hiigh though we need a version of MSI afterburner that can undervolt the mem and core voltages.  Power use is around 150 watts for 27.7 mhs if i crank up the memory it will probbaly go higher

[mirny]

be careful with oc, 480 can push 85w+ through pciex at stock clocks

[reb0rn21]

@Marvell1
The speed of mining is OK, but the power usage is disaster, even if new afterburner or some tool manage to lower the voltage we can not know if the card would be stable and would not cripple the speed to match

Now you should try to push memory as more you can, it does not use excessive power and ethereum is always hungry for it

For instance 79xx and 280x were gr8 for undervolting but I had no luck with 290/390 sure they can undervolt but most are unstable with even -50mV

[64dimensions]

This card isn't really worth it if it does 24MH/s using 150Watts. My 7950 uses 150Watts and hashes at 20MH/s.

The 7950 is Tahiti based @ 28nm, ~3 generations behind the rx 480 and about 4 years old (released in 2012).
Tahiti uses a lot of power at max load unless you undervolt or downclock.  


The rx 480 is @ 14nm and power utilization benchmarks already indicate it uses significantly less power at max load.
Comparing the 480x to the Tonga (which is newer and more efficient than the Tahiti), it is at least 17% more power efficient in watts and hashes Ethereum ~20% faster.
And well the Tonga is 20-35% more power efficient than the latest Tahiti chips.

By this data, the rx 480 will likely be at least 20-55% more power efficient than the 7950 hashing Ethereum and hash around 20% faster than a 7950.

A less facile view of the problem:

1) As I pointed out in an earlier post, initially, RX480 will have thermal management problems:


1) In going from "28nm" fab technology to "14nm" fab technology, the the chip area has shrunk by about 1/2 to 1/4 of the area of a roughly similar 28nm fabbed device.

2) Say you have a 100W chip. If you shrink its area by half and everything else is the same, the amount of thermal energy per unit area that has to be dissipated has doubled. The problem is that you may have to use a technology such as water cooling to handle this increased heat dissipation per unit area. The heat transfer properties of passive radiators/conductors using metals in conjunction with thermal grease have their own limits in terms of how much heat flow they can sustain per unit area.


The thermal management issues will take months to cheaply work out. What shrinking the die does in this case is increase the heat flow per unit area and lower the thermal mass. So whatever system is controlling the chip temp has to be faster and heat dissipation is now a real issue because the thermal resistance of the chip-grease-cooler stack hasn't changed.

Most likely OC'ing this chip and not using the OEM software to do this will void the warranty. I would also void the warranty is the cooler is separated from the chip.

The 7950 owner is quite right in holding on to this card until these issues are settled.

Right now the older cards are useful in places where electricity is cheap and at least with these 1st RX cards air conditioning is free.

Finally, in winter I'm using the Phil Ma solution for my older cards and bringing them inside to provide heat.

[greaterninja]

This card isn't really worth it if it does 24MH/s using 150Watts. My 7950 uses 150Watts and hashes at 20MH/s.

The 7950 is Tahiti based @ 28nm, ~3 generations behind the rx 480 and about 4 years old (released in 2012).
Tahiti uses a lot of power at max load unless you undervolt or downclock.  


The rx 480 is @ 14nm and power utilization benchmarks already indicate it uses significantly less power at max load.
Comparing the 480x to the Tonga (which is newer and more efficient than the Tahiti), it is at least 17% more power efficient in watts and hashes Ethereum ~20% faster.
And well the Tonga is 20-35% more power efficient than the latest Tahiti chips.

By this data, the rx 480 will likely be at least 20-55% more power efficient than the 7950 hashing Ethereum and hash around 20% faster than a 7950.

A less facile view of the problem:

1) As I pointed out in an earlier post, initially, RX480 will have thermal management problems:


1) In going from "28nm" fab technology to "14nm" fab technology, the the chip area has shrunk by about 1/2 to 1/4 of the area of a roughly similar 28nm fabbed device.

2) Say you have a 100W chip. If you shrink its area by half and everything else is the same, the amount of thermal energy per unit area that has to be dissipated has doubled. The problem is that you may have to use a technology such as water cooling to handle this increased heat dissipation per unit area. The heat transfer properties of passive radiators/conductors using metals in conjunction with thermal grease have their own limits in terms of how much heat flow they can sustain per unit area.


The thermal management issues will take months to cheaply work out. What shrinking the die does in this case is increase the heat flow per unit area and lower the thermal mass. So whatever system is controlling the chip temp has to be faster and heat dissipation is now a real issue because the thermal resistance of the chip-grease-cooler stack hasn't changed.

Most likely OC'ing this chip and not using the OEM software to do this will void the warranty. I would also void the warranty is the cooler is separated from the chip.

The 7950 owner is quite right in holding on to this card until these issues are settled.

Right now the older cards are useful in places where electricity is cheap and at least with these 1st RX cards air conditioning is free.

Finally, in winter I'm using the Phil Ma solution for my older cards and bringing them inside to provide heat.

I've worked in semiconductors for 3 years...specifically in chip packaging for companies such as Apple and AMD.  Going from 28nm process to 14nm does not imply thermal energy per area stays the same.  It also does not imply that the chip area/die shrinks.  It usually means transistor dimensions used to build the chip get smaller and heat generated per area usually goes down as the process becomes more efficient.  I think your argument is a little flawed as you are associating the 28 vs 14 nm process to the chip's area when it directly relates to the smallest feature of a chip...in this case the transistors.

[reb0rn21]

Maybe he mean problem to transfer heat from small die to cooler might be a problem, like with sandy bridge i7 CPU and lack of direct die contact

Anyway Nvidia managed to rise the frequency a lot with 16nm but still power usage is fine even with first chip on 16nm node, maybe Samsung 14nm node is not so gr8 for monolithic large GPU

[jstefanop]

Getting about 26-27 MH on mine with basic settings that can be adjusted via crimson. Ill measure exact power draw in a bit, but running around 80c with very quiet fan profile (50% fan speed).

This should easily hit 30 MH once non reference cards hit.

[hawkfish007]

Getting about 26-27 MH on mine with basic settings that can be adjusted via crimson. Ill measure exact power draw in a bit, but running around 80c with very quiet fan profile (50% fan speed).

This should easily hit 30 MH once non reference cards hit.

Nice, what mining software are you using? I will have to wait till tomorrow to get mine 

[jmumich]

I've worked in semiconductors for 3 years...specifically in chip packaging for companies such as Apple and AMD.  Going from 28nm process to 14nm does not imply thermal energy per area stays the same.  It also does not imply that the chip area/die shrinks.  It usually means transistor dimensions used to build the chip get smaller and heat generated per area usually goes down as the process becomes more efficient.  I think your argument is a little flawed as you are associating the 28 vs 14 nm process to the chip's area when it directly relates to the smallest feature of a chip...in this case the transistors.

Thanks for this explanation. Here, it looks like (according to Wikipedia anyway) the die size is 232 mm^2, with 5.7 billion transistors. That's somewhere between Tonga (R9 380) with 5 billion transistors and a 359 mm^2 die size and Hawaii (R9 390) with 6.2 billion with a 438mm^2 die size.

So it looks like they did shrink the die size to closer to Trinidad (R7 370, 2.8 billion transistors, 212 mm^2 die size).

If I understand this right, there's some room to grow.

[jstefanop]

Looks like its pulling around 180 watts for me...which is pretty horrible considering this is supposed to be a 150 watt card. I guess considering the higher hashrate its not bad, but definitely higher than i expected.

[Tmdz]

Wow that seems like a lot considering my 290s with a undervolt pull only 220 watts at the wall, which is only 40 watts more than this card.  I paid $60 cheaper for them per card that means I can mine on my 290's for 6+ months before I even see that savings in the power cost. 

Seems hard to justify at this point to get on the rx 480 game.

[philipma1957]

Looks like its pulling around 180 watts for me...which is pretty horrible considering this is supposed to be a 150 watt card. I guess considering the higher hashrate its not bad, but definitely higher than i expected.

and what are your settings>  standard clocks?   are these cards voltage  locked?   no under volting?

which one do you have?

and is it the 4gb or the 8gb