One Power Hungry GPU: AMD Radeon Vega 56 Revisited in 2020

[Julio Franco]

Today we're looking at AMD's Radeon RX Vega 56, originally a $400 product that was meant to compete against the GTX 1070 in mid 2017. If you're wondering why is this relevant, we'd say the main reason is that to this very day, you can buy a Vega 56 graphics cards for as little as $250.

Read the full article here.

https://www.techspot.com/review/1993-amd-vega-56-revisited/

 

[QuantumPhysics]

Hopefully AMD’s next line of GPU fair better.

It always seems to me that AMD favors “more” over “efficiency “ and refinement.

You’ve got 20 cores? Well I’ve got 30!!!

It works well for workstations but it’s terrible for gaming when energy efficiency is a concern.

 

[Puiu]

Hopefully AMD’s next line of GPU fair better

IPC won't change that much (although they did say that RDNA has a lot of things that can be improved), but they'll definitely make the die bigger

 

[redhat]

A question, Rdna cards are more efficient than nvidia‘s cards but when taking in consideration the total system power then it consume more power?

 

[QuantumPhysics]

A question, Rdna cards are more efficient than nvidia‘s cards but when taking in consideration the total system power then it consume more power?

Energy efficiency is coming from having smaller circuit prints but also from optimization.

 

[NightAntilli]

AMD has made some great advances. Too bad they aren't really valued for their efforts by the gaming community, but trashed left and right.

 

[Puiu]

A question, Rdna cards are more efficient than nvidia‘s cards but when taking in consideration the total system power then it consume more power?

They're about the same efficiency wise (2000 series vs 5000 series), but Nvidia is still using the older 14nm process node so they can gain a further 20-30% power usage reduction with their 7nm next gen parts.

AMD managed 50% because GCN (and Vega) was an all-in-one architecture for compute+gaming (saving money on r&d) which drew a lot of power at high clock speeds. Nvidia will prolly not gain that much, but I do hope they can prove me wrong.

 

[krizby]

A question, Rdna cards are more efficient than nvidia‘s cards but when taking in consideration the total system power then it consume more power?

CPU and RAM also come into equation when measuring total systems power, AMD's driver just seem to utilize more CPU cycles than Nvidia. Also faster GPU will require more CPU cycles.

 

[deksman2]

A power pig?
If you left the voltages at stock levels perhaps, but not with an undervolt and an overclock (both of which make V56 perform the same or better than GTX 1080 at LOWER power consumption than 1080.

5600XT would probably still be more efficient, but regardless, AMD has an issue of overvolting GPU's out of the factory... but its not just that.
Vega uArch as it exists on V56 has much larger compute performance (and that is a serious power hog).

 

[Judespace]

They're about the same efficiency wise (2000 series vs 5000 series), but Nvidia is still using the older 14nm process node so they can gain a further 20-30% power usage reduction with their 7nm next gen parts.

AMD managed 50% because GCN (and Vega) was an all-in-one architecture for compute+gaming (saving money on r&d) which drew a lot of power at high clock speeds. Nvidia will prolly not gain that much, but I do hope they can prove me wrong.

Turing is on 12nm finfet

 

[grumblguts]

Only if you let it.
I drop my vega card down 20% on power and I only loose a few % in performance.
Goes from 550 watt down to sub 400 at the wall.
Its funny you are not showing the 1080 or 1070 cards power usage.
You would think you are trying to hide something.

 

[Theinsanegamer]

Anyone buying a vega 56 brand new is a total mong. You can buy them for less then $200 on ebay. That's where the real value is.

 

[Squid Surprise]

While I appreciate the article, the intro compares it pricewise to the Nvidia 1660 and states that it originally competed with the Nvidia 1070...

It would have been nice to have included those 2 cards here as well...

And, for the troll who will inevitably post this, I am aware that those benchmarks have been done in different articles - but for readability and convenience, they should be included in this one.

 

[rrwards]

Anyone buying a vega 56 brand new is a total mong. You can buy them for less then $200 on ebay. That's where the real value is.

Absolutely this. The whole point is to buy them used and undervolt em. Although, I understand why they didn't do this, it would be nice to have it included as a comparison.

 

[Puiu]

Turing is on 12nm finfet

The names don't matter much, it's the same generation (marketing gimmicks). You can call it 14/16nm+ if you want. (it's actually just a rebranded 16nm process with a bit better better gate density and few optimisations)

 

[deksman2]

Only if you let it.
I drop my vega card down 20% on power and I only loose a few % in performance.
Goes from 550 watt down to sub 400 at the wall.
Its funny you are not showing the 1080 or 1070 cards power usage.
You would think you are trying to hide something.

This.
When undervolted and overclocked, Vega 56 actually matches or surpasses GTX 1080 and draws less power than 1080.
Most of the performance V56 can gain is from increasing HBM frequencies by about 50-100MhZ without increasing power draw of the VRAM by more than 5 or 10W
Besides, HBM and core have separate voltages... if the core is undervolted and slightly overclocked with HBM having a larger overclock, it results in a very nice performance boost at much lower power draw (as I already mentioned)

One of the benefits of AMD GPU's is that AMD gives the user an option to undervolt, whereas NV doesn't allow for similar ease of voltage modification since core and voltage on NV gpu's are often tied together.

Overall speaking, V56 is a really good GPU when manually modified and definitely still more than capable (especially with latest drivers and features AMD enabled).

 

[deksman2]

Absolutely this. The whole point is to buy them used and undervolt em. Although, I understand why they didn't do this, it would be nice to have it included as a comparison.

For some reason, official reviewers don't undervolt GPU's to test them out like that.
They just give you comparisons at stock levels.
Some reviewers won't be shy of overclocking the GPU's to test them, but they almost never undervolt.

 

[dirtyferret]

The review could have started and ended with the "total system power consumption" chart.

 

[Lew Zealand]

A power pig?
If you left the voltages at stock levels perhaps, but not with an undervolt and an overclock (both of which make V56 perform the same or better than GTX 1080 at LOWER power consumption than 1080.

5600XT would probably still be more efficient, but regardless, AMD has an issue of overvolting GPU's out of the factory... but its not just that.
Vega uArch as it exists on V56 has much larger compute performance (and that is a serious power hog).

Why do you assume the 1080 doesn't undervolt? I undervolt my 1080 all the time.

Runs at 1822 MHz 0.85V averaging 125W use. Or 1923 MHz at 0.9V averaging 140W use. And that's with the crappiest PNY 1080 card available. Anyone with a better cooler (which is everyone) would get another bin higher clocks.

Let's see your Vega 56 numbers.

 

[CharmsD]

The names don't matter much

Then why debate it?

it's the same generation (marketing gimmicks). You can call it 14/16nm+ if you want. (it's actually just a rebranded 16nm process with a bit better better gate density and few optimisations)

Turing is 12nm FinFet

"Jen-Hsun is still talking about working on node designs with TSMC, there is still the possibility that it’s adapting a 7nm process with the company’s traditional pureplay foundry partner. “In terms of process nodes we tend to design our own process with TSMC,” explains Huang. “Buying an off-the-shelf process is something that we can surely do, but we want to do much more than that.” "​

​

 

[Disparity]

Did you actually test the Vega or pull a random figure from the internet. The 56 shipped with a core power limit around 180W which meant 210W peak on the base cards.

Your data suggests 350W, 200W over the 5600XT (150W) which would need modded power tables and a big overclock.

Something doesn't add up.

 

[deathtrap]

https://tpucdn.com/review/powercolor-radeon-rx-5600-xt-red-devil/images/power-gaming-average.png

A Stock Vega56 does not slurp more power than a RX590, fact.

 

[Beerfloat]

Only if you let it.
I drop my vega card down 20% on power and I only loose a few % in performance.
Goes from 550 watt down to sub 400 at the wall.
Its funny you are not showing the 1080 or 1070 cards power usage.
You would think you are trying to hide something.

These kinds of posts are really tone deaf.

This article covers mass produced consumer electronics devices and the kind of experience you can expect from them in the configuration that their vendors see fit to back with a warranty.

Some individual samples may or may not be able to operate slightly differently with undervolting, overclocking, spit shining and/or go faster stripes painted on, and while that may well go some way towards making the owner of such a sample feel better about their purchase, it is also completely irrelevant to an article such as this one.

 

[Theinsanegamer]

Did you actually test the Vega or pull a random figure from the internet. The 56 shipped with a core power limit around 180W which meant 210W peak on the base cards.

Your data suggests 350W, 200W over the 5600XT (150W) which would need modded power tables and a big overclock.

Something doesn't add up.

210 watt peak for a vega 56...

LMFAO.

Anyone that owns a vega GPU will tell you that is pure BS. Vegas blow right past 180 watts. You cant reduce power consumption of a vega gpu by 100+ watts if they were only pulling 180 watt to begin with. Vega 64s (I know, as I own one) push 400 watts at full bore with the fan speed pushed up to allow higher clock rates. 56s are not much smaller in terms of power draw.

 

[Colonel Blimp]

Got my Saphire Pulse 56 for £260 brand new from Overclockers Uk a year ago.
Massively happy with it.
No plans to change it for a year or 2.
I like what AMD has done with Navi but whilst performance per watt has improved, performance per dollar hasnt much.
A little undervolt helps it stay at max frequency for longer and consumes less power.
No driver issues for me either.