AMD Releases Statement On Radeon RX 480 Power Consumption; More Details Tuesday
by Ryan Smith on July 2, 2016 5:05 AM EST
Following up on this week's Radeon RX 480 launch, there has been some questions raised about the power consumption of the card. This is after some sites whom directly tap the power rails feeding the card discovered that at least some of their samples were pulling more than the standard-allowed 75W over the PCIe slot and/or 6-pin PCIe external power connector.
To that end, it would appear that AMD's staff is working weekend duty, and they have just sent over the following statement.
As you know, we continuously tune our GPUs in order to maximize their performance within their given power envelopes and the speed of the memory interface, which in this case is an unprecedented 8Gbps for GDDR5. Recently, we identified select scenarios where the tuning of some RX 480 boards was not optimal. Fortunately, we can adjust the GPU's tuning via software in order to resolve this issue. We are already testing a driver that implements a fix, and we will provide an update to the community on our progress on Tuesday (July 5, 2016).
If some of the data is to be believed, these cards are exceeding 150W total at times, which would mean there is either something causing them to run in the wrong power state, or they are just outright exeeding their power limit and need to be throttled back. As we don't do per-rail testing I don't have anything meaningful to add at this second, but it will be very interesting to see how AMD responds next week.
Update 07/06: AMD has since released their status update, which you can find here.
Facebook
Twitter
RSS
181 Comments
View All Comments
atlantico - Sunday, July 3, 2016 - link
Wow, so the Asrock Extreme Gaming4 mobo can't run overclocked GPUs. A likely story. What's with all these tech-wienies and little girls freaking out over exceeding the PCIe spec? Here's some news for you, everything from the GTX 960 to just about every OCd GPU out there violates the PCIe spec - and you believe a gaming motherboard can't handle that.Cool story brah.
atlantico - Sunday, July 3, 2016 - link
If you ever overclocked, or used an overclocked GPU, you probably went way over the PCIe spec. Mobo manufacturers for the last 10 years at least have designed gaming mobos for OCd GPUs.So duh. I've been running 2x OCd 7950s for 3 years, both sucking way more from the PCIe than the spec "allows", because: overclock.
No power issues, motherboards completely unaffected and working as intended. The PCIe standard is ignored every day by users all over the world with all sorts of GPUs and it just doesn't matter.
mutantmagnet - Sunday, July 3, 2016 - link
The only thing ridiculous is your ability to jump to conclusions. I was mostly pointing out that you now have to take into account what motherboard you're using if you decide to use a reference 480.Everything else you stated is based on your imagination.
bill.rookard - Saturday, July 2, 2016 - link
I actually read quite a bit about what's going on and there are a few things to note about it.1) It's not actually an issue with the motherboard itself as long as it's a good motherboard. Where the problem lies is the actual physical PCIe connector. If you have a motherboard which is Crossfire/SLI rated then you should be fine with a single card. The max spec is 75w per slot. Thus a SLI/Crossfire setup could theoretically pull 150w through the motherboard, causing damage, -but- since it's only pulling in the range of 80-90w, the limit you're hitting is the single slot pulling the power.
2) the '90 watt' number that's being pulled in general is a -spike-, as in a single point based on the resolution of the oscilloscope doing the reading (many times a second). If you've watched the power delivery of a GPU in fine resolution, it's all over the place up and down. What's important is the average. If you look at the average, then yes, it's running a bit high, but not drastically so (maybe 5w), while the spike is what causes the 90w number. Should it be that way? Probably not, most cards draw 30-50w through the slot, the remainder through the PEG.
3) Can they issue a software fix? Hopefully. If they can issue a fix (either VBIOS or driver) which changes the balance of power to pull more through the PEG and less through the slot that would be better. Yes, it's temporarily putting more power but if you look at actual wire capacities, an 18 gauge conductor can flow 10A @ 12v, 16 gauge can flow 20A @ 12v. That means a 6 pin 18AWG PEG (12v x 2, gnd x 3) will be capable of a combined wattage of 12v * (10A + 10A) = 240w. Why it's only rated for 30% of it's potential capacity is beyond me, unless it's the physical connection which is the limiter. Still, even if they went with 50% of wire capacity, it should easily handle 120w.
shabby - Saturday, July 2, 2016 - link
1. How many slots you have is irrelevant, there are 5 pins/traces in the pcie slot that pull the power, if all those pins/traces are overloaded on the motherboard by 30-40% for long periods of time they're going to pop or something around them is going to pop.2. In the pcper's oc'ed results its running at 80-90 watts constantly, its not a spike, limit is 66 watts.
3. That's most likely the easiest solution, but its still technically out of spec on the 6 pin connector, aib partners will probably remedy this with an 8 pin connector. Ati should just scrap the reference design.
lmotaku - Saturday, July 2, 2016 - link
Your friend should worry about brand / model / retail price originally of his board.990fx isn't enough to go on.
ie: Asus Sabertooth is fine. My M5a97 R2.0 has some quirks but has a lot of power controls and protections. MSI gaming boards, and some Gigabyte gaming boards should also be fine. I only trust those three, anyway, MSI and Asus being at the top, Gigabye, then Asrock as a very, very last. Asrock is okay for budget focused builds, but you have to be sure it's their highest of high gaming builds, but even then, I might not trust it still. Only two boards I've heard of having issues so far is a Foxconn AM2 OEM board and some Asrock board(don't remember them saying which), but it had multiple PCI-E slots and burned them all out.
A beta driver for 480 was just released for 480, which will give a consumer more fine-tuned control until they update their bios to run a bit differently for those who paid under $80 for their mobos.
Impulses - Saturday, July 2, 2016 - link
This does not bode well for AMD, last thing they need right now is a scandal, or to let this fester for days.Impulses - Saturday, July 2, 2016 - link
The card isn't even very efficient to begin with... And I'm not one to complain about that when the performance is there (after all, I had CF 6950 and currently have CF R9 290), but the 480 is basically on a bang/buck tightrope as it is.Eden-K121D - Saturday, July 2, 2016 - link
This card is like Maxwell efficiencymedi03 - Saturday, July 2, 2016 - link
Better, although, roughly in that ballpark.From nVidia's own (shitty FUDdy skewed scale) slides, 1060 is about 40% better at perf/w.
But again, that's at "reference" 480 clock.
Note that even vs AMD's previous gen, that's only about 1.8 improvement.
While Raja was claiming that 1.7 comes from process shrink, and they pushed it further up to 2.8, by "optimizing architecture".
Slides that shows that number mentions it's 470 vs 270x, though.