RX 480 DC Benchmarks

Senture

Senture

Senture

Associate
Joined
1 May 2007
Posts
2,354
Location
London
I expect there will be a few people interested in these cards for distributed computing work. I have a pair of these which will be delivered tomorrow.

I'm planning to give them a blast with Folding@Home for a while first. What other benchmarks would people like to see?

Test system configuration:

Dual Xeon E5-2670 V1 HT Off 4% Overclock
64GB (8x8GB) DDR3 ECC 1600Mhz
2x Gigabyte RX 480 @ Stock. Radeon Crimson driver 16.6.2. Cards spaced 2 slots apart
240GB Kingston Fury SSD for OS, 160GB WD Velociraptor for DC Storage
Lian-Li PC-A76X with 3 120mm Intake and 3 120mm Exhaust fans
Corsair AX1500i
Windows 10 Professional 64 Bit build 10586

Corsair Link used for power draw. Power Out (component power draw total excluding efficiency) value used for data analysis.

Folding@Home Performance

Core Project (Run/Clone/Gen) TPF Estimated PPD Power Draw (Min/Max) Config Base Credit
21 9640 (2/7/99) 01:56 307703 233w/239w Single Card 4300
21 11419 (3/16/11) 05:41 277277 440w/452w Dual Card 25110 <<<Thermal Throttle
21 9704 (24/0/176) 02:11 294128 440w/452w Dual Card 9000
21 10495 (6/54/55) 02:57 232621 442w/455w Dual Card 14421
21 9442 (34/0/283) 02:21 305172 432w/440w Dual Card 10400
21 10494 (11/7/92) 05:06 300351 432w/440w Dual Card 23137

Moo! Wrapper Performance

All recorded units ran below the target temperature of 80 degrees, but below the maximum boost clock with 75% fan speed. This indicates that the card is throttling due to power usage limits.

Further units can be observed here

Project WU Type Run Time (Seconds) Credit Units Per Card Core Clock
Moo! Wrapper AMD 833.59 6144 1 1070-1100
Moo! Wrapper AMD 877.59 6144 1 1117-1153
Moo! Wrapper AMD 889.99 6144 1 1145-1161
Moo! Wrapper AMD 933.28 6584 1130-1158

Collatz Conjecture Performance

More units can be observed here

Project WU Type Run Time (Seconds) Credit Units Per Card Core Clock
Collatz Conjecture Collatz Sieve 1.21 198.51 4295.37 1 1229-1246
Collatz Conjecture Collatz Sieve 1.21 196.56 5094.7 1 1229-1246
Collatz Conjecture Collatz Sieve 1.21 194.55 4421.14 1 1229-1246
Collatz Conjecture Collatz Sieve 1.21 186.14 4706.79 1 1229-1246
Collatz Conjecture Collatz Sieve 1.21 194.55 4521.14 1 1229-1246

Poem@Home Performance

More units can be observed here

Project WU Type Run Time (Seconds) Credit Units Per Card Core Clock
Poem@Home Poem++ OpenCL 2.31 979.56 5500 1 1236-1255
Poem@Home Poem++ OpenCL 2.31 1491.17 9100 1 1236-1255
Poem@Home Poem++ OpenCL 2.31 1519.95 9100 1 1236-1255
Poem@Home Poem++ OpenCL 2.31 730.05 5500 1 1236-1255
Poem@Home Poem++ OpenCL 2.31 885.55 5500 1 1236-1255
Poem@Home Poem++ OpenCL 2.31 659.99 5500 1 1236-1255
Poem@Home Poem++ OpenCL 2.31 1607.72 9600 1 1236-1255
 
Last edited:
Senture said:
I expect there will be a few people interested in these cards for distributed computing work. I have a pair of these which will be delivered tomorrow.

I'm planning to give them a blast with Folding@Home for a while first. What other benchmarks would people like to see?
Click to expand...

Moo wrapper, collatz conjecture and poem when you have time please :)
 
I'm looking forward to seeing the ppd for fah senture, and also if possible the approximate power draw from both of them if you have an energy meter lying around.
 
hanluc said:
Moo wrapper, collatz conjecture and poem when you have time please :)
Click to expand...

I'll get these done as soon as I can :)

currob said:
I'm looking forward to seeing the ppd for fah senture, and also if possible the approximate power draw from both of them if you have an energy meter lying around.
Click to expand...

I don't happen to have one, but I can swipe the AX1500i from my other computer and use Corsair Link to read the power draw from the PSU its self.
 
Managed to pinch the AX1500i from my other machine and fit it into the test rig. Updated the first post with details of the test rig also.

Dome some initial testing to get numbers of power draw before any GPU goes in. I'm doing the tests with HT off for a couple of reasons.

#1 I use this machine for video rendering along with DC stuff, the video rendering only scales to 20 threads then virtually stops scaling past 24. #2 with HT on running DC work it kicks out a considerable amount of heat. This makes my room incredibly uncomfortable at night so I'd prefer to produce 75% of the points 100% of the time than 100% of the points 50% of the time.

System State Power Draw Minimum / Power Draw Maximum

Idle (No GPU Installed) 65w/75w

1 Core No GPU Installed 97w/118w
2 Cores No GPU Installed 113w /125w
3 Cores No GPU Installed 122w /138w
4 Cores No GPU Installed 140w/154w
5 Cores No GPU Installed 150w /166w
6 Cores No GPU Installed 158w /176w
7 Cores No GPU Installed 166w /188w
8 Cores No GPU Installed 184w /200w
9 Cores No GPU Installed 204w /209w
10 Cores No GPU Installed 214w /221w
11 Cores No GPU Installed 227w /233w
12 Cores No GPU Installed 235w /241w
13 Cores No GPU Installed 247w /251w
14 Cores No GPU Installed 256w /260w
15 Cores No GPU Installed 264w /268w
16 Cores No GPU Installed 270w /274w

32 Cores (HT On) No GPU 300w/308w
 
These cards seem to perform at about a GTX 970's level in Folding@Home, which is quite consistent with other gaming related benchmarks. Each card pulls about 150w with occasional spikes up to 160w. All the benchmarks are being ran at stock settings, with one exception: a more aggressive fan curve.

The default fan curve is sufficient for gaming, but runs into thermal throttling issues with DC work (as you might expect). The default profile optimises for a target temperature of 80 degrees, and throttles at 90 degrees. During the single card test, the stock profile kept the fan speed at about 50% while maintaining 79-81 degrees on the core. Adding the second card had some negative impact on cooling (even though they are 2 slots apart) making the top card reach 89 degrees with 58% fan speed and thermal throttle.

Fan noise is not bad for a reference blower cooler. Noise levels are reasonable up to 50% when it starts to become loud. Temps are kept in check at about 80 degrees with 70% fan speed maintaining the rated boost clock of 1266Mhz. Noise levels are still lower than my Gigabyte GTX 980 G1 Gaming cards.

On the whole I am quite impressed with the performance from these cards, and the significantly lower power draw compared to the R9 3XX series. I would recommend to people interested in these cards to wait for the AIB productions coming later on. I have not tried overclocking these cards yet, but given other reviewers have barely got a 6% overclock out of them (typically in the 1340-1360Mhz range) and that these cards hit their thermal limits easily in DC work it seems ill advised to even attempt to (unless you run the fan at 100% which then sounds like a hair dryer).

More benchmarks to come!
 
hanluc said:
Thanks for that Senture, I'll do some comparisions against my older AMD cards soon, at first glance it looks pretty good, esp. considering the power draw.
Click to expand...

You're welcome. These are the first AMD cards I've had since the 4850. I've always preferred the nvidia drivers for some of their features and general organisation, but the AMD Crimson drivers have been really good with 1 exception: for love nor money I cannot get these cards to crossfire. I've tried every slot configuration possible, different BIOS options and reinstalled the drivers about half a dozen times, no dice. But since I wasn't really intending to run these in crossfire I'm not too fussed about it.

I did laugh when I calculated the Moo! Wrapper PPD from these cards at about 1.2 Million a day. Collatz should be even higher, but I didn't calculate it.
 
is that 1.2m per card? I had a 280x running moo wrapper and that was saying 500k ppd
I may need some amd cards as we have pretty much been taken off the board on moo wrapper :(
 
Problem I have is that I only have AMD in my Mac Pro and it seems shiahtay instablilityville at anything OpenCL on the gpu.

Its probably better on projects I have not interest in like collatz and Moo but like I said solving number puzzles just doesn't grab me like medical projects.
 
You must log in or register to reply here.
Back
Top Bottom