Overclocking, Thermals, and Power Consumption
AMD's Fiji Pro and Sapphire's Nitro
About six months ago the cut down variant of the Fiji based GPU known as Fiji Pro was unleashed to the world. Known as Fiji Pro and packing 3584 stream processors along with 4gb of HBM on a 512bit bus we saw limited launch with only two graphics cards, one of which we previously reviewed here.
| WCCFTech | AMD Radeon R9 Fury X2 | AMD Radeon R9 Fury X | AMD Radeon R9 Nano | AMD Radeon R9 Fury | AMD Radeon R9 290X |
|---|---|---|---|---|---|
| GPU | Fiji XT x 2 | Fiji XT | Fiji XT | Fiji Pro | Hawaii XT |
| Stream Processors | 8192 | 4096 | 4096 | 3584 | 2816 |
| GCN Compute Units | 128 | 64 | 64 | 56 | 44 |
| Render Output Units | 128 | 64 | 64 | 64 | 64 |
| Texture Mapping Units | 512 | 256 | 256 | 224 | 176 |
| GPU Frequency | TBA | Up to 1050Mhz | Up to 1000 MHz | Up to 1000 MHz | 1000Mhz |
| Memory | 8GB HBM (4 GB Per Chip) | 4GB HBM | 4GB HBM | 4GB HBM | 4GB GDDR5 |
| Memory Interface | 4096-bit x 2 | 4096bit | 4096bit | 4096bit | 512bit |
| Memory Frequency | 500Mhz | 500Mhz | 500 MHz | 500Mhz | 1250Mhz |
| Effective Memory Speed | 1Gbps | 1Gbps | 1Gbps | 1Gbps | 5Gbps |
| Memory Bandwidth | 1024 GB/s | 512GB/s | 512GB/s | 512GB/s | 320GB/s |
| Cooling | Liquid | Liquid, 120mm Radiator | Air, Single Fan. Custom AIB Solutions Later | Air, Custom AIB Solutions | Air, Single Blower Fan |
| Performance (SPFP) | 17.2 TFLOPS | 8.6 TFLOPS | 8.19 TFLOPS | 7.2 TFLOPS | 5.6 TFLOPS |
| TDP | TBA | 275W | 175W | 275W | 250W |
| Power Connectors | Dual 8-Pin | Dual 8-Pin | 8-Pin | Dual 8-Pin | 6+8 Pin |
| GFLOPS/Watt | TBA | 31.3 | 47.1 | 26.2 | 19.3 |
| New Prices | TBA | $649 | $499 | $549 | $299 (Retail) |
| Launch Date | Early 2016 | 24th June 2015 | 7th September 2015 | 10th July 2015 | 24th October 2013 |
But why a re-launch six months later? For us to really understand why we need to compare the previous Tri-X model and the new updated Nitro model. That all starts with the PCB, while the original featured the same reference pcb layout that the fury x employed, while that made for easy waterblock compatibility it also limited what Sapphire was able to do with the card itself and forced them to build around a foundation. With the Nitro card Sapphire was able to build it from the ground up to their demands with an enhanced 6-phase circuit and an 8 layer pcb that helps lower vrm temperatures (unable to validate due to sensors not reading in GPUZ). They were also able to incorporate a dual bios switch that can regulate power draw; the default BIOS limits the TDP to 260w and a target temperature of 75C, the second BIOS option allows for 300w and a target temperature of 80C. Sapphire also changed the chokes to Black Diamond in order to combat the known coil whine exhibited by gpu’s in the Fury lineup I am happy to report that outside of loading screens in FarCry 4 and Hitman Absolution I experienced no noticeable coil whine.
This version of the Tri-X cooling solution has been beefed up as well and unfortunately extends down into an extra slot essentially making this a 2 ½ slot cooling solution. The results however are more than worth this sacrifice. Sapphire is touting this to be their “most efficient cooling solution ever built by the company”. Featuring an array of heatpipes including a 10mm one connected via a solid copper plate and cooled by three 90mm fans that sit idle until the gpu hits a certain temperature. They use a die cast bracket to connect the heatsink to the card to aid in supporting the 12” monster. And on the backside feature a love it or hate it designed aluminum backplate.
Coming in at core clock of 1050mhz makes this the fastest factory clocked Fury card on the market at an MSRP of $549, it does come in a standard version as well for an MSRP of $539 for those who want to save a few dollars and overclock on their own.
Sapphire Radeon R9 Fury Nitro Specifications:
Test System
| Component | Selection |
| CPU | Intel Core i7-4770k (4.2ghz) |
| Motherboard | Asus Maximus VI Hero |
| Power Supply | Cooler Master V1200platinum |
| HDD | Crucial MX100 512gb |
| Storage Disk | Seagate 4TB SSHD |
| Memory | 8GB AMD Radeon R9 Gamer Ram DDR3 2400 |
| Monitor | Monoprice Crystalview 4k |
| Video Cards | Sapphire Nitro R9 Fury, Geforce GTX 980 |
| Operating System | Window 10 64-Bit |
The latest drivers were used with both GPUS. Nitro Fury = Radeon Software 16.1, GTX 980 Geforce 361.43. We would like to thank Youtuber Brian P. (www.twitter.com/badseedtech) for allowing us to borrow his GTX 980 for comparative purposes in this review.
AMD Radeon R9 Fury Block Diagram:
The AMD Fiji GPU is based on a 28nm process and is the single largest chip AMD has ever produced. The chip which measures around 596mm2 has a total of 8.9 Billion transistors crammed under its hood which means a whole lot of performance and the card really does deliver a amazing amount of improvement in performance and efficiency over previous generation GPUs. Compared to Hawaii with its 6.3 billion transistors and 5.6 TFlops of performance, the Fiji GPU crunches out 7.2 TFlops of gpu performance from the depths of its heart. Also comapred to Hawaii, AMD has managed to deliver several Micro architecture improvements along with enhanced power management features and a totally refocused look at board designs. The whole die measures a massive 1011mm2 while the interposer is designed with a 65nm process.
The AMD Radeon R9 Fury is the only card in the Fury series that comes with custom designs. The other two cards, the Radeon R9 Fury X and the Radeon R9 Nano are currently offered as reference only designs and while AMD has made it clear that they will have the flagship Fury X kept the same way, the Radeon R9 Nano will get some custom cooled variants in the near future. The Radeon R9 Fury is also the only cut down version of the Fiji GPU based lineup that comes with 54 compute units which pack 64 stream processors each, total of these amount to 3596 stream processors, 64 Raster operation units and 224 texture mapping units. The engine clock for the Radeon R9 Fury WindForce 3X is maintained at 1010 MHz that is clocked 10 MHz faster than the stock variant but can allow additional overclocking thanks to the better PCB components.
The texture fill rate is rated at 224.0 GT/s and 64.0 GP/s Pixel Fill rate. The card churns out a total of 7.2 TFlops of compute performance. The card also features 4 GB of HBM VRAM that is clocked at a 500 MHz (1.0 Gbps data rate). The memory runs across a 4096-bit bus interface has has a bandwidth of up to 512 GB/s. The card is powered by dual 8-Pin connectors and has a board power of 275W. Display outputs include HDMI, single HDMI and three display ports. The card has support for next generation API such as DX12 and supports Freesync, VSR and Frame Rate Target Control.
Gaming Benchmarks
GTAV
GTAV was run at 1080p, 1440p, and UHD using the in game benchmark at the highest settings possible in the basic graphics menu, meaning if High was the highest setting that’s what was used, if Ultra that was used. MSAA x2 was used and Water MSAA was left at 0. Non of the Extended Graphics settings were used. However the Shadow settings were set to each vendors specific rendering option.
*Note: the min fps was observed with other Radeon GPUs unsure of cause but it was repeatable so it stayed.
Middle Earth: Shadow of Mordor
Shadow of Mordor was tested at 1080p, 1440p, and UHD all at the Ultra Preset using the in game benchmarking utility
Dirt Rally
Dirt Rally was tested 1080p, 1440p, and UHD at the Ultra Preset with 2x MSAA using the in game benchmarking utility
Metro: Last Light
Metro Last light was tested at 1080p, 1440p, and UHD using Very High quality, texture filtering at 16x, Motion Blur at Normal, and Tesselation set to Normal. This was done using the in game benchmarking utility
Tomb Raider
Tomb Raider was tested at 1080p, 1440p, and UHD using the Ultimate preset including TressFX using the in game benchmarking utility.
The Witcher 3
The Witcher 3 was tested at 1080p, 1440p, and UHD using the Ultra preset, but disabling all hairworks. Since the game does not feature a benchmark utility I set up a course in The White Orchard and rode the horse from the Ransacked Village to the Sawmill.
Battlefront
Battlefront was tested at 1080p, 1440p, and UHD using the Ultra Preset and running 5 waves on Hoth Survival Mode.
Synthetic Benchmarks
Firestrike
Unigine Heaven
Using the Extreme preset but changing the Resolution option is how testing was done for Unigine Heaven. I expected this test to be a runaway for the GTX 980 as Maxwell handles tessellation so well, but it came out much closer than I expected.
Overclocking, Thermals and Power Consumption
A big selling factor of the Nitro Fury is the overclocking potential. Now with unlocked voltage and a freshly redesigned power delivery system expert overclockers are likely hit beyond the speeds I did. Using MSI Afterburner and simply increasing the power limit to +50 I was able to easily reach 1150mhz on the core, Voltage and Memory overclocking was left alone as this model had to be returned to Sapphire. Now this wasn’t a massive overclock by any means, especially when you consider our reference GTX 980 was able to hit 1456mhz on the core and 7894mhz on the memory. But as the results show it only extends the lead the Nitro Fury has over the GTX 980
Thermals stay in check with the Nitro staying under its target temperature at all times even while in case with the side panel on. Even when overclocked and the fan profile barely ramping up at all it still manages to keep it as cool as a stock clocked reference GTX 980 and just as quiet. Thermals were tested with the side panel on the case and 30 minutes of looping Unigine Valley, sound was measured with a meter one foot from side of case immediately after the panel was removed once temperatures were noted.
Where the Nitro gets left in the dust however is the power consumption charts at stock clocks it pulls roughly 75 watts more than the stock GTX 980 under a 1440p gaming load. Overclock it and it pulls nearly 100 watts more than an overclocked GTX 980 system. But, there is a bit of ray of light for red team fans. The overclocked GTX 980 closes in the gap vs the stock Nitro Fury and at that point you’re looking at only a much more palatable 48w difference.
Conclusion
In the end it’s safe to say that Sapphire has created the premier and possibly the best Fiji Pro based graphics card. The noise level is so low one would question whether the fans ever come on and the temperatures stay in check leaving little concern for heat. I dare say going to a higher tier single card may just be an e-peen situation as I have a R9 295x2 and Nano handy and prefer running this gpu over both of those. The GTX 980 offers up a compelling argument at a few dollars cheaper and (at the time of this review) bundles Rise of the Tomb Raider. Heavily overclocked the GTX 980 does come close in performance, but then again you still get a faster experience ‘out of the box’ with the Sapphire Nitro Fury. If you’ve got the space and you have power you should seriously consider this GPU for your 1440p, and with some tweaks and entry level 4k gaming experience.
Contents
Follow Wccftech on Google to get more of our news coverage in your feeds.
