AMD Radeon HD 7970 3GB Review

[SKYMTL]

Benchmarking the Benefits of PCI-E 3.0

Benchmarking the Benefits of PCI-E 3.0

One of the points about the Southern Islands products that are bound to come up in countless marketing documents is their use of the new PCI-E 3.0 specification. Much like the switch between PCI-E 1.1 to 2.0, this certification incorporates several new features that could prove to be hugely beneficial to graphics card manufacturers. First and foremost among the improvements is a brand new encoding scheme that helps it effectively double the bandwidth of PCI-E 2.0 to 8 GT/s.

While current generation single and dual GPU cards still fit well within the bandwidth limitations of PCI-E 2.0 x16 slots, Tahiti’s creators will be trumpeting PCI-E 3.0 to anyone within earshot. However, does it really make a difference in gaming scenarios? More importantly, will upgrading to a PCI-E 3.0 motherboard allow the HD 7970 to pull further ahead of the competition? To find out, we used an ASUS P9X79 motherboard (which allows users to switch between Gen 2 and Gen 3 PCI-E functionality in its BIOS) along with our usual stable of games and benchmarks. To ensure outside factors didn’t play into the equation, the highest resolution and image quality settings were used in every game.

Starting things off was 3DMark 11 and the differences between a Gen2 and Gen3 interface was effectively nil. From the perspective of 3DMark, it really doesn’t seem to care whether the HD 7970 runs on a higher bandwidth interface or not.

The results couldn’t be any clearer: the HD 7970 doesn’t benefit in any way from a PCI-E 3.0 link to the CPU. Granted, there were a few small increases (and decreases) here and there but these are a result of averaging out the results from multiple manual benchmark runs and fall well within our margin of error. So to the relief of gamers and reviewers alike (hey, we still love our X58 test systems!), a next generation PCI-E interface isn’t needed to get the most out of the Tahiti XT core.

However there are a few caveats that should be mentioned. Even though a single HD 7970 may not saturate a PCI-E 2.0 x16 slot, things could change drastically once two of these cards are installed into a Z68 motherboard which uses a dual 8x setup for Crossfire and SLI. Things will likely get even more complicated when AMD’s new dual GPU product is released in the first half of 2012. We’ll likely revisit this topic throughout 2012 but for the time being, rest assured knowing that AMD’s HD 7970 is in no way limited by current PCI-E certifications.

 

[SKYMTL]

GPGPU Performance; OpenCL & DirectCompute

GPGPU Performance; OpenCL & DirectCompute

Like many other modern GPU cores, Southern Islands has been designed from the ground up to provide exceptional compute performance. With dual DMA engines that have the capability to saturate the PCI-E 3.0 bus and a highly parallelized Graphics Core Next architecture, it is no wonder why AMD has been talking up these cards’ advances in this field.

In order to put some of these claims to the test, we used a number of synthetic OpenCL and DirectCompute benchmarks with all cards running on a PCI-E 3.0 slot. In order to cut the CPU out of the equation, we focused strictly upon stressing the shaders.

First up in our testing regimen was LuxMark, a simple benchmark that uses an OpenCL calculated ray tracing routine to render a highly complex scene. As with many of these benchmarks you will see in this section, its results are largely theoretical but could be replicated if programmers choose to implement highly optimized OpenCL code into their applications. With that being said, the HD 7970 walks all over its predecessor by posting a 65% improvement.

SISoft’s Sandra platform is a one stop shop for benchmarkers and a simple OpenCL engine was recently included and later perfected. As with LuxMark, it can truly show the Tahiti XT’s potential in compute tasks and how much this architecture has improved over the previous generation. To put this into perspective, the HD 7970 has only about 35% more compute units than a HD 6970 but its performance has more than doubled in some cases.

For DirectCompute performance, we used a number of different benchmarks ranging from some typical Mandelbrot sets to more complex 3D dynamic fluid calculations. All of the scenarios were measured in frames per second using FRAPS over a 3 minute period for each.

Again we see some eye opening improvements from one generation to the next and this time the differences between new and old is significant. The HD 7970 is really able to stretch its legs in some of the more complex scenarios.

Unfortunately, we couldn’t get any Folding@Home numbers for this review since Stanford’s new V7 beta GPU client (which supports OpenCL) failed to recognize our card. Hopefully this will be rectified in the next release since as we saw above, the HD 7970 has the potential to be a Folding powerhouse if the OpenCL coding can be successfully ported over to the GCN architecture. Until then, we’ll continue looking for some real world applications that can highlight the benefits of OpenCL and DirectCompute performance.

 

[SKYMTL]

Temperature & Acoustical Testing / Power Consumption

Temperature Analysis

For all temperature testing, the cards were placed on an open test bench with a single 120mm 1200RPM fan placed ~8” away from the heatsink. The ambient temperature was kept at a constant 22°C (+/- 0.5°C). If the ambient temperatures rose above 23°C at any time throughout the test, all benchmarking was stopped. For this test we use the 3DMark Batch Size test at its highest triangle count with 4xAA and 16xAF enabled and looped it for one hour to determine the peak load temperature as measured by GPU-Z.

For Idle tests, we let the system idle at the Windows 7 desktop for 15 minutes and recorded the peak temperature.

With an efficient 28nm core beating at its heart the HD 7970 posts some very good temperature numbers, especially when compared to the HD 6970. This should also allow for some reasonable overclocking headroom at near default fan speeds. However, these low temperatures do come at a price….

Acoustical Testing

Yes, we have finally added decibel testing to our repertoire and this section will expand in future reviews. What you see below are the baseline idle dB(A) results attained for a relatively quiet open-case system (specs are in the Methodology section) sans GPU along with the attained results for each individual card in idle and load scenarios. The meter we use has been calibrated and is placed at seated ear-level exactly 12” away from the GPU’s fan. For the load scenarios, a loop of Unigine Heave 2.5 is used in order to generate a constant load on the GPU(s) over the course of 20 minutes.

AMD claims their larger fan with its upsized fins should provide a quieter gaming experience but we didn’t see any of that during testing. The HD 7970 quite loud when compared to the NVIDIA cards. Its fan doesn’t shriek like a gut shot banshee (a la HD 6990) but the noise it puts out will certainly be heard throughout its immediate vicinity and any adjoining rooms if your case doesn’t have some form of acoustical dampening.

Interestingly enough, the fan doesn’t go above 40% so it is completely possible that one our engineering sample’s thermal sensors is slightly buggy but after speaking to some of our colleagues, several are achieving the same results. On the flip side of that coin some others have reported a perfectly quiet computing experience. Take this as you may but from our perspective, additional testing will certainly have to be done on retail products.

System Power Consumption

For this test we hooked up our power supply to a UPM power meter that will log the power consumption of the whole system twice every second. In order to stress the GPU as much as possible we once again use the Batch Render test in 3DMark06 and let it run for 30 minutes to determine the peak power consumption while letting the card sit at a stable Windows desktop for 30 minutes to determine the peak idle power consumption. We have also included several other tests as well.

Please note that after extensive testing, we have found that simply plugging in a power meter to a wall outlet or UPS will NOT give you accurate power consumption numbers due to slight changes in the input voltage. Thus we use a Tripp-Lite 1800W line conditioner between the 120V outlet and the power meter.

While we were disappointed in the acoustical results, there’s no denying that AMD hit the nail on the head in the efficiency department. Not only was our HD 7970 far more efficient than the GTX 580 but its in-game performance was significantly better as well.

When compared against the HD 6970, we honestly had to rerun the tests several times to validate the results because at first, they looked to be a mistake. Here we had a graphics card that beat out the previous generation flagship by a good 40% across the board and yet it consumed less power. Something didn’t look right but after hours of testing, it seems our initial impressions held out. AMD claims the HD 7970 should consume about as much as the Cayman XT so we may have just received a low leakage core but nonetheless, kudos to the design team on this one.

 

[SKYMTL]

Overclocking Results

Overclocking Results

When it comes to higher end cards, AMD’s focus has gradually shifted towards an overclocker-friendly mentality. The inclusion of a dual BIOS switch along with a fair amount of claimed overhead should allow enthusiasts to customize these cards’ clock speeds to their hearts’ content.

Much like the HD 6900-series, we were told the default clock speeds on the HD 7970 were set at conservative levels. Supposedly, 1Ghz and higher clock speeds would be easily achievable and few ASICs could even make it to the 1.2GHz mark. The memory modules have also been chosen for their overhead since their default speed is 6Gbps rather than the 5.5Gbps they actually run at.

AMD does give users a number of basic tools for HD 7970 overclocking but sets limits on how far you can actually push clock speeds. The core speeds can go up to 1125MHz while the memory tops out at 1575MHz (6300MHz in QDR mode) but we highly doubt may will be able to get much further without voltage adjustments. If you’re in the mood to push things even further, a supporting version of MSI’s AfterBurner software is only a few weeks away from release.

In our experience, overclocking our HD 7970 was relatively easy but our theory about having received a low leakage core seems to have held true since we couldn’t push the card past the 1.078Ghz mark while maintaining long term stability. In order to ensure we weren’t held back by PowerTune, its level was increased +20% and we decided to keep the default –yet aggressive- fan speed profile. For the record, temperatures never went above the 80 degree mark throughout gaming while the core was overclocked.

Memory clocking with GDDR5 is an interesting experience since its advanced error correction means artifacts will rarely pop up. Rather, increasing its speeds past a certain point will just result in flatlined performance numbers. We got up to the 6.21Gbps mark before framerates hit a plateau. Naturally, the performance was mind blowing.

Final Clock Speeds:

Core: 1078MHz
Memory: 6214MHz (QDR)

 

[SKYMTL]

Conclusion

Conclusion

The introduction of a new architecture always carries with it a fair amount of hope and in many cases, a healthy dose of apprehension as well. On one hand, if all the right pieces fall into place a company could be well positioned for the future but if things don’t turn out quite as planned, being tied at the hip to a failed design could have far reaching repercussions. While AMD’s missed opportunity with Bulldozer will go down in history as a disappointment that could stick around for a long time, their Southern Islands family is off to one hell of a roaring start. Not only did the Tahiti XT exceed almost every one of our expectations, with the right amount of massaging this architecture still has plenty of room for future growth.

The list of improvements AMD built into the Graphics Core Next architecture is impressive to say the least, especially on the compute front. Gone is the slightly antiquated VLIW instruction set and in its place is a highly parallelized design that (in some cases) allows for mind blowing performance when compared against previous generations of Radeon cards. This bodes particularly well for any HPC-oriented products based off of the Southern Islands family even though most home users just won’t benefit from high level GPGPU capabilities.

Most of you reading this are interested in gaming performance and let’s be perfectly clear: the HD 7970 3GB represents a giant leap forward for AMD GPU performance. It left the HD 6970 in the dust and handily beat NVIDIA’s GTX 580 in nearly every single game. By looking at the results in individual cases, it’s apparent that AMD’s new flagship card excels in Shogun 2, Deus Ex and The Witcher 2 but it still retains a significant advantage over the GTX 580 in Battlefield 3 and Crysis 2, two areas where driver optimizations could improve performance further. The GTX 580 3GB makes things interesting at extreme detail settings but the result is still a rather convincing win for the HD 7970.

The HD 7970's framerates at 1920 x 1200 may be class leading but it almost begs for an ultra high resolution 27” or 30” monitor in order to really shine. Indeed, when detail settings are cranked and MSAA is increased to higher levels, the 3GB of GDDR5 is able to stretch its legs and the card pulls well ahead of almost every other reference spec single GPU product on the market. At times the HD 7970 even came close to GTX 590 performance levels. So Tahiti XT does have what it takes to pull well ahead of the current competition but the real question here is whether this is enough to stay in the lead throughout 2012.

With gamers and enthusiasts becoming all too aware of the ballooning power requirements and thermal output from modern graphics cards, it was high time something changed. AMD’s approach to finding a solution took several different paths but when combined they really are revolutionary. The new 2D idle power states along with ZeroCore Power will make the HD 7970 all that much easier to live with, particularly for anyone who wants a multi card setup. However, the real star of the show from a gaming standpoint is the efficiency brought to the table by the 28nm manufacturing process. In our testing the HD 7970 consumed less power under load and idle than its predecessor -granted, this could have been due to a low leakage core- which is an incredible accomplishment.

Efficiency may be a huge selling point these days but this GPU tends to stumble in one key area: acoustics. AMD claims to have reworked many of their heatsink’s design elements but we’d still classify the HD 7970 as an overly loud card. The core temperatures may have been great and overclocking came quite easily but without end user input, the fan ramps up far too quickly and makes an absolute racket. We would have much rather seen thermals go into the low 80s than experience decibel levels in the mid 50s.

Let’s be honest here, the HD 7970’s price will likely come as a shock to many but it accomplishes the one thing that AMD’s last generation failed to do: convincingly beat the incumbent NVIDIA flagship in most gaming scenarios. We’re all used to seeing Radeon branded products undercutting the competition but this time AMD has a bona fide top tier card and it’s priced accordingly. That shouldn’t take anything away from the fact that $549 is certainly a hefty amount to pay and retailers may take further advantage of limited supply. Nonetheless, once the initial surprise wears off it becomes clear that the HD 7970’s cost is in line with its relative performance against the current crop of modern GPUs.

For the time being AMD’s HD 7970 sits in a preeminent position within the GPU market and we’d highly recommend it to anyone looking for a high end graphics card. Its combination of extreme performance at every resolution, advanced features and a relatively low power signature make it one of the best products released in the last two years. There are however some nagging questions which mostly revolve around availability and final cost. Aside from the fact that AMD has another paper launch on their hands, supply of 28nm GPUs will be tight for the foreseeable future so we don’t have high expectations for the HD 7970’s initial availability once it actually hits the retail channel. And yet if the planets align and products are ready to buy at a fair price, AMD could very well have an insurmountable lead by the time the competition launches their answering salvo.

http://www.hardwarecanucks.com/forum/video-cards/49655-comment-thread-amd-hd-7970-3gb-review.html​