XFX MERC 310 RX 7900 XT Review: AMD’s Back in the Game

Thanks to XFX, we have the MERC 310 BLACK RX 7900 XT on the test bench today and is also our first GPU from AMD’s RDNA3 line of cards. The new RX 7900 XT is based on the NAVI 31 die and according to AMD, their new design delivers nearly 1.5x the performance per watt. With the new GPUs from AMD, we see a new approach (well, sort of) using a chiplet design. There are some similarities to the ZEN processors, but using chiplets on a GPU is very different compared to a CPU. We’ll try and break this down into more detail below and cover the features of the XDX MERC in our hands.

The MERC 310 RX 7900 XT brings a robust VRM design, a large aluminum triple-fan shroud with a plated copper heatsink, and an overclock out of the box. The 2.7-slot cooling solution includes an LED on the card’s facing edge showing the “XFX” logo. A full-length aluminum backplate cleans up the card’s backside while adding rigidity and heat dissipation. Overall, it’s a beautiful card and fits well with most build themes.

Performance-wise, this overclocked card performed well in our testing, besting the last generation RTX 3090 Ti and was ahead of the RTX 4080 in some tests or close behind in most. This card would be ideal for high-refresh FHD (1920 x 1080) and more than able to handle 144 Hz WQHD (2560 x 1440) screens. In our 4K UHD testing, the RX 7900 XT proved very capable, scoring 90 FPS or better across all benchmarks.

The MSRP for the XFX MERC 310 RX 7900 XT Black Edition is set at $949.00, a $50 premium over the 7900 XT reference model’s MSRP of $899.99. This premium has benefits: a factory overclock, improved power delivery, solid aluminum construction, a premium copper-based vapor chamber cooling solution, and a three-year warranty.

Since this is a new architecture launch, we need to discuss RDNA 3 and AMD’s new chiplet design; this may get a bit dry, so grab a coffee and settle in.

RDNA3 Chiplet Design

AMD uses a similar approach for NAVI 31 and the Zen CPU, separating the memory/IO portion from the computational part of the die. Zen4 CPUs contain a large Input/Output Die (IOD) manufactured on TSMC N6 that connects to the system memory and PCIe interface. This IOD is connected via the Infinity Fabric to one or two much smaller Core Compute Dies (CCD) manufactured using TSMC N5. Specific processes, such as memory interfaces and Cache, don’t scale performance-wise when shrinking nodes. On top of this, the TSMC N5 is more costly to produce than TSMC N6 leading AMD to mix processes in the CPUs for efficiency and cost reasons.

CPUs and GPUs are very different in how they function and communicate with the memory and the rest of the system. Typically CPUs have a limited number of cores when compared to GPUs. Take the Ryzen 9 7950X, for example, with 16 cores and 32 threads; this is a drop in the bucket compared to the RX 7900 XTX with 96 CUs for a total of 6,144 Stream Processors. Feeding all these cores takes an enormous amount of memory bandwidth.

To accommodate this bandwidth, AMD took an opposite approach with a larger Graphics Compute Die (GCD) housing thousands of Stream Processors, video codec hardware, display interfaces, and the PCIe connection. Connected to the GCD are up to six smaller Memory Controller Dies (MCD) containing the large Infinity Cache, Infinity Fabric links, and GDDR6 memory controllers. The GCD measures 300 square millimeters and contains 45.7 Billion transistors. Each MCD measures 37 square millimeters with 2.05 Billion transistors. In the specifications chart, the transistor count combines the GCD + MCD.

AMD still had another hurdle in their GPU chiplet design. The Infinity Fabric that AMD uses in their CPUs works well, but scaling it up to accommodate a 384-bit interface would consume too much power to make it feasible.

AMD needed a replacement for their organic substrate interconnect and developed something completely new, the Infinity Fanout Links. This new tech from AMD packs double the number of connections in less than 2% of the same area as the Infinity Fabric used. These links operate at 9.2 Gb/s, offering ten times the bandwidth of AMD’s Infinity Fabric for an industry-leading 5.3 TB/s. These links also account for less than 5% of the total GPU power consumption, thanks to AMD’s innovation.

RDNA 3 Architecture

On top of a total GPU redesign to accommodate the chiplet layout, AMD has also made improvements to most, if not all, aspects of its architecture. The compute units, Cache, and Ray tracing have all seen some tweaking. According to AMD, RDNA 3 GPUs can hit the same frequency as RDNA 2 while using half the power or 1.3 times the frequency while using the same power.

Although the official boost speeds seem modest, AMD claims that RDNA 3 has been created to reach speeds of 3 GHz. During our testing, we did see some peaks over the 3 GHz mark, but the card doesn’t run at that speed continuously.

Like the previous Generation, RDNA3 uses Compute Unit pairs, but the new generation has some significant changes. Aside from the 20% increase in Compute units over RDNA 2, we also have updates to the L0/L1/L2 cache sizes, more SIMD32 registers for FP32 and matrix workloads, plus wider and faster interfaces.

Looking at the scheduler for RDNA 3, we now have double the floating point 32 and Matrix SIMD 32 instruction capability. This should give RDNA 3 GPUs double the raw FP32 throughput of RDNA2. We should also see double the AI Matrix throughput, although resources are shared with FP32 instructions. New to the AI units is BF16 (brain-float 16-bit) support, as well as INT4 WMMA Dot4 instructions (Wave Matrix Multiply Accumulate); AMD claims there’s an overall 2.7x increase in matrix operation speed.

The Cache system for RDNA 3 has seen a complete overhaul as well. Compared to RDNA 2, the L0 cache has doubled to 32 KB, the L1 Cache has also doubled to 256 KB, and the L2 Cache is now up to 6 MB compared to 4 MB from the last generation. The L0, L1, and L2 Cache links are also 1.5x wider than RDNA 2, improving the bandwidth.

The L3 (Infinity Cache) has shrunk compared to the last gen. It’s also divided up between the MCDs in 16 MB chunks. The RX 7900 XTX with six MCDs would see the maximum Infinity Cache of 96 MB. The 7900 XT we have here today has 80 MB of Infinity Cache. The second generation can deliver up to double the bandwidth, 3494.4 GB/s for the RX 7900 XTX, compared to 1793.5 GB/s over the last generation’s RX 6950 XT.

AMD’s RDNA 3 Compute Units evolve their high-performance ray tracing acceleration with several hardware and software optimizations. They employ methods to skip parts of the acceleration structure during transversal to execute the workload efficiently. An enhanced cache structure allows more rays to be in flight for faster performance. Overall, thanks to the new optimizations, higher frequencies, and increased number of Ray Accelerators, AMD says RDNA 3 should deliver up to a 1.8x the performance for ray tracing compared to RDNA 2.

Also, with RDNA 3, we see the introduction of AMD’s new Radiance Display Engine featuring the industry’s first Display Port 2.1 interface in a high-performance graphics card. Display Port 2.1 has a display link bandwidth of 54 Gbps doubling the display bandwidth per port over RDNA 2, supporting unprecedentedly high refresh rates.

Specifications and Features

The XFX MERC 310 RX 7900 XT BE uses the Navi 31 core measuring 300 mm² with 47.5 billion transistors. The XT variant of the RX 7900 offers 5,376 stream processors, 192 ROPs, and 336 TMUs, slightly less than the RX 7900 XTX. The 7900 XT also has 80 MB of AMD’s second-generation Infinity cache with 20 GB of GDDR6 running at 20 Gbps over a 320-bit bus for 800 GB/s total bandwidth.

Typically, the difference between the reference cards and AIBs is the cooler, power delivery, and clock speeds. In this case, the MERC 310 BLACK is spec’d at 2220 MHz Game Clock and 2560 MHz Boost Clock, which is the same as the reference design. The game clock is where AMD says the card will run most of the time but checking our results in the synthetic benchmarks; the Merc 310 7900 XT ran at or above the boost clock consistently.

XFX’s Speedster lineup for the RX 7900 XT cards is unknown because of NDAs. I’m confident, as usual, there will be multiple GPUs available at different trim levels.

Our XFX RX 7900 XT is the MERC 310 Black Edition which uses the 310 heatsink – equipped with three 100 x 15 mm fans. The new 13-blade fans are designed for high efficiency and longer life thanks to the dual ball bearings and blade shape. This combination of three fans delivers high airflow and low noise during stock operation. Once we overclocked this GPU, we tested the fans at full speed (3300 RPM), which was quite loud. Luckily when left at auto, they never reached these speeds very often.

XFX also redesigned the heatsink to improve airflow dynamics and cooling efficiency. The premium 2.7-slot aluminum heatsink has 913,966 square mm of surface area for maximum heat dissipation. We also have a full plated-copper heat spreader, vapor chamber, and six 6 mm heat pipes that snake through the heat sink. Additionally, that card uses a lot of high-quality thermal tape, aiding in thermal transfer from the core, memory, and power bits. In our cooling tests, the MERC 310 RX 7900 XT BE peaked at 58°C under load at stock speeds and was quiet.

XFX doesn’t have any additional software for overclocking, but Radeon Adrenalin works fine, and third-party utilities such as MSI afterburner for tweaking the core and memory speeds. The MERC 310 doesn’t include any controllable RGB LEDs but has a subtle backlit LED light bar​​ showing the XFX logo.

	SOLID BACKPLATE In addition to protecting the card, the solid aluminum backplate is being utilized as an additional heat spreader. Thermal pads have been added to critical locations on the backside of the PCB to transfer heat to the backplate. The backplate also features 33% more surface area when compared to the MERC319 6950 XT backplate
	FULL DIGITAL PWM POWER DELIVERY The XFX MERC 310 7900 XT Black utilizes a 13+1 phase design. The default BIOS delivers 290 watts of power to the GPU.
	Vapor Chamber and Thermal Pads The nickel-plated copper vapor chamber is utilized for both GPU and GDDR6 modules. XFX utilizes a thorough placement of thermal pads across all GDDR, VRM, and other vital components, including the backside of the PCB (a total of 20 thermal pads).
	Vapor Chamber and Heatpipes Plated Copper heat sink with a vapor chamber and six 6 mm Heatpipes offer tremendous cooling across GPU and memory.
	TRIPLE FAN DESIGN 13 Blade High-Efficiency Dual Ball Bearing Fans three 100 mm x 15 mm fans
	IT’S MORE THAN JUST DESIGN Every iteration isn’t just designed for design’s sake; we take everything we learned from the last model and fine-tune, adjust, and modify it to make our GPU design as honest as possible—nothing fancy or unnecessary; just what is needed to increase performance and improve quality.

Retail Packaging and Accessories

This slideshow requires JavaScript.

Retail packaging for the XFX MERC 310 RX 7900 XT has a vertical theme. It includes the XFX branding and MERC 310 model on the front with some features (Ray Tracing support, 20 GB VRAM, PCIe 4.0). Turning the package around offers a picture of the card and a more extensive list of features.

Inside the box, the card sits snug in form-fitting foam to keep it secure during shipping. The card is wrapped in an anti-static bag, and there are no included accessories besides the “Z” Bar support bracket pictured above.

This support bracket is included to help support the weight of this larger graphics card. The frame is fixed to the rear of the card, and the case with screws.

Meet the XFX Speedster MERC 310 RX 7900 XT Black Edition

The XFX MERC 310 Black RX 7900 XT has a clean and neat appearance. Most of the card is black, with some brushed aluminum accents around the edges. The three fans also have a brushed aluminum ring with a half-moon-type logo on the hubs. As mentioned, the MERC 310 has some backlighting on the outer edge where we have a white XFX logo. XFX has also included an Aluminum backplate with lots of thermal tape, offering excellent heat dissipation, PCB protection, and rigidity. The MERC 310 shouldn’t have any issues fitting in with most build themes.

A Closer Look at the XFX MERC 310 Black RX 7900 XT

Zooming in on the I/O layout, we’re greeted by three DisplayPorts (v2.1) and a single HDMI (v2.1) port—the latter supporting up to 4K@120Hz. The card’s maximum resolution is 8K (7680 x 4320) @ 165 Hz. While the card does not exhaust air directly out of the I/O plate, there are holes to let some warm air out. Power is sent to the GPU through two 8-pin PCIe connectors. Combined with the PCIe slot, the card provides up to 375 W of in-spec power, more than enough for this 315 W (TBP) card even while overclocked. Next to the PCIe power connectors is a dual-position BIOS switch. According to XFX, the position closest to the IO shield is the total power 290 W (TGP), while the other is a “Quiet” BIOS.

XFX uses a premium aluminum heatsink on this MERC 310 RX 7900 XT. The 2.7-slot solution uses six 6 mm heat pipes, with a total length of almost 2 meters, soldered to a plated copper vapor chamber which contacts the GPU die and memory ICs. We also have three 100 mm, 13-blade dual ball bearing fans to push plenty of air through the heatsink contained in a solid aluminum shroud that adds rigidity and aids in heat dissipation. This cooling solution pushes the card’s dimensions to 13.5 x 5.04 x 2.17 inches (344 x 128 x 55 mm).

The pictures above and below are the thermal tape that XFX has added to help cool the GPU die, MOSFETs, chokes, and memory ICs.

After entirely removing the heatsink, we can finally see the PCB. We see an unmarked Navi 31 silicon and its 522 mm² combined die area. Surrounding the GPU are 10 Hynix memory ICs (Hynix H56G42AS8) specified to run at 20 Gbps. The memory uses a 2-phase design with an 8-phase Monolithic Power Systems MP2856 controller feeding MP87997 70A MOSFETs. The GPU core is controlled via a 12-phase Monolithic Power Systems MP2857 and ten MP87997 70 A MOSFETs. There are three additional phases for the SOC using the same MOSFETs and an MP2856 controller. This is how they come up with a 13+1 phase design.

​Below are closeups of the power bits, Hynix GDDR6, and the control ICs.

This slideshow requires JavaScript.

Below is a screenshot of GPU-Z showing the clocks we achieved at stock speeds. Out of the box, the MERC 310 RX 7900 XT has a listed game clock of 2225 MHz, with the boost clocks to 2550 MHz. The card ran fairly consistently around 2600 MHz during more challenging benchmarks and, as you can see below, boosted above 2900 MHz at times.

XFX MERC 310 RX 7900 XT Black Edition on the test bench…

Test System and Benchmark Methods

Our test system is based on the latest (at the time of publishing) mainstream Intel z690 platform and uses the i9-12900K 8P,8E/24t CPU. The DRAM is in a 2×16 GB configuration at 6000 MHz with CL30 timings, a middle-of-the-road option balancing performance, and cost. The CPU runs stock.

Since the last update, we have added a few new titles and dropped some older games. More details can be found in the GPU Testing Procedure article, which we have updated with our latest benchmarks. Below is a quick summary for easy reference.

• UL 3DMark Time Spy – Default settings
• UL 3DMark Fire Strike (Extreme) – Default settings
• UL 3DMark Port Royal – Default Settings (Ray Tracing capable cards only)
• Unigine Superposition – Performance, 1080p High
• Shadow of the Tomb Raider – DX12, “Highest” preset
• The Division 2 – DX12, Ultra preset, VSync Off
• Assassin’s Creed: Odyssey – Ultra High preset, VSync Off
• Far Cry 6 – Ultra defaults
• F1 2021 – DX12, Very High defaults, Bahrain track
• Metro: Exodus – DX12, Ultra defaults

Synthetic Benchmarks

Our first set of benchmarks hails from Underwriters Laboratories, which acquired Futuremark in 2014. Earlier in 2018, a rebrand occurred, and since that time, Futuremark is now UL. The benchmarks have not changed, just the name. We chose to stick with 3DMark Fire Strike (Extreme) and 3DMark Time Spy as these tests give users a good idea of performance on modern titles. We also include 3DMark Port Royal as a Ray Tracing benchmark.

3DMark Fire Strike (Extreme) is a DX11-based test that runs at 1080p resolution. UL says the graphics are rendered with detail and complexity far beyond other DX11 benchmarks and games. 3DMark Time Spy is a DX12 benchmark designed for Windows 10 PCs. It supports new API features such as asynchronous computing, explicit multi-adapter, and multi-threading and runs at 2560×1440. 3DMark Port Royal is the first Ray Tracing benchmark designed for Windows 10 PCs and graphics cards with Microsoft DirectX Raytracing capabilities.

Looking at the charts above, the XFX MERC 310 RX 7900 XT is holding its own in 3DMark’s Fire Strike Extreme and Time Spy compared to the RTX 4080. The RX 7900 XT falls behind all the RTX 4000 cards in Port Royal, a Ray Tracing benchmark, but the RX 7900 XT is ahead of the RTX 3090 and is showing a 50% gain compared to the last generation RX 6900 XT. Just as Fire Strike Extreme seems AMD-friendly, Unigine’s Superposition is the opposite, with the XFX MERC 310 RX 7900 XT slipping behind the RTX 4080.

Gaming Benchmarks

We have updated our testing suite for gaming benchmarks to bring more modern titles into the mix. Gone are Battlefield V, F1 2020, Far Cry: New Dawn, AOTS:e, and World of Tanks, which have been replaced with Metro Exodus, Assassin’s Creed: Odyssey, F1 2021, and Far Cry 6. We kept The Division 2 and Shadow of the Tomb Raider. The games should provide a good view of the card’s overall performance. Many of these are DX12 games.

1920 x 1080 (1080p) Results

The XFX MERC 310 RX 7900 XT ran a tight race with the rest of the pack throughout most of our testing. Aside from Assassin’s Creed: Odyssey, the RX 7900 XT put on a great show, finishing at or near the top, even without the overclock.

2560 x 1440 and 4K UHD Results

Below are the higher resolution results starting with 2560 x 1440 and gaining in popularity, 3840 x 2160 (4K UHD). These resolutions are a bit of a stretch for lower-end cards, but the XFX MERC 310 RX 7900 XT proves it is more than capable of 4 K gaming.

Moving up in resolution, the XFX MERC 310 RX 7900 XT is losing ground to the Ada Lovelace cards from Nvidia. The 7900 XT performs well, maintaining 90 FPS or better throughout all of our 4 K tests, even besting the RTX 4080 in a couple of AMD-friendly titles, but it appears as the resolutions go up, the 7900 XT starts to fall behind a bit.

Ray Tracing and FSR Testing

Below, we tested Far Cry 6 with ray tracing at 1440p and 4K. The XFX MERC 310 RX 7900 XT performed very well here, but FC6 is an AMD title, so this is to be expected. One thing to note is the improvement made over the last generation RX 6900 XT. Comparing the 4K results, the RX 7900 XT shows a 50% gain over the 6900 XT.

We have also included FidelityFX Super Resolution results to give you an idea of the performance improvement it can offer.

Far Cry 6 @ 1440p with Ray Tracing FSR off

Far Cry 6 @ 1440p with Ray Tracing FSR on

Far Cry 6 @ 4K with Ray Tracing FSR off

Far Cry 6 @ 4K with Ray Tracing FSR on

Overclocking the XFX MERC 310 RX 7900 XT Black Edition

XFX doesn’t offer an overclocking utility, so we used the Adrenalin Tuning software in AMD’s Catalyst Control Center. The newer AMD cards operate in a highly dynamic way, and as such, there is no way to set a “static” operating frequency with their overclocking software. Instead, we have a lower and upper range setting for the GPU frequency. We also have sliders for adjusting the VRAM frequency and power levels and fine-tuned fan controls for custom cooling settings.

Using AMD’s Radeon software, we set out to overclock the MERC 310 RX 7900 XT. Overclocking this card took more time than expected. Typically when overclocking, you raise the frequencies until the GPU gets unstable. Stability didn’t seem to be an issue – performance was. We could raise the lower frequency limit to 2900 MHz, and the card would still run, and it also “appeared” to run at that speed when checking 3DMark logs. However, the resulting scores were terrible, typically worse than stock operation. This took a bit of time and extensive testing to settle on a minimum frequency level of 2704 MHz. We also added 100 MHz to the memory speed to finish our overclocked settings.

The XFX card peaked at 3079 MHz and consistently ran around 2750 MHz during most synthetic benchmarks. This shows the variance in overclocking the new NAVI 31; the core speed is very dependent on the load. In the end, this overclock gave us about a 3-5% increase in performance which was consistent throughout most of our benchmarks.

Temperatures and Power Use

We test power consumption by running through the game benchmarks of Shadow of the Tomb Raider and F1 2021 at stock speeds and while overclocked. We monitor temperatures throughout this testing, with the peak temperature listed in the data below. The benchmarks are extended (time) to allow the card to settle to simulate actual gaming conditions more accurately.

Temperatures on the XFX MERC 310 RX 7900 XT peaked at 58°C in F1 2021 and 56°C SOTR at stock settings. After overclocking the card, temperatures increased slightly in each game, peaking at 60°C for both titles. These excellent results demonstrate just how well XFX’s cooling solution works. The fans ramped up slowly, then settled at 50% (1600 RPM) when overclocked and were barely audible over the rest of the system.

Power use for the XFX MERC 310 RX 7900 XT peaked at 582 W (total system power) at stock, reaching 644 W while overclocked (both in SOTR). A quality 850 W power supply for most systems will be adequate for your needs, even while overclocking the card and CPU.

Conclusion

Once again, AMD has gambled with a new approach, stepping outside of the box, and that appears to have paid off for them. Applying a chiplet design to a GPU raised its own issues, such as data transfer and memory bandwidth. Using multiple MCDs solved the bandwidth issue, but data transfer between dies required something completely new, and AMD came through again with their Infinity Fanout links. Overall, their design has paid off. They wanted a high-performance graphics card for under a thousand dollars, and that’s precisely what they have.

The RX 7900 XT was close to but not quite up to the performance level of the RTX 4080, but it also comes in $300 less, assuming MSRP holds and retail prices don’t skyrocket as we’ve seen in the recent past. Comparing the 7900 XT to its predecessor, the 6900 XT, we see a massive leap in raw performance and considerable gains in the Ray Tracing arena.

XFX’s MERC 310 RX 7900 XT takes AMD’s reference design and puts its own mark on it. The MERC 310 version has a 13+2+1-phase power delivery system and 375 W of available power from the two 8-PCIe connectors. The heatsink is effective and quiet, with an appealing appearance overall. The XFX MERC 310 RX 7900 XT is a big card measuring 344 mm (13.5″) long, so always check your dimensions before purchasing. The drawback to having these features is a meager price increase. The reference model AMD RX 7900 XT’s MSRP is $899, and XFX’s MERC 310 will run you $949, a modest increase of $50 over the reference MSRP.

XFX has done a great job with the MERC 310 RX 7900 XT Black Edition, and it is worth considering if you have the money to spend. This card would excel at 1440p/144Hz and is very capable at 4K. Here at Overclocker’s, we have no qualms about giving it our seal of approval!

– Shawn Jennings (Johan45)

• MSI RTX 4090 Suprim Liquid X Review
• MSI RTX 4080 Suprim X Review
• MSI RTX 3090 Ti Review
• MSI Radeon RX 6900 XT Gaming X Trio Review
• XFX MERC319 BLACK RX 6700 XT Review
• MSI RTX 3060 Ti Gaming X Trio Review
• AMD RX 6800 XT and RX 6800 Review
• ASUS ROG Strix RTX 3080 OC Review
• MSI RTX 3090 Gaming X Trio Review