Lisuan LX 7G100 Performs Like RX 580, Fails 6 of 20 Windows Games

Few WindowsForum readers will buy one. PCGH notes that the LX 7G100 is not officially available outside of Asia. It is still the clearest measurement yet of how close a new, independently designed GPU architecture can get to Nvidia, AMD and Intel on Windows. It also shows why a spec sheet tells you so little about a new graphics card.

Lisuan LX 7G100 benchmarks land between the GTX 1060 and RTX 2060​

PCGH built a cut-down version of its standard performance index for the card: 14 of its 20 rasterization games, at native 1920×1080 and maximum detail. Because of that narrower scope, every card in the chart scores differently than in PCGH's regular GPU rankings. Tom's Hardware transcribed the resulting index as follows, with each card shown as a percentage of the GeForce RTX 5060.

The ratios in this table match every comparison PCGH makes in its own text. PCGH says the RTX 2060 6GB is 55% faster than the LX 7G100, and 17.2 divided by 11.1 gives exactly that. Tom's Hardware describes the same gap as the Lisuan trailing by about 35%. The two figures describe one result from opposite starting points, so they don't conflict. The RX 6600's 80% lead also checks out (20.0 against 11.1). So does Tom's "more than 2.2X" deficit to the RTX 4060. Against the RTX 5060, the LX 7G100 manages about a third of the performance.

One figure in PCGH's English-language review is worded the wrong way round. The text says the card "trails" the GTX 1060 6GB by 23%, but PCGH's own index puts the Lisuan ahead, 11.1 to 9.0, which is a 23% lead. Tom's Hardware reports it as a lead, and so does ComputerBase's separate testing. The Lisuan is faster than the GTX 1060 and only just ahead of the RX 580. PCGH sums up the position this way: "when comparing fundamentally different architectures, the raw TFLOPS figure is of limited significance", and in practice the card's gaming performance is roughly where AMD and Nvidia were in 2015.

ComputerBase's LX7G100 test confirms the gap and adds frame-pacing problems​

The two outlets shared a single imported Founders Edition card but ran their own tests with different methods, so ComputerBase's results are real corroboration and not a republished chart. Its numbers point the same way. In Full HD, the LX7G100 is far from the performance of a seven-year-old RTX 2060, with a 46 percent difference, and the current entry-level RTX 5060 is about three times as fast, the RTX 5050 still two and a half times as fast.

ComputerBase also separates out how much of the result comes from memory. The Lisuan has 12GB of VRAM, while the old Nvidia comparison cards have 6GB, and that helps in some games. With those games removed from the rating, the LX7G100 sits slightly above the average performance level of the GTX 1060. With them included, the LX7G100 is 31 percent ahead of the GTX 1060 and 32 percent behind the RTX 2060 in average FPS.

Average frame rates also flatter the card. ComputerBase found that the LX7G100 has significantly more frame-pacing problems than GeForce and Radeon. On 1%-percentile frame rates, which track stutter and are closer to how smooth a game feels, even with the memory advantage it only reaches GTX 1060 level. The magazine's conclusion is blunt: even the now clearly aging GeForce RTX 2060 is, with one exception, the clearly better product in every respect.

TrueGPU's 24.6 TFLOPS figure depends on dual-issue working​

Much of the pre-launch hype came from the LX 7G100's specifications. PCGH lists a TrueGPU-architecture chip built on TSMC's N6 process, running on PCIe 4.0 ×16, with 192 texture units, 96 ROPs, a 2,000MHz boost clock, and 12GB of GDDR6 at 18Gbps on a 192-bit bus, which gives 432GB/s of bandwidth. Board power is listed at 225W. On theoretical FP32 throughput, the LX 7G100 ranks on par with a GeForce RTX 5060 Ti or Radeon RX 9060 XT.

Tom's Hardware and other reports give the chip 6,144 shading units and about 24.5 TFLOPS. The shader count is unconfirmed. PCGH points out that Lisuan never publishes an ALU count at all, only a theoretical figure of 24.6 TFLOPS. ComputerBase agrees that the company speaks of 48 compute units, while how these are composed remains unclear, and that other sources give 6,144 FP32 ALUs, which would amount to 128 FP32 units per compute unit, an unusually high number.

PCGH's reading, based on its own synthetic tests and a feature the driver panel calls "Intelligent Dual-Issue", is that the chip has 3,072 FP32 ALUs. Under good conditions each one can process two instructions per clock, in the same way AMD designs have since RDNA 3. When dual-issue can be used, that gives 24.6 TFLOPS. When it can't, the card has 12.3 TFLOPS. AIDA64 measured 11.0 TFLOPS at stock, which fits the lower figure far better than the headline one.

This explains most of the paper-versus-practice gap. A card whose real throughput in shader-heavy games sits near 12 TFLOPS will not perform like one rated at 24. PCGH adds that shader utilization, front-end design, cache hierarchy, memory access and driver quality all decide how much theoretical compute reaches the screen, and that is especially true for a brand-new architecture.

The hardware also lacks features. PCGH lists no dedicated ray-tracing units and no matrix (AI) units. Games that require ray tracing are therefore ruled out, and so are upscalers that need matrix hardware, while vendor-agnostic options such as FSR 3.1 appear in ComputerBase's charts. Display output is DisplayPort 1.4a, and HDMI is only available through an adapter.

Windows game compatibility: 14 of 20 run, 13 run cleanly​

Tom's Hardware credits Lisuan with stable DirectX 12 gaming compatibility from launch, and says it was the first Chinese chipmaker to get WHQL certification for a graphics card. The WHQL part holds up: Lisuan's own download page lists WHQL-certified Windows drivers. No other outlet has confirmed the "first" claim. The testing gives a more mixed picture of how stable that compatibility is.

PCGH found that 14 of the 20 games in its rasterization suite worked. Two of those showed graphical glitches and one was extremely unstable. Vulkan games couldn't be measured with the usual tools: PCGH tried several versions of PresentMon and CapFrameX and still got no frame-rate logs for Baldur's Gate 3 or X4: Timelines. Baldur's Gate 3 stayed in the index by switching to its DirectX 11 fallback. X4 was dropped, though PCGH saw it running at 20–30 FPS. Ray-tracing tests were left out completely.

ComputerBase tested a different game list and reached almost the same count. Of 20 games that should have run without issue, six either failed to start or crashed immediately. Of the 14 that ran, one showed severe graphical errors and was barely playable, so just 13 of 20 games run without problems at all. The most common failure it describes is usually a black screen, with no way forward. ComputerBase compares this directly with Intel's Arc A380 launch. That card had a rough start and took months to improve, and ComputerBase says the Lisuan LX7G100 reaches a whole other level, with far more bugs and much that doesn't work at all.

Both outlets also frame this as real progress. PCGH says that two years ago, when it tested Moore Threads' MTT S80, hardly any games ran and those that did played like a slideshow. With moderately reduced detail, PCGH found the LX 7G100 smooth in Baldur's Gate 3, Battlefield 6, Cyberpunk 2077, Days Gone Remastered, Dying Light: The Beast, Forza Horizon 5 and Kingdom Come: Deliverance 2. In less demanding games it mostly beat the MTT S80 by a wide margin, though Moore Threads stayed ahead in reduced-detail Crysis and in World of Warcraft. In Dota 2 at 1440p, the Lisuan kept pace with the Radeon RX 6600 and Arc A580.

Lisuan's WHQL driver cadence and the "flatlined" performance claim​

Tom's Hardware says that despite ongoing driver work, four months after launch the card's performance "looks like it has flatlined". Lisuan's driver history supports the first part: the company is releasing drivers regularly. Its download page for the LX 7G100 lists these Windows releases:

• WHQL v29.0.2260.76 and Pilot v30.0.2260.101, both dated May 18, 2026.
• WHQL v29.0.2260.186 and Pilot v30.0.2260.214, both dated June 16, 2026.
• WHQL v29.0.2260.305 and Pilot v30.0.2260.376, both dated July 17, 2026; the listing says retail cards must use v30.0.2260.376 or later.
• WHQL v29.0.2260.515 and Pilot v30.0.2260.552, both dated August 18, 2026.

That matches Tom's count of three WHQL updates after the first release. Lisuan appears to run two tracks: a certified v29 branch and a newer, uncertified "Pilot" v30 branch released on the same day.

The "flatlined" part is harder to support. Neither review ran the card across driver versions to track performance over time. What they give is a detailed snapshot on current drivers. That snapshot is poor, but it doesn't show whether each monthly release adds a few percent. The fair reading is narrower: several months of drivers have not moved the card anywhere near the RTX 3060 12GB, which is where its bandwidth and memory configuration would suggest it could land. By PCGH's index, the RTX 3060 12GB is about twice as fast.

Overclock Mode, power draw and noise on the LX 7G100 Founders Edition​

The driver includes a one-click Overclock Mode at the bottom of its Settings section, confirmed with a second click. It warns that power use and temperature will rise but gives no figures. PCGH tested it in six games that already ran well and measured gains of at least 8%, rising to 10% in Cyberpunk 2077 and Kingdom Come: Deliverance 2, while AIDA64's FP32 reading went from 11.0 to 12.3 TFLOPS. PCGH estimates the mode raises the GPU clock to about 2.2GHz, but calls that an educated guess because Lisuan doesn't document it. ComputerBase measured a smaller average gain: with the OC switch in the driver, performance increases by seven percent.

ComputerBase's power and noise tests show where the design is weakest. The card appears to have no separate low-load power states: in dual-monitor use, a window-movement test and 4K60 SDR video, it drew exactly 59 watts in every scenario. The Founders Edition has no fan-stop mode, and the card's telemetry misreports fan speed. ComputerBase measured it at 34.5 decibels, louder than the far better Asus GeForce RTX 5060 Prime under full load. With Overclock Mode on, sustained gaming draw stayed below the 225W rating at 201W, but the LX7G100 draws up to 269 watts in load spikes, and 284 watts with the OC switch.

What the LX 7G100 results mean for buyers and PC builders​

For almost every Windows gamer, there's nothing to decide here. The card isn't officially sold outside Asia. Even where it is sold, older and cheaper cards match or beat it without the compatibility risk. A PCGH reader comment gives a German-VAT equivalent price of about €370 before import costs. That's one reader's calculation, not a verified retail price, but even at that figure the value case falls apart. The LX 7G100 matters more as a measurement of how far a new GPU architecture has come than as something to buy.

• Treat the LX 7G100 as RX 580 or GTX 1060 class in rasterized 1080p games, with its 12GB of VRAM helping only in memory-heavy titles.
• Don't judge it by its 24.6 TFLOPS rating, because PCGH's tests show real throughput closer to 12 TFLOPS unless dual-issue can be used.
• Expect about three in ten tested games to fail to start, crash or render badly, and no ray tracing, since the hardware has no RT units.
• If you do run one, use the retail-required Pilot v30.0.2260.376 or a later driver, and be aware that Vulkan games may not work with PresentMon or CapFrameX for frame-rate logging.
• Budget for about 59W at idle and load spikes near 270–285W when choosing a power supply and planning for noise, as ComputerBase measured.
• Use Overclock Mode only as a small top-up, since independent tests put the gain at 7–10% with higher power draw and an undocumented clock target.

The LX 7G100 has moved Chinese gaming GPUs from barely working to working, but slowly, and on current drivers it sits between the GTX 1060 and RTX 2060. Lisuan is shipping WHQL and Pilot drivers roughly every month, so any gains should be measurable. The reviews make clear what it will take to be competitive: fixing the six-in-twenty game failures and getting far more out of dual-issue. Monthly releases of 7–10% improvements won't close a 2x gap to the RTX 3060 on their own.