See full bio

Usman Sheikh

As a Senior Hardware Reviewer at Tech4Gamers with over a decade of hands-on experience, I dive deep into benchmarking GPUs, CPU coolers, and motherboards, pushing hardware to its limits. Passionate about overclocking and fine-tuning performance.

See full bio

Nvidia has just released the third card in its latest RTX Blackwell lineup. This is the RTX 5070, which is aimed at 1440P gaming (but with 4K a ready possibility), just like the RTX 5070 Ti. It does, however, use a less powerful GB205 chip with fairly toned-down specifications. It’s aimed at competing with AMD’s RX 9070 (XT), the older 7900 XTX, and Nvidia’s RTX 4070 Super and RTX 3090 cards.

With this, Colorful has launched the latest editions of its custom cards, with a total of four variants. The iGame Ultra W OC models, including the RTX 5080, 5070 Ti, and 5070, feature a unique white and pink theme—a design I have a lot to say about! 

I will be comparing the RTX 5070 with AMD’s RX 9070 throughout this review, as the two are very similarly priced, competing graphics cards. I have also tested and reviewed the RX 9070, make sure to check that out as well.

Key Takeaways

• The Colorful iGame RTX 5070 Ultra is a stylish mid-range GPU with solid performance at 1440p, but it’s limited memory and minor generational improvements make it a tough sell against AMD’s competition.
• You should buy the RTX 5070 if you are not interested in rasterization performance and want to play 1440p with MFG and DLSS.
• Similarly, you should skip the RTX 5070 if you want great raster performance, you play at high resolutions, or if you’re paying over $550 for the card. 

Here are the specifications:

The RTX 5070 has a couple of more Stream Multiprocessors (SMs) than its predecessor and, so, two more RT cores. This results in a slightly higher TMU, ROP, and Tensor Core Count. These core components are still dialed down in number compared to the RTX 4070 Super and Ti models, though. The Founder’s Edition GPU is priced at $550, the same as AMD’s Radeon RX 9070.

We also see more shared L2 cache and a TDP 50W higher than the predecessor; the latter part doesn’t excite us.

Architecture

The RTX 5000 cards are based on the new RTX 5000 Blackwell architecture, or Blackwell 2.0. The RTX 5070 uses the GB205 chip over the previous generation’s AD104 (there’s no GB204). Like the RTX 5080, the chip has a reduced number of transistors due to its smaller size and persistence with the same 4N node. However, this one is a significant decrease, down to 31.1b from 35.8b of the AD104. 

We get a PCIe 5.0 x16 host interface within and an upgrade to GDDR7 memory from the previous generation’s GDDR6X. This doubles bandwidth and halves the power consumption of GDDR6. The memory is clocked at a much faster 1750 MHz (vs. 1313 MHz), leading to an effective memory clock of 28 Gbps. With this, the bandwidth comes out to 672 GB/s with its 192-bit memory bus. 

Improvements

Nvidia Blackwell architecture

1 of 4

1. Stream Multiprocessors: The new Blackwell SMs contain FP32/INT32 capability on all shaders, unlike Ada, where half were capable of FP32 and the other half of both FP32 and INT32. With that, INT32 throughput is theoretically doubled. The Tensor Cores will be accessible through the shaders, so-called neural shaders.
2. Tensor Cores: The new 5th generation Tensor Cores support INT4 and FP4. These are lower-precision formats that enable higher throughput with lower memory utilization but at the loss of some precision. 
3. Ray Tracing Cores: Nvidia’s 4th generation RT cores add new components, namely, the Triangle Cluster Intersection and Triangle Cluster Decompression engines, which will enable effective ray tracing for Mega Geometry. The ray tracing cores have double the ray triangle intersect speed compared to the previous generation.
4. Display and Video Engines: With an updated Display Engine, the new cards support DisplayPort 2.1, the next-gen display output supporting speeds of up to 80 Gbps. Furthermore, high-speed hardware flip metering speeds up frame pacing. This feature is particularly helpful for Multi Frame Generation. The latest video engine supports AV1 UHQ and twice the speed of H.264 decode. Furthermore, it supports the MV-HEVC format and 4.2.2 Encode/Decode. 
5. AI-management Processor: The new AI-management processor will efficiently dedicate resources to the GPU’s Tensor cores for improved performance, especially in DLSS applications. 

Neural Rendering 

Neural rendering is a process that uses the new Neural shaders for different components of graphics rendering. This process will encompass Neural Textures, Neural Materials, Neural Volumes, Neural Radiation Fields, and Neural Radiation Cache. Leveraging this process, materials and textures can be compressed neurally, reducing their memory usage and computational cost. 

Nvidia worked with Microsoft to develop an API for DirectX (Cooperative Vectors API), that will allow developers to integrate neural techniques in their game design. This will allow better game optimization and performance. 

DLSS 4 And Reflex 

With RTX 5000, Nvidia also introduced DLSS 4. In fact, Nvidia’s marketing of RTX 5070 performance (beating the RTX 4090) was using Multi Frame Generation, a DLSS 4 feature. Speaking of which, this is a technique that can AI-generate three frames for every traditionally rendered frame. With DLSS Super Resolution set to Performance and Multi Frame Gen at 4X, 15 of 16 pixels are generated by AI, boosting performance enormously. The enhanced flip metering that we just discussed will help smoothen this process.

DLSS 4 also introduces a Transformer Model, replacing the previously used convolutional neural networks. This model is used for DLSS-powered Super Resolution, ray reconstruction, and anti-aliasing processes. The result is vastly improved image quality and even potentially more framerate through Super Resolution! What’s better is that this model is not dependent on a specific hardware component and will be supported by all RTX GPUs. 

Lastly, we have Nvidia Reflex 2, which uses Frame Warping to show you the frame on your screen quicker. This works by warping the frame that is to be generated according to the direction in which you move your mouse. You can see the output of your movement up to 75% quicker through this process. All RTX GPUs also support Reflex 2. 

Packaging and Unboxing

 The GPU is shipped inside a cardboard box with a white theme containing pink elements. 

The front side has the GPU name and the main VRAM specs. On the backside, we get some notes about the design, some platform features from Nvidia, and a diagram of the display ports on the card. 

Inside, the card is wrapped in foam packaging and further protected by an antistatic bag. There’s not much else inside: just a quick guide and a 16-pin to 2x 8-pin PCIe adapter. This will be helpful if your power supply doesn’t have the 16-pin connector but does have two 8-pin ones. 

Design

The front side reveals a white shroud with a pink outline around the frame and the fans. A close-up reveals different patterns and curved lines in between the pink outlines. We have the “Ultra” text on the bottom right with a graffiti-like font. The fans feature curved blades, a dual ball bearing design, and a smart auto-stop feature. The left and right fan frames feature the same text. Meanwhile, the central fan has the Colorful iGame logo.

This is a dual-directional design, with the left and right fans spinning anti-clockwise while the central fan spins clockwise. Thanks to this design, turbulence and noise are reduced, and airflow is improved

Moving to the backside, we have a similar-looking metallic backplate. The “Ultra” logo, in the same graffiti style, is visible but partially blocked off by the GPU’s support bracket, a lovely look. On the left side, a cutout exposes the heatsink for better heat dissipation. A sticker on the top right reveals the GPU’s serial number. 

Moving toward the top, we have the Ultra branding again, along with the iGame Ultra branding, which is RGB backlit. The lighting can be controlled by the iGame Center software and synced with your other components or peripherals. Toward the right side is the 16-pin (12V-2×6) power connector. The connector provides up to 600W, while the card is rated at 250W. 

Here, we can also see the heatsink with its densely packed fins. The cooling design consists of this heatsink and seven heat pipes running across the length of the GPU. 

On the right side, we have more Ultra branding (and curved lines!), including an embossing on the lower side. The GPU has four display ports: three DisplayPort 2.1b and one HDMI 2.1b. This is a dual-slot design. 

Test Setup

• CPU: AMD Ryzen 7 9700X, PBO Tuned, TDP 170W, Curve Optimizer -5
• RAM: XPG Lancer 16×2 6400MT/s CL28 (Tuned)
• Motherboard: MSI B650 TOMAHAWK WIFI
• Cooler: Deepcool Mystique 360 Liquid Cooler (ARGB fans)
• GPU: ZOTAC RTX 5080 | Colorful iGame Ultra RTX 5070 | ASUS Dual RTX 4070 Super | ASRock RX 9070 Steel Legend OC
• BIOS: 7D75v1L AGESA ComboAM5 1.2.0.2b
• Drivers: GeForce Game Ready 572.65 (RTX 5080, RTX 4070 Super), AMD Adrenalin-Edition-24.30.31.03 (RX 9070)
• Windows Version: Windows 11 24H2 (OS Build 26100.3194)
• Storage: WD SN 770 1TB, Samsung 970 EVO 500GB for OS
• Benchmarks: CapframeX v1.7.4, Nvidia Frame View v1.6.10, HWINFO Pro v8.22

Methodology

Testing Setup and Specifications

We tested all games and GPUs using the same settings, drivers, and software mentioned above. Every GPU ran the same version of each game, and we measured gaming performance at the game’s native resolution—without using any upscaling or frame-generation tools.

Upscaling and Anti-Aliasing

For games that don’t have their own upscaling or Temporal Anti-Aliasing (TAA), we used FSR Native because it works with every GPU. For example, this was the case with Alan Wake II.

Settings

We didn’t change any settings in the control panel; all adjustments were made directly within the game.

Performance and Measurement Tools

• CapframeX recorded performance metrics.
• Nvidia Frame View measured latency during Frame-Generation and DLSS tests on RTX GeForce graphics cards.
• HWINFO Pro tracked clock speeds and measured both average and maximum power consumption.
• NvAPP: NVIDIA shadow-play was completely disabled on RTX GeForce cards.

Benchmarking Methods

Some games used their built-in benchmarks, while others were tested with custom settings.

GPU Configuration

None of the GPUs were modified, overclocked, or had their power limits increased during the tests.

Screen Resolutions Tested

• FHD (1920×1080)
• QHD (2560×1440)
• UHD (3840×2160)

DLSS and Frame-Generation Benchmarks

For DLSS tests, we used the transformer model, and if the game supported it, frame-generation (including modes at 2x, 3x, and 4x) was also used.

Rasterization Performance

Gaming 1080p Benchmarks (Native)

Raster 1080p Performance

1 of 8

At the native 1080p raster performance, the RTX 5070 offers only 4% more performance than the RTX 4070 Super. These cards are very neck-to-neck in gaming performance at this resolution, which is quite disappointing. In Cyberpunk 2077, the lead is now 7%. In Forza Horizon 5, RX 5070 is 5% faster. In Hogwarts Legacy, the lead is 6%. However, there are no double-digit gains in the 1080p resolution whatsoever.

Gaming 1440p Benchmarks (Native)

Raster 1440p Performance

1 of 8

At the native 1440p raster performance, it’s a similar story as 1080p; the RTX 5070 is only 6% faster than the RTX 4070 Super. It is again very disappointing to see the RTX 5070 face defeat in one of the titles, Forza Horizon 5, where the RTX 4070 Super outpaced the RTX 5070 by 3%. However, Cyberpunk 2077 shows promising gains on RTX 5070, almost 19% faster than RTX 4070 Super.

Gaming 2160p Benchmarks (Native)

Raster 2160p Performance

1 of 8

At the native 2160p raster performance, RTX 5070 fails to hit the golden 60 FPS mark on average in a combined seven-game result. However, we are seeing double-digit gains at 4K resolution with RTX 5070, 33% faster in Cyberpunk 2077 and 24% faster in Hogwarts Legacy, which is impressive. If we look at the big picture, RTX 5070 is now leading by 10% as compared to RTX 4070 Super.

Ray Tracing Performance

Gaming 1080p Benchmarks (RT)

Ray-Tracing 1080p

1 of 7

Here comes the ray-tracing where GeForce GPUs truly shine; the 5070 was much slower than RX 9070 in raster, but RTX 5070 closed the ray-tracing gap. The difference is negligible; however, in some titles, such as Black Myth Wukong, RX 5070 outshines the RX 9070 by 80%, which is a massive victory. This is the only title where AMD couldn’t catch up with NVIDIA. However, RX 5070 provides 18% more performance than RTX 4070 Super in Forza Horizon 5. Overall, the RTX 4070 Super performs nearly as well as the RTX 5070.

Gaming 1440p Benchmarks (RT)

Ray-Tracing 1440p

1 of 7

At 1440p Ray-Tracing, the RTX 5070 is 9% faster than the RTX 4070 Super. Another big victory over AMD was in Black Myth Wukong; overall, the results were still very disappointing. RTX 5070 could have done better in the RT. It feels almost like copying and pasting an RTX 4070 Super but overclocked.

Gaming 2160p Benchmarks (RT)

Ray-Tracing 2160p

1 of 7

Here, we are looking at ray-tracing 2160p benchmarks; the RTX 5070 suffered due to the video memory. The 12GB is not sufficient for the RT 2160p. However, AMD has outclassed the RTX 5070 in every title except for Cyberpunk 2077, and Alan Wake II, with RTX, runs at 5 FPS; even Black Myth Wukong couldn’t save NVIDIA here; it seems like NVIDIA has hit themselves in the foot by not giving enough video ram capacity. If you want to experience 4K Ultra + RTX, we will not recommend RTX 5070. You should be looking at something higher in class; RTX 5080 would be at least required here for the smooth gaming experience

Synthetic Test Results

Synthetic Benchmarks

1 of 3

Synthetic Benchmarks

1 of 3

In 3DMark Port Royal, the GeForce RTX 5070 is 11.6% faster than the RTX 4070 Super. In 3DMark Speed Way, it delivers a 15.6% performance boost, and in 3DMark Steel Nomad, it runs 13.3% faster than the RTX 4070 Super.

DLSS 4 & Multi-Frame Generation

We’re utilizing the DLSS Transformer Model in our tests and compared it with NVIDIA’s Native DLAA, and here are some latency numbers as well; we’ll discuss the sweet spot you should be aiming at for the optimal gaming experience. To test the DLSS, we have chosen the Cyberpunk 2077 RT 1440p for our benchmarks. For your clarity: Here are the internal render resolutions DLSS uses to create 1080p, 1440p, and 4K results:

DLSS 4 ”Quality” + MFG Benchmarks

We’re using the DLSS Quality (Transformer) Model on 1440p, which means the internal render resolution on the 1440p with the ‘Quality’ preset is 66.7% and the game is being rendered at the 1707 x 960. The DLSS Quality mode is 73% faster than the Native DLAA mode. However, when we enable the frame-generation FG 2x, that’s almost 188% more performance than the DLAA, which is 2.88x faster which is incredible with the frame-generation lag of only 8ms, for some individuals it might not be very noticeable.

Here’s a structured table based on the relevant data on DLSS Quality + FG on 1440p

DLSS 4 ”Balanced” + MFG Benchmarks

In the DLSS balanced mode, we are seeing 92% performance uplift over the DLAA model, The game now has the internal resolution of 835p, with the 2x FG mode, we are seeing 3x more performance gains, 4.4x on FG 3x, and 5.6x on FG 4x.

Here’s a structured table based on the relevant data on DLSS Balanced + FG on 1440p

DLSS 4 ”Performance” + MFG Benchmarks

In the DLSS Performance Mode, we are seeing 143% performance uplift over the DLAA model, The game now has the internal resolution of 720p, with the 2x FG mode, we are seeing 3.7x more performance gains, 5.1x on FG 3x, and 6.3x on FG 4x.

Here’s a structured table based on the relevant data on DLSS Performance + FG on 1440p

The Quality Mode seems like a good sweet spot to play the games at 1440p; there are a few instances where we couldn’t see the difference vs the native resolution. However, the extra performance is worth it. We suggest that our users use the NVIDIA app and the latest preset in the game settings in the DLSS super-resolution.

However, applying the multi-frame generation is just like the cherry on top. If you’re using the 144hz Monitor, we highly recommend using the FG 2x mode, as our tests demonstrated that the average FPS is up to 130 FPS with the DLSS Quality. However, the latency penalty is roughly 8ms.

To use the frame generation as intended, you need to make sure your base FPS is in the 50 to 60 FPS range, which, in our case, we have achieved 78 FPS on average with the quality mode. If your base FPS is too low, such as below 40 FPS, you might run into extreme system lag where the game would not feel very responsive, and you might see artifacts as there aren’t enough frames for the frame generation to work with.

Overclocking Results & Dual BIOS Results

Overclocking a Colorful iGame RTX 5070 Ultra wasn’t tricky, but it was worthwhile; we gained another 10% more performance in Cyberpunk 2077 by blending the memory and core overclocks. We’re using our good old trusty MSI After-Burner to overclock the GPU. Overclocking NVIDIA cards has always been relatively easy compared to AMD GPUs. Remember that we are using the ‘Turbo’ BIOS on this GPU to overclock the card, which can be enabled by pressing the GPU and then tweaking the core and memory clocks.

How Did We Overclock An RX 5070?

1. Opened the MSI After-Burner
2. Unlock the Power-Limit, Increase it to the maximum (108%)
3. Head to the Core Clock (MHz). Increase the core clock frequency by (+300MHz)
4. Head to the Memory Clock (MHz). Increase the memory clock frequency by (+1000MHz)

OC Settings

1 of 2

Overclocking Performance Benchmarks

The performance gains are really massive here, almost double digits. We have gained another 10% more free performance in Cyberpunk 2077. NVIDIA offers good overclocking headroom for the extra performance in RTX 5070. However, the results may vary on your RTX 5070 as the silicons are not binned equally. You may have less overclocking headroom or more.

BIOS Comparison

Performance BIOS

The Colorful iGame RTX 5070 has the ‘One Key Overclock’ Button on the GPU I/O shield, which provides an extra 45MHz over the boost clock of 2512MHz. It’s basically a BIOS switch. However, the performance BIOS has a 4% extra power limit totaling 108%, as compared to the base BIOS, which has a total power limit of 104%.

Normal BIOS

The Normal BIOS comes with a power limit adjustment range of -30% to +4%; in our tests, while enabling the power BIOS, we noticed the card clocking -50MHz, lower the performance mode, and fans spinning at slightly less RPM.

Power Consumption & Clock Speed

Our sample Colorful iGame RTX 5070 Ultra has effective core clock speeds of 2,811MHz (Q) and 2,854MHz (P) in 3D Mark: Speed Way, roughly 10% more factory overclock (Q) and 12% more over the base clock of 2,511MHz, and averaging 223W (Q) and 231W (P) in power, as reported by HWINFO.

We have Cyberpunk 2077 power and clock testing numbers in the Gaming Test. The Colorful iGame RTX 5070 Ultra has an effective core clock speed of 2,787MHz (Q) and 2830MHz (P) while consuming an average of 208W (Q) and 207W (P). The performance BIOS maintains a +40-50 higher clock at a similar power, which is fantastic.

Temperatures & Fan Speed

The Colorful iGame RTX 5070 Ultra does exceptionally well in temperature but is loud at the fans. The GPU core stayed at 65c in both BIOS. However, the fans are taking extra force in the Performance BIOS, which is set to auto in both.

In the gaming test session, we peaked at 64c on the core temp on both performance and quiet BIOS. We’re impressed by the cooling here.

Should You Buy It?

Buy It If

✅You want to make the use of latest DLSS Transformer + MFG: The multi-frame generation does well on the RTX 5070 at 1440p; the 3x mode with the quality can match the high refresh rate on your monitor. The DLSS transformer model is much easier to apply as compared to FSR4

✅You want higher efficiency: While the RTX 5070 might not be a really good performer compared to the RX 9070, it draws less power in gaming and is even lower when DLSS technologies are enabled.

Don’t Buy It If

❌ You want the higher raster performance: The RTX 5070 struggles in raster benchmarks as compared to the competitor RX 9070 16GB.

❌ You intend to play at higher resolutions with RT: RX 5070 offers 12GB of video memory as compared to the competitor RX 9070 16GB.

❌ You’re spending over $550: I don’t recommend buying the RTX 5070 if it costs more than $550.

Conclusion

The launch of the RTX 5070 is questionable as it doesn’t provide much comfort for RTX 4070/Super users and lacks significant generational improvement. However, it may be a decent upgrade for RTX 3070 users. NVIDIA is placing its bet on AI features, such as DLSS and Frame Generation, in the RTX 50 series.

Unfortunately, the raw performance of the RTX 5070 is disappointing; in our tests, it closely matches that of the RTX 4070 Super. NVIDIA has made bold marketing claims about delivering RTX 4090 performance at a price of $579. In reality, they enabled multi-frame generation (4x frame generation) compared to the RTX 4090’s 2x frame generation, emphasizing AI to interpolate artificial frames to compete with the previous flagship GPU, the RTX 4090.

Another major drawback of the RTX 5070 is its 12GB of video memory. It would have been preferable if they offered at least 16GB, especially since their competitor, AMD’s RX 9070, provides 16GB at a similar MSRP.

Today’s games are increasingly memory-intensive; the card performs well at 1440p and 1080p resolutions. However, achieving 4K with ray tracing may not be feasible in most games. Availability of the RTX 5070 may also be an issue. If anyone is looking to upgrade from their old GPUs, they are not missing out much. RTX 4070 Super at the discounted price would still be a better choice, or if anyone can find the RTX 5070 at the MSRP and wants to unlock further AI capabilities such as multi-frame-generation, this card will be a great purchase for them Its overclocking potential is impressive; we achieved an additional 10% performance just by adjusting the slider.