Tag: future of ai

The RTX 5070 – the $550 GeForce 50 series GPU that Jensen confidently told gamers during the Nvidia CES keynote would deliver performance equal to the RTX 4090, the previous generation’s $1,600 flagship product. After the announcement, comments were flooded with users celebrating this exciting new generation – 4090 performance for $550? Yeah, that does sound pretty amazing.

Of course, the excitement was quickly squashed after it became clear that Nvidia was using multi-frame generation to make these extremely misleading and, frankly, false marketing claims. We were quick to point this out in our CES coverage, but many still came away from the Nvidia announcement expecting great things from the RTX 5070. Now, we finally have it, and it is unbelievably underwhelming.

The RTX 5070 is so boring, disappointing, and ultimately pointless that we are going to shorten our review. While we conducted extensive testing, we won’t go over all the individual game data because it’s highly repetitive and none of it is exciting.

But first, let’s address Nvidia’s blatant lie – the one we just mentioned – where they proudly claimed the RTX 5070 would deliver 4090 performance. Obviously, this is a massive falsehood, and it’s a damaging one that undermines features like frame generation by weaponizing them to mislead gamers.

In simple terms, Multi-Frame Generation is just a more advanced version of Single-Frame Generation. Instead of generating one frame, it can generate up to three, increasing smoothness – provided your monitor has a high enough refresh rate to display these frames. What it doesn’t do is increase performance.

In simple terms, Multi-Frame Generation is a more advanced version of Single-Frame Generation – it can increase smoothness – what it doesn’t do is increase performance.

Unlike rendering normal frames, generating frames doesn’t lower latency. In fact, if anything, the overhead of generating frames makes latency worse. No reduction in latency means no performance increase. The game won’t feel or play any faster, but it will look smoother – with some artifacts thrown in. That’s just frame smoothing, not a genuine performance boost.

So, again – generating frames doesn’t boost performance. No matter how many frames are generated, the RTX 5070 cannot outperform the RTX 4090. These comparisons must be made before enabling frame generation, so let’s quickly do that now.

RTX 4090 vs. RTX 5070

In reality, the RTX 4090 is, on average, 63% faster than the RTX 5070 across our 16-game sample at 1440p. But the deception gets even worse when we turn to ray tracing, as there are cases where the RTX 5070 doesn’t work at all due to its much more limited 12GB VRAM buffer.

One such example is Indiana Jones and the Great Circle, a game Tim strongly recommended we add to our ray tracing tests because of its excellent use of path tracing, resulting in a truly transformative experience. So, we included it in this latest round of GPU testing. The only problem – at least for the RTX 5070 – is that it can’t actually run the game under these conditions, rendering just 13 FPS on average, while the older RTX 4070 Ti Super is good for 47 fps.

This is because the RTX 5070 – a $550 GPU being released in 2025 – only comes with a measly 12GB VRAM buffer. By now, we would have hoped this kind of memory capacity was reserved for entry-level products, and that an RTX 5070 would include at least 16GB, but that’s not the case.

As a result, in this example, the RTX 4090 is 462% faster, and even frame generation can’t salvage a win for the RTX 5070. Also, keep in mind that this isn’t even native 1440p – we are enabling DLSS Quality upscaling to try and boost performance.

You can’t even load the game at 4K using these quality settings – it immediately crashes to the desktop.

The RTX 5070’s Position in the Market

Whatever opinions exist about the RTX 5090, 5080, and 5070 Ti, we can confidently say all three of those products are much better than the RTX 5070.

Things are already looking pretty bad for the RTX 5070, and while one could argue that the GeForce 50 series has been a flop so far, whatever opinions exist about the RTX 5090, 5080, and 5070 Ti, we can confidently say all three of those products are much better than the RTX 5070. For those who aren’t up to speed on what the RTX 5070 actually is, let’s quickly go over the specs.

	RTX 5070 Ti	RTX 4070 Ti Super	RTX 5070	RTX 4070 Super	RTX 4070
Price MSRP	$750	$800	$550	$600
Release Date	Feb 2025	Jan 2024	Feb 2025	Jan 2024	April 2023
Process			TSMC 4N
Die Size (mm²)	378 mm²	378.6 mm²	263 mm²	294.5 mm²
Core Config	8960 : 280 : 96	8448 : 264 : 112	6144 : 192 : 80	7168 : 224 : 80	5888 : 184 : 64
L2 Cache (MB)	48 MB			36 MB
GPU Boost Clock	2452 MHz	2610 MHz	2512 MHz	2475 MHz
Memory Capacity	16 GB			12 GB
Memory Speed	28 Gbps		28 Gbps	21 Gbps
Memory Type	GDDR7		GDDR7	GDDR6X
Bus Type / Bandwidth	256-bit, 896 GB/s	256-bit, 672 GB/s	192-bit, 672 GB/s	192-bit, 504 GB/s
Total Board Power	300W	285W	250W	220W	200W

For $550, you get a Blackwell GPU with 6,144 CUDA cores, 192 texture mapping units, and probably 80 ROPs. That’s 31% fewer cores than the 5070 Ti and just 4% more than the original RTX 4080 released two years ago.

Interestingly, it also has 14% fewer cores than the updated RTX 4070 Super, with a core clock speed that is largely the same. However, making up for the reduced core count is a 33% increase in memory bandwidth. The RTX 5070 uses 28 Gbps GDDR7 memory on a 192-bit wide memory bus, providing 672 GB/s of bandwidth – 14% fewer cores than the 4070 Super but 33% more bandwidth, and an 8% discount at MSRP.

However, as we’ve already pointed out, one aspect that remains unchanged is memory capacity. Like the original RTX 4070 and the updated 4070 Super, the RTX 5070 also only features 12GB of VRAM. While the VRAM is faster, the capacity remains the same – and no, speed or bandwidth cannot compensate for capacity. It just doesn’t work that way.

At this point, the RTX 5070 is looking like little more than a slightly discounted 4070 Super with $50 knocked off the MSRP. Of course, we have tested a variety of games to confirm this, so let’s take a look at some of them and then go over the full performance breakdown with graphs.

Test System Specs

CPU	AMD Ryzen 7 9800X3D
Motherboard	MSI MPG X870E Carbon WiFi (BIOS 7E49v1A23 – ReBAR enabled)
Memory	G.Skill Trident Z5 RGB DDR5-6000 [CL30-38-38-96]
Graphics Cards	GeForce RTX 4070 GeForce RTX 4070 Super GeForce RTX 4070 Ti GeForce RTX 4070 Ti Super GeForce RTX 4080 GeForce RTX 4080 Super GeForce RTX 4090 GeForce RTX 5070 GeForce RTX 5080 GeForce RTX 5090 Radeon RX 7700 XT Radeon RX 7800 XT Radeon RX 7900 GRE Radeon RX 7900 XT Radeon RX 7900 XTX
ATX Case	MSI MEG Maestro 700L PZ
Power Supply	MSI MPG A 1000G ATX 3.0 80 Plus Gold 1000W
Storage	MSI Spatium 1TB M470 PCIe 4.0 NVMe M.2
Operating System	Windows 11 24H2
Display Driver	Nvidia GeForce Game Ready 572.60 AMD Radeon Adrenalin 24.12.1

Benchmarks

Marvel Rivals

Starting with Marvel Rivals at 1440p, we see that the RTX 5070 is only 3% faster than the 4070 Super and just 15% faster than the two-year-old RTX 4070. We haven’t had a chance to update our 7900 GRE data, but compared to the 7800 XT, the 5070 is 16% faster and 6% slower than the 7900 XT.

At 4K, the 5070 and 4070 Super are neck and neck, delivering virtually identical performance, averaging just 44 FPS.

Stalker 2: Heart of Chornobyl

Next, we have Stalker 2, where the 5070 actually performs worse than the 4070 Super, coming in 9% slower at 1440p. This meant it was just 11% faster than the 7800 XT and 12% slower than the 7900 XT.

The 4K results were slightly better, but even then, the best that could be said is that the 5070 managed to match the 4070 Super, averaging just 33 FPS. Overall, a pretty disappointing showing.

Counter-Strike 2

Moving on to Counter-Strike 2, we see very similar results again. The RTX 5070 is essentially an RTX 4070 Super, as both delivered nearly identical performance at 1440p.

The 5070 did pull slightly ahead at 4K, but even then, it only managed a 6% performance uplift.

Cyberpunk 2077: Phantom Liberty

In Cyberpunk 2077 at 1440p, the 5070 came in slightly behind the 4070 Super, though the difference was just 2 FPS in terms of average frame rate. Still, that meant the 5070 was only 7% faster than the 7800 XT in this example.

At 4K, the 5070 performed slightly better relative to the 4070 Super, pulling ahead by a 6% margin with 51 FPS on average. This was without any RT effects enabled, using the second-highest rasterization preset.

Hogwarts Legacy

Hogwarts Legacy is a bandwidth-sensitive game, which works in the RTX 5070’s favor. At 1440p, it outperformed the 4070 Super by an impressive 21% margin, showing that under the right conditions, there are some notable gains. However, this only allowed the 5070 to match the 7800 XT.

Oddly, the margin shrank significantly at 4K, with the 5070 only 8% faster than the 4070 Super and the 7800 XT, averaging 66 FPS.

The Last of Us Part I

The Last of Us Part I delivered more typical results. At 1440p, the RTX 5070 performed identically to the 4070 Super, with both averaging around 90 FPS. This made the 5070 just 2% faster than the 7800 XT.

At 4K, the 5070 pulled slightly ahead, but the average frame rate was only 4% greater than that of the 4070 Super – a very underwhelming performance boost.

Starfield

The last game we’ll cover is Starfield, another case where the 4070 Super outperformed the new 5070 at 1440p. Granted, the performance was nearly the same, but it was still disappointing to see the 5070 coming in 3% slower.

At 4K, the 5070 was again 6% slower than the 4070 Super, averaging just 45 FPS. Another set of disappointing results.

Performance Summary

1440p

Across the 16 games tested at 1440p, the RTX 5070 was, on average, just 1% faster than the 4070 Super. That settles it – the 5070 is basically a 4070 Super. Realistically, for this to be considered a next-gen GPU, it should have been called the 5060. But we’ve been down this road already with the 5070 Ti and 5080, and to some extent, even the 5090, which is really just a 4090 Ti.

4K

It’s a similar story at 4K. The RTX 5070 was, on average, just 5% faster than the 4070 Super, delivering RTX 3090-like performance. So maybe Jensen meant the 5070 would match the 3090, not the 4090, and he simply misspoke – yeah, that must be it.

Power Consumption

We’re not going to dive too deep into power consumption, as power usage tends to be a boring topic at the best of times. However, believe it or not, the 5070 is once again a 4070 Super in this regard, consuming roughly the same amount of power across the games we tested.

Ray Tracing Performance

RT – Metro Exodus Enhanced

Now for the ray tracing benchmarks, starting with Metro Exodus Enhanced, where the 5070 matched the 4070 Super at 1440p. Well, technically, it was 4% faster, which is quickly starting to feel like a significant win for this new GeForce GPU.

At 4K, it was 8% faster, averaging 53 FPS – which is an embarrassingly low level of performance for a $550 GeForce GPU released in 2025. Of course, Radeon GPUs fare even worse, though none of them were released this year.

RT – Alan Wake II

Next, we have Alan Wake II, and the 5070 struggles at 1440p even with upscaling, only matching the original 4070 and falling 15% behind the 4070 Super. That’s a disaster, considering we’re only looking at 39 FPS.

At 4K with quality upscaling enabled, the 5070 did manage to match the 4070 Super, but at just 22 FPS on average – not exactly a win.

RT – Cyberpunk 2077

Moving on to Cyberpunk 2077: Phantom Liberty, the RTX 5070 continues its trend of extremely underwhelming ray tracing performance, coming in 12% slower than the 4070 Super with an average of 58 FPS.

Once again, the 5070 is only able to match the 4070 Super at 4K with quality upscaling, but now we’re looking at just 31 FPS on average – hardly an acceptable level of performance.

RT – Marvel’s Spider-Man

The Spider-Man Remastered results at 1440p are CPU-limited, but even so, the 5070 ended up 14% faster than the 4070 Super. Like Hogwarts Legacy, this game is very bandwidth-sensitive, which explains why the 5070 performs better here.

That said, there seems to be another bottleneck at 4K, as the 5070 drops back down to 4070 Super-like performance. While 91 FPS is still a solid result, it’s hardly impressive when compared to previous-generation GPUs.

RT – Dying Light 2 Stay Human

In Dying Light 2, the RTX 5070 managed to edge out the 4070 Super, but only by a mere 5% margin.

At 4K, the gap remained small, with the 5070 being just 7% faster at an average of 44 FPS – not exactly a game-changing improvement.

RT – Black Myth: Wukong

The RTX 5070 once again performs like a 4070 Super in Black Myth: Wukong, managing just 46 FPS at 1440p with upscaling. That said, at least it’s not a Radeon GPU.

At 4K with upscaling, the RTX 5070 becomes completely useless, only matching the RTX 4070 Super at 25 FPS.

RT – Indiana Jones and the Great Circle

Lastly, we have Indiana Jones and the Great Circle, a late addition to this review. Tim suggested including it to highlight the RT performance of these new GeForce GPUs, and it certainly paints a bleak picture for the 12GB RTX 5070 – just 13 FPS on average at 1440p with upscaling.

Of course, you can lower the quality settings to avoid maxing out the VRAM, but these are the ray tracing settings Tim recommends for a truly transformative experience in this game.

Naturally, 4K is completely out of the question. In fact, it’s so far out of the question that the game will crash and refuse to relaunch on a 12GB GPU with full RT enabled. To get back in, you’ll have to start in safe mode and lower the settings.

So, how future-proof is the RTX 5070’s RT performance – and perhaps its performance in general – as a 12GB GPU released in 2025? Not very, by the looks of it.

Ray Tracing Performance Summary

1440p

Since the RTX 5070 completely crapped out in Indiana Jones and the Great Circle with full RT enabled, we have removed that game from the average, leaving us with six titles. Even then, the 5070 still fell just behind the RTX 4070 Super – though, to be fair, overall performance was nearly identical.

4K

At 4K, the 5070 was slightly faster than the 4070 Super, but again, overall performance was nearly identical. Even with the help of upscaling, frame rates were generally poor.

Cost per Frame

MSRP

In a perfect world where MSRP actually meant something, the RTX 5070 would be a decent value product – if you overlook its obvious VRAM limitations, which will no doubt become a bigger issue in the coming years.

But ignoring that reality, the 5070 at $550 looks good – not amazing, but good. It offers 8% better value than the 5070 Ti and 20% better value than the RTX 5080. However, both of those products come with more VRAM. While it may only be the minimum amount of VRAM we’d want to see on a mid-range or better product, at least those cards meet that minimum.

The 5070 is also 13% better value than the RTX 4070 Super. Again, that’s not an amazing generational uplift, but if available at MSRP, it stacks up fairly well.

Retail

However, the RTX 4070 Super saw some small discounts in 2024. If you bought one back then, the RTX 5070 only ends up being 10% better value. At that point, we’d rather give up 10% in efficiency for an extra 6 – 12 months of use, making an RTX 4070 Super purchase a year ago the smarter choice.

Australian Retail

The RTX 5070 will have to compete with the Radeon RX 9070, so it’s likely to hit the $550 MSRP sooner rather than later. That said, in Australia where I live, the RTX 5070 is expected to cost at least $1,250 AUD, which is a terrible price point. That makes it 8% more expensive in terms of cost per frame compared to the RTX 4070 Super, which was recently available for $1,100 AUD.

It also makes the RTX 5070 worse value than the Radeon RX 7900 XTX, 7900 XT, and significantly worse than the 7800 XT. The cost per frame compared to the 7800 XT is 31% higher, despite the 5070 having less VRAM – obviously a terrible trade-off.

Operating Behavior

Before wrapping up this review, let’s take a quick look at how Nvidia’s Founders Edition version of the RTX 5070 compares to the Asus Prime and Gigabyte Eagle models.

Under full load, the FE model peaked at a GPU temperature of 72°C, while the Eagle was significantly cooler at 63°C, and the Prime ran at 62°C. It was a similar story with memory temperatures – 76°C for the FE model, while the Eagle peaked at 64°C and the Prime at 68°C.

Oddly, the FE model reported a fan speed of 2,350 RPM, yet it didn’t seem loud, measuring just 39 dBA. Typically, at that fan speed, we’d expect noise levels to be well above 40 dBA. Meanwhile, the Eagle and Prime were much quieter at 35 dBA and 34 dBA, respectively.

The Eagle consumed 20W less than the FE model, which led to a slightly lower clock speed of 2,760 MHz – odd considering this is the OC version of the Eagle, yet it clocks lower than Nvidia’s reference model. The Prime exhibited a similar pattern, though its power consumption was reported to be higher despite having a slightly lower core clock.

Bottom Line

We are not impressed with the GeForce RTX 5070. It’s a compromised product that will struggle to fully utilize the RTX feature set due to its limited 12GB VRAM buffer. And while this may not be a major issue right now, it’s likely to become one within the realistic lifespan of this product.

In terms of general performance, there’s nothing exciting here. The RTX 5070 is essentially a refreshed RTX 4070 Super with $50 knocked off the MSRP. It can occasionally compete with the RTX 4070 Ti Super, but again, with just 12GB of VRAM, it’s not a favorable comparison.

Realistically, you would have been better off buying an RTX 4070 Super months ago if you wanted this level of performance. Six months ago, the 4070 Super was available for $585, and a year ago, it was selling for $600. Waiting an entire year to save $50 while only gaining 5% more performance hardly seems worth it – another reason why the RTX 5070 is so underwhelming.

For those looking to spend around $550 on a GPU right now, there isn’t a better option – at least not yet. However, the Radeon RX 9070 series launches tomorrow, and AMD is confident they can deliver models at that price. The RX 9070 could be faster, and it certainly offers more VRAM, making it a serious competitor. We have a detailed review of the new Radeons coming, and a dedicated look at FSR4 coming up.

In other words, if you’re considering a $550 GPU purchase, it would be wise to wait a little longer to see which card is truly worth your money. That said, we suspect that spending an extra $50 on the 9070 XT will be the best move

The RTX 5070 hasn’t impressed us – in fact, it’s extremely underwhelming. We’ll wait to see what AMD brings to the table before making any recommendations at this price point.

Shopping Shortcuts:

• Nvidia GeForce RTX 5070 on Amazon
• Nvidia GeForce RTX 5070 Ti on Amazon
• Nvidia GeForce RTX 5080 on Amazon
• Nvidia GeForce RTX 5090 on Amazon
• AMD Radeon RX 7900 XTX on Amazon
• Nvidia GeForce RTX 4070 Super on Amazon
• AMD Radeon RX 7800 XT on Amazon
• AMD Radeon RX 7900 XT on Amazon