CEOInterviews.AI
Start App
Get an email when Steven Walton gives a new interview
One email per new interview. Unsubscribe in one click.
Steven Walton
Co-founder & Host, Hardware Unboxed, Hardware Unboxed

AMD Scams Gamers, Their Dodgiest GPU Ever

📅 Sep 18, 2026 Hardware Unboxed 26 MIN 296156 VIEWS 33 SEGMENTS · 1 SPEAKERS
Thermal Grizzly: https://www.thermal-grizzly.com/en/wi... https://www.thermal-grizzly.com/en/du... Support us on Patreon:   / hardwareunboxed   YT Membership:    / @hardwareunboxed   Buy relevant products from Amazon, Newegg and others below: GeForce RTX 5090 - https://geni.us/JyDFBvM GeForce RTX 5080 - https://geni.us/dRy7y07 GeForce RTX 5070 Ti - https://geni.us/AuxCu GeForce RTX 5070 - https://geni.us/Wg4vPDv GeForce RTX 5060 Ti 16GB - https://geni.us/5Kgs2I GeForce RTX 5060 - https://geni.us/xmaeB Radeon RX 9070 XT - https://geni.us/CH0CxfR Radeon RX 9070 - https://geni.us/3neQ6nF Radeon...
Steven Walton 0:05 ↗
AMD announced their latest RDNA4-based budget GPU back in July, the Radeon RX 9050, and it was suggested to cost $280 US, though it wouldn't actually be sold in the US despite getting a US MSRP. So, that's a bit odd. Anyway, not to get too caught up in that, because there is a far bigger problem with the Radeon RX 9050, and it is one that we're going to explore in today's video. And boy, did it cost me a lot of Aussie dollars to get my hands on this truly diabolical graphics card. So, in an effort to try and recoup some of those lost funds, today's sponsor spot is brought to you by Thermal Grizzly and their Wire View Pro 2, the best way to safeguard your 12V high-powered connector from melting. This little device plugs in between your graphics card's power connector and the power supply's cable to measure current, voltage, temperatures, and the power of each individual pin. It has a built-in display with configurable views for per-pin current, total power, and temperature, plus audible and visual alarms if thresholds are exceeded. The CNC machined aluminium body includes a semi-passive fan for cooling, automatic data logging, and optional automatic system shutdown via the included splitter cable for extra safety. Thermal Grizzly are so confident that the Wire View Pro 2 will keep your graphics card safe that they're backing the product with an extended warranty covering the GPU connector for two years. That is a crazy level of support. So, for more information, please check the link in the video description.
Okay, so in all seriousness, it did cost me about $1,700 Australian to get my hands on this budget entry-level graphics card, and that's because you can't buy it here in Australia. And in regions where you can buy it, you can't buy just the graphics card. That's because it's an OEM-only product, so you have to buy it in a complete PC. And to make matters worse, it is a super dodgy OEM product. I'll explain why that is very shortly. As it stands right now, you can actually purchase the Radeon RX 9050 as a standalone retail product in multiple regions. For example, over in Japan, it can be had for about 54,000 yen, and a direct currency conversion puts that at about $345 US or $490 Australian. That said though, Intel's Arc B570 can be had for 42,000 yen, which works out to be about $270 US or $380 AUD.
But what we are talking about here is the retail 8 GB version of the RX 9050. And sadly, there are two versions of this product which share the exact same name, a bit like the RX 9060 XT, which can be purchased with either 8 or 16 GB of VRAM. And as much as I don't like that, I consider it to be quite dodgy, the RX 9050 manages somehow to be even worse. And that's because the horrible 4 GB version can only be sold as an OEM product according to the rules set by AMD, meaning it can only be offered in pre-built computers. And that's how I went about getting mine. I bought CyberPowerPC's Gaming Master Gaming Desktop from Best Buy for $750 US. This PC is advertised to include AMD Radeon RX 9050 graphics, but if you pay close attention, you'll notice it only has 4 GB of GDDR6 dedicated video memory, assuming anyone buying this thing actually knows what that means.
So, what does this mean? Well, to state the obvious, it has just 4 GB of VRAM, whereas all retail versions of the 9050 come with 8 GB. This is a major problem because if anyone was to research what a Radeon RX 9050 is, they're almost guaranteed to be looking at results for the faster 8 GB model. And in fact, if you go to the official product page, that's exactly what you'll find. So, let's say a person does their due diligence and researches if the RX 9050 will be fast enough to play the games they're interested in playing. You'd find data from AMD suggesting that 60 FPS in 007 First Light is possible, almost 100 FPS in Cyberpunk 2077 using medium settings, or 206 FPS in Rainbow Six Siege. That all sounds pretty great for a budget graphics card. Sadly though, if you buy the 4 GB version, you'll be looking at significantly lower performance than what's advertised here.
The 4 GB 9050 is even worse than you might think because not only does it have just 4 GB of VRAM, half that of the retail version, but it also has half the memory bandwidth. The Navi 44 die that the RX 9050 is based on supports up to a 128-bit wide memory bus, which is achieved by using four 2 GB GDDR6 chips. In order to create a 4 GB model, AMD can only use two 2 GB GDDR6 memory chips. Each has a 32-bit wide interface, meaning two of them enables a 64-bit wide memory bus, half that of the 8 GB model, and therefore you get half the memory bandwidth. So in short, whereas the 8 GB 9050 has a memory bandwidth of 288 GB/s, the 4 GB model's been reduced to the same 144 GB/s bandwidth of the original Radeon RX 6500 XT models.
This factor alone is going to significantly reduce performance relative to the 8 GB 9050. And not only that, but most modern games are probably going to struggle with just 4 GB of VRAM. So to recap, the 4 GB version of the RX 9050 shares the exact same product name as the 8 GB model, but has half the memory bandwidth, and on top of that is being hidden away as an OEM-only product, allowing AMD to quietly palm them off to unsuspecting gamers at an undisclosed price, at least for the individual component. Of course, you will know how much you're paying for the computer that you're buying it in. The only question that remains now is just how bad is the 4 GB RX 9050?
To find out, I'm comparing it to the 8 GB version, and I've also thrown in the Arc B570 for comparison, along with the 4 and 8 GB versions of the 6500 XT. Please note, all testing was conducted on our standard GPU test system, which features the Ryzen 7 9800X3D. And this will be an absolute best-case showing for the 4 GB 9050, as this system supports PCI Express 5.0 and high-speed DDR5 memory, both of which help to mitigate the effects of running out of VRAM. In a future video, I will be testing the 4 GB RX 9050 on a much weaker system, but for today, we're giving it the best chance possible. I'm also testing every game using the lowest possible quality preset at 1080p, so this really is the best performance that you're going to be able to get out of this graphics card. Okay, let's get into it.
Starting with Forza Horizon 6, we find some pretty dismal results for the 4 GB version of the RX 9050. Using the medium preset, the 8 GB model offered 47% more performance, and then 40% more using the very low preset. Not only that, but the Arc B570 was 32% faster using the medium settings, and that margin will only grow as you increase visuals, which you could do with the B570 given it averaged 116 FPS.
Next up, we have Assassin's Creed Black Flag Resynced, and the results here are also horrible for the 4 GB 9050, as the 8 GB version was around 45% faster for both the low and ultra-low quality tests. And that means the B570 was also much faster, offering 46% greater performance with the low preset, and more crucially, it was able to average 60 FPS.
Testing with The Last of Us Part II Remastered saw the 8 GB RX 9050 offering 31 to 36% more performance when compared to the OEM 4 GB version. Even worse for AMD was the fact that Intel's Arc B570 was around 38% faster. And again, you could increase visual quality with the B570 and even max out textures as it has 10 GB of VRAM.
Moving on to Crimson Desert, we find some interesting results. Firstly, note how the 4 and 8 GB versions of the old 6500 XT delivered similar frame rates in this title, at least when using the low and minimum quality presets. This is because the game isn't using more than 4 GB of VRAM, so the 4 GB 6500 XT doesn't suffer from a lack of VRAM. However, that's not the case for the 4 GB RX 9050, as the 8 GB model was seen to be 31% faster with the low preset and even 15% faster when using the lowest possible quality settings. This performance hit isn't due to a lack of VRAM, but rather VRAM bandwidth, meaning the 4 GB 9050 will be slower than the 8 GB model even when using less than 4 GB of VRAM, and this will be true for all games. Finally, I'll just note that with the low-quality preset enabled, Intel's Arc B570 was a whopping 51% faster, rendering 68 FPS on average.
Resident Evil Requiem has what looks to be pretty typical results for us, as the 8 GB version of the RX 9050 was seen to be roughly 38% faster than the 4 GB model using the low and lowest quality presets. It has to be said, Intel's Arc B570 doesn't perform particularly well in this game, only matching the 4 GB 9050 using the lowest quality settings, though it was 20% faster with the low preset, and I expect that margin would only grow as we increase the visuals.
Next, we have 007 First Light, and this is a particularly bad example for the 4 GB RX 9050, as it was only able to render 38 FPS using the low preset, which is the lowest quality settings you can actually use in this game. So, what we have here is a modern AAA title that can't be played anywhere near 60 FPS on a brand new Radeon RX 9050 4 GB discrete graphics card at 1080p using the lowest possible quality settings. The 8 GB version was 53% faster, while the Arc B570, which averaged 59 FPS, was 55% faster with 96% greater 1% low performance. So, needless to say, the experience was significantly better with the Arc GPU. The only way to improve performance was to enable FSR balanced upscaling, and let me tell you, at 1080p it looks horrid. But it did allow the RX 9050 4 GB to reach 60 FPS. So, that's where budget PC gaming is in 2026, I guess. Thanks AMD and Nvidia.
Moving on to F1 2025. This is a game you can easily play using low-end hardware even with medium quality settings. Though here the 8 GB 9050 was still 41% faster than the 4 GB version, despite the OEM model delivering a highly playable 134 FPS. Still, in terms of value, that's probably not great given the Arc B570 was 51% faster with 202 FPS. The ultra-low preset increased the margin for the 8 GB model over the 4 GB version, the 9050 to 45%, while the Arc B570 was 40% faster. All of that said, you'd probably want to aim for high quality settings in this game given that we are already at very high frame rates.
Cities Skylines II provides us with an example of why having the bandwidth of the RX 9050 with the 4 GB model is detrimental to performance. Whereas the 4 and 8 GB versions of the 6500 XT delivered the exact same level of performance in this title, the 8 GB 9050 was 35% faster using the low settings and a massive 66% faster using the very low settings. The Arc B570 was also anywhere from 44 to 69% faster than the 4 GB 9050.
Taking a look at Space Marine 2, we see that the severe performance penalty of the 4 GB RX 9050 is on full display yet again. Using the medium quality settings at 1080p, the 8 GB version of the 9050 was roughly 46% faster, delivering 70 FPS on average compared to just 48 FPS. The Intel Arc B570 also comfortably outclassed the 4 GB 9050, coming in 29% faster with a playable 62 FPS. Dropping down to the low quality preset didn't completely save the 4 GB 9050 either. The 8 GB model maintained a commanding 37% lead by pushing 81 FPS, while the Arc B570 was roughly 19% faster when compared to the cut-down Radeon GPU with 70 FPS. Interestingly, we also saw a notable performance gap between the older 6500 XT models, with the 8 GB version outpacing the 4 GB card by 26% on average using these lower settings. This clearly indicated that the limited 4 GB buffer was severely holding back performance in this title.
Fortnite presented some very interesting scaling behavior depending on the rendering mode used. Using the DirectX 12 low-quality settings, the Arc B570 was roughly 17% faster than the 8 GB RX 9050, hitting 362 FPS, which in turn was about 20% faster than the 4 GB 9050. However, switching over to the performance mode completely transforms this data. Under these conditions, the 8 GB 9050 matched the Arc B570 blow for blow, with both GPUs delivering massive average frame rates of 528 and 538 FPS respectively. Unfortunately, the 4 GB RX 9050 couldn't keep pace in this mode, falling severely behind as the 8 GB model offered 36% more performance. This once again highlighted the crippling nature of the 4 GB 9050's memory configuration in high frame rate scenarios.
Testing with Cyberpunk 2077 Phantom Liberty showed a slightly narrower gap between the new Radeon RX 9050 models, though the 4 GB version did still struggle. At 1080p medium, the 8 GB RX 9050 was just 8% faster than the 4 GB model, yielding 80 FPS compared to 74 FPS. However, Intel's Arc B570 was the undisputed budget champ, delivering a highly playable 98 FPS, making it 22% faster than the 8 GB 9050. The 6500 XT 8 GB also comfortably beat the 4 GB variant here by about 38%, showing that memory capacity still mattered in this example. Dropping the visual quality to low allowed for all of the current generation GPUs to clear the 60 FPS hurdle easily. The B570 continued to lead the pack at 117 FPS, while the 4 GB 9050 fell marginally behind its 8 GB sibling.
Pragmata gave us a dead tie between the Arc B570 and 8 GB RX 9050 using the balanced quality preset at 1080p. Both of these GPUs averaged exactly 84 FPS with identical 1% lows of 72 FPS. The 4 GB 9050 though, it simply couldn't keep up, and as a result, the 8 GB model was 31% faster. Even when using the less demanding minimum preset, the B570 and 8 GB 9050 remain neck and neck at the top of our graph. Meanwhile, the 4 GB 9050 was left in the dust once again as the 8 GB model was 20% faster.
Moving on to Lego Batman: Legacy of the Dark Knight. The 1080p medium preset favored the Arc B570 and 8 GB RX 9050, which trade blows at 181 and 175 FPS respectively. The 4 GB 9050, however, fell off a cliff, allowing the 8 GB model to be a massive 42% faster. The VRAM capacity limits were also glaringly obvious on the older 6500 XTs, where the 8 GB version was nearly twice as fast as the 4 GB variant, rendering 112 FPS compared to just 59 FPS. Then when we drop down to the low preset, the 8 GB 9050 was able to render 311 FPS, making it 39% faster than the 4 GB version. Still, the 4 GB model was able to average 224 FPS, but it has to be said, Lego Batman looks truly awful using these quality settings.
Counter-Strike 2 provided blisteringly high frame rates, but still perfectly illustrates the hardware differences. Using the medium preset, the 8 GB RX 9050 secured a 25% performance advantage over the 4 GB model, pushing well past the 300 FPS mark. And yet again, the Arc B570 was entirely in a league of its own, averaging an incredible 368 FPS to completely smoke the Radeon lineup. We also saw the 6500 XT models tie exactly at 213 FPS, reiterating that bandwidth, not capacity, was holding back the 4 GB 9050. Then when we switch over to the low preset, the performance gap between the 9050s shrank dramatically, but the Arc B570 remained at the top of our graph with a staggering 400 FPS average, offering unbeatable value for high refresh rate competitive gaming.
Moving on to Tom Clancy's Rainbow Six Siege X, the performance deficit of the 4 GB RX 9050 was glaringly obvious here. Using the high preset at 1080p, the 8 GB version delivered an impressive 180 FPS on average, making it 62% faster than the 4 GB version. The Intel Arc B570 took top honors here, pushing 192 FPS to lead the pack. Interestingly though, the 4 GB and 8 GB versions of the older 6500 XTs offered identical performance with 92 FPS averages, proving this game wasn't spilling over the 4 GB VRAM buffer. Instead, this perfectly exposes the severe memory bandwidth limitations inherent to the 4 GB 9050. Even dropping to the low preset couldn't save the 4 GB model, as the 8 GB card remained roughly 35% faster, delivering 243 FPS compared to just 180 FPS.
Up next is Battlefield 6, and the results here were quite punishing for the 4 GB graphics cards across the board. Using the medium settings, the 8 GB RX 9050 was roughly 38% faster than its 4 GB counterpart, completely outclassing the cut-down model with 98 FPS compared to just 71 FPS. The Arc B570 looked incredibly strong here, outpacing the 8 GB 9050 by 17% with an average of 115 FPS. Switching to the low preset tightened the race considerably, but the B570 still claimed the top spot at 123 FPS, while the 8 GB 9050 was still 29% faster than the 4 GB model, and the B570 was 35% faster.
Finally, we have Marvel Rivals, which mirrors the results seen in Pragmata almost perfectly with the medium quality settings enabled. The Arc B570 and 8 GB RX 9050 were virtually indistinguishable, delivering 70 and 69 FPS respectively, with identical 56 FPS 1% lows. The 4 GB 9050 was significantly slower in this scenario, allowing the 8 GB model to boast a comfortable 43% performance advantage. Switching to the low preset naturally boosted frame rates across the board, but the overarching trend remains. The B570 and 8 GB 9050 fought tightly for first place, while the 4 GB 9050 trailed significantly, with the Arc GPU offering 40% greater performance, proving yet again that the memory bandwidth cuts are heavily detrimental to performance.
Finally, we'll wrap things up with our 18-game average. We can clearly see the brutal reality of modern gaming on a 4 GB graphics card. Looking first at the 1080p low data, the Intel Arc B570 10 GB took the overall crown, edging out the 8 GB Radeon RX 9050 by roughly 5% with a solid 112 FPS average. But the real story was the massive performance gulf between the two 9050 variants. The fully equipped 8 GB model was a staggering 37% faster than the cut-down 4 GB OEM version, firmly cementing how detrimental that sliced memory bandwidth really is. In fact, it's likely much worse than what we're showing here, given that all of this testing was conducted on our modern GPU test system, which features PCI Express 5.0 support and high-bandwidth DDR5 memory, which can help to mask just how bad performance can get when running out of VRAM.
Now, when dropping down to the lowest quality preset, that did tighten the race to the top. The 8 GB RX 9050 actually managed to squeak past the Arc B570 here by a razor-thin 3% margin, managing 147 FPS. However, the 4 GB 9050 still couldn't catch a break. The 8 GB model was 32% faster, which is quite a significant margin when looking across a reasonable range of games. And ultimately, across this range of games using mixed titles, this data proves that the heavily limited memory bandwidth and the 4 GB VRAM buffer simply doesn't cut it anymore for a smooth, consistent PC gaming experience.
So, is the 4 GB version of the RX 9050 as bad as it was made out to be, or at least as bad as it was made out to be by us? Well, yeah, I'd say it very much is. This is a very bad product. It was really shocking to see the performance drop off in a game like Rainbow Six Siege where you're not actually using anywhere near 4 GB of VRAM, it's more like 2 GB. And just the drop off was caused by the heavily reduced memory bandwidth, going from about 180 FPS with the 8 GB version down to 111 FPS with this 4 GB model. It's absolutely insane to see, especially when we saw no difference between the 4 and 8 GB versions of the 6500 XT, because those models have the same amount of memory bandwidth, it's just a difference in memory capacity. I'm sorry to say, but there's no two ways about it. The 4 GB RX 9050 is a scam, plain and simple. AMD should be embarrassed by this, and the community backlash should be swift and brutal. There is no excuse for why this shouldn't be called the RX 9040.
Whether it's a retail product or not, it does need a different name. I'm sure we'll see some people trying their hardest to defend this scam, perhaps claiming that it actually works and it works okay. Like, 111 FPS in a game like Rainbow Six Siege is still very playable and enjoyable, so an affordable product like the RX 9050 makes sense. And look, I have absolutely no issue with budget products. In fact, I really love them. They interest me far more than an RTX 5090. The RX 9050 does technically work, yes, sure, I'll give you that, but that's a pretty low bar. It's still horrible value and not a product budget gamers should even entertain. Intel's Arc B570, for example, was an average 44% faster and right now can be had for $260 US or $340 Australian.
Of course, we don't know the actual price of the 4 GB RX 9050 given this is an OEM-only product, but we can take a guess. The CyberPowerPC that I purchased for $750 US features the Ryzen 5 5500, 8 GB of DDR4 memory, an AM4 MicroATX motherboard, a 500 GB SSD, and of course the 4 GB Radeon RX 9050. The CPU can be had for $80, the lowest retail price though has been $65. A single 8 GB DDR4 module now costs around $50. You can get a MicroATX AM4 motherboard for $70, and a 500 GB SSD that'll set you back about $80. A comparable case and power supply that'll cost less than $100, but let's just say $100. And a cheap tower cooler is about $20. So, let's say the retail cost for all of that hardware is $400 US. That leaves another $350 US for the graphics card, Windows 11 Home, and then of course profit. And again, this is all based on retail pricing, not the prices that CyberPowerPC are probably getting. But based on all of that, I'm willing to bet AMD are selling the 4 GB version of the RX 9050 for at least $250 US, so about the same price as Intel's Arc B570. But of course, this is all just a guess.
The point is though, you could build this system for less and include the Arc B570, so probably do that if you're on an extreme budget. That said, I wouldn't recommend this hardware configuration. The 5500 is not a good CPU, the 5600's much better, and I'd probably get at least 16 GB of memory. But that's not really what this video is about.
Something to explore in a future video though will be the performance of the 4 GB 9050 in the CyberPowerPC, and of course I'll compare that with some other GPUs. But given that the Ryzen 5 5500 limits the graphics card to PCI Express 3.0, and then with just 8 GB of single-channel DDR4 memory, running out of VRAM will be an unmitigated disaster on that configuration. So that should be fun to witness.
Anyway, the 4 GB RX 9050 is a scam. It's designed to trick buyers and it's deceptive, so that alone makes it a scam. Of course, let me know what you think in the comments section. I'll be sure to read all of the comments and yeah, see what you guys think about this GPU, and if you're fed up with AMD and Nvidia being very anti-consumer and dodgy with products such as this one.
And that is going to do it for this video. If you liked it, you know what to do. There's a subscribe button, and we also have Patreon or the join button. So, signing up, hitting, doing the thing with either of those will give you access to our exclusive Discord server, monthly live streams, Q&A stuff, and behind-the-scenes content. So, check that out if you're interested. If not, that's perfectly fine. And of course, I would like to thank you for watching this video. I'm your host, Steve. I'll see you again next time.

Cite this transcript

APA, MLA, BibTeX
APA

Walton, S. (2026, September 18). AMD Scams Gamers, Their Dodgiest GPU Ever [Interview transcript]. Hardware Unboxed. CEOInterviews.AI. https://ceointerviews.ai/interview/3229006/

MLA

Steven Walton. "AMD Scams Gamers, Their Dodgiest GPU Ever." Hardware Unboxed, 18 Sep. 2026. Transcript, CEOInterviews.AI, https://ceointerviews.ai/interview/3229006/.

BibTeX
@misc{walton2026_3229006,
  author       = {Steven Walton},
  title        = {AMD Scams Gamers, Their Dodgiest GPU Ever},
  howpublished = {Interview transcript, Hardware Unboxed. CEOInterviews.AI},
  year         = {2026},
  month        = {sep},
  url          = {https://ceointerviews.ai/interview/3229006/},
  note         = {Speaker-attributed transcript with timestamps}
}