Sunday, September 20th 2026
AMD Radeon RX 9050 4 GB Benchmarks Surface, 8 GB Model 37% Faster
Hardware Unboxed and Toasty Bros have both tested the mysterious Radeon RX 9050 4 GB, and the results look rough. AMD only sells this version in OEM systems, so both channels had to buy CyberPowerPC prebuilts to get one. In Hardware Unboxed's 18-game test at 1080p Low, the 4 GB card averaged 78 FPS with 55 FPS 1% lows, against 107 FPS and 78 FPS for the RX 9050 8 GB. That makes the 8 GB model about 37% faster. At the lowest settings tested, it was still around 32% ahead, at 147 FPS versus 111 FPS. Both cards share 1,024 Stream Processors and up to a 2.6 GHz boost clock, so the difference comes from memory. As AMD confirmed, the 4 GB model has a 64-bit bus, 144 GB/s of bandwidth, and 16 MB of Infinity Cache, exactly half of what the 8 GB model has.
Toasty Bros opened their prebuilt and found an XFX dual-fan card with a single 8-pin connector, so it looks like ASRock is not the only partner making one. AMD has said it developed the Radeon RX 9050 4 GB version for a specific partner system aimed at entry-level gaming customers, and that it has no plans to sell it outside that OEM PC. The card was first spotted in a CyberPowerPC Gamer Master prebuilt at Best Buy, which was listed at $889.99. Keep in mind that this is still a small sample, with only two outlets testing the card so far, so it's important to wait for more independent results before drawing firm conclusions. Still, AMD launching a card with only 4 GB of VRAM in 2026 is not looking like a great decision so far.
Sources:
Hardware Unboxed, Toasty Bros, via Videocardz
Toasty Bros opened their prebuilt and found an XFX dual-fan card with a single 8-pin connector, so it looks like ASRock is not the only partner making one. AMD has said it developed the Radeon RX 9050 4 GB version for a specific partner system aimed at entry-level gaming customers, and that it has no plans to sell it outside that OEM PC. The card was first spotted in a CyberPowerPC Gamer Master prebuilt at Best Buy, which was listed at $889.99. Keep in mind that this is still a small sample, with only two outlets testing the card so far, so it's important to wait for more independent results before drawing firm conclusions. Still, AMD launching a card with only 4 GB of VRAM in 2026 is not looking like a great decision so far.


88 Comments on AMD Radeon RX 9050 4 GB Benchmarks Surface, 8 GB Model 37% Faster
The storm is almost escaping my teacup here
@TheinsanegamerN
Don't feed the troll. Take advantage of your ignore list
Some people throw rocks at others, when confronted they cry to try to make out they're the victim, more people will eventually see through the facade
I agree product shouldn't exist, but then I feel that way about a larger part of every GPU stack.
The issue has always been the naming scheme. This is not a 9050. You can not put this silicon into a normal 8 GB 9050 PCB assembly and have it work without issue. There are physical changes to the silicon (half cache, half bus), presumably as a result of binning. That's fine. But still calling it a 9050—without even a "f" or other differentiating designation—points to nothing else but deliberate obfuscation of actual hardware capabilities. I wonder why, considering the 37% performance difference in a best case scenario (1080p low).
Guess what AMD calls versions of the same CPU with different levels of cache? Different names, vastly different prices. Massive marketing campaigns drawing attention to the fact that cache changes performance, and here's why more is great.It offers at least two tiers lower performance. For direct GPU comparison, a ~37% faster 8 GB version is roughly the same jump as between the 5070 and 5080, a three figure vs a four figure cost graphics card. Having the same name and at least having relatively similar performance with different hardware would be one thing, still bad. But this is bigger than the difference between three CPU generations on two platforms—5800X3D and the 9850X3D, where there is a ~30% gaming performance difference.
Again, if AMD wants to bet on releasing a cache/bus gimped 4 GB card to the segment where buyers know the least about the hardware, during a RAM/VRAMocalypse, fine. Burn the bridges to facilitate a cheaper product, I understand the logic. Not our problem.
Call it something else.
Something else that's worth considering - VRAM and die wafer supply is limited. For every x "9050" 4 GB made, there could be more cards that aren't such borderline ewaste. The 9050 8 GB will probably still be usable five years from now if you manage settings in new games. This thing? As useless as most entry level prebuilts (not even worth upgrading them because the PSU is a semi-dangerous joke, the cooling/RAM config is a joke, the motherboard is a joke, and we've reached the point where the GPU is also a joke. Just the CPU left then).
I would not be surprised to see that current iGPU machines with enough memory outperform this "9050".
In comparision the other RX 9050 with 8GB would be an alright entry level card performance wise, albeit still not great given the nonsense $300+ price it has where its avaliable often costing more than a RX 7600 and B580 where it is avaliable.
If NVIDIA really needs to be bring up here, for some reason, the cancelled RTX 4080 12GB model was rightfully criticized by everyone to the point NVIDIA was forced to cancel it and launch it with a more proper name.
What if any reason is there to justify this RX 9050 4GB in its current state and name?Just because it's AMD people shouldn't be softer with them when they do bad products or bad decissions, they are just another big corporation and AMD isn't a struggling with negative profits smaller in importance company like they were in the early 2010s.
Regarding the RX 9050 4GB, we've said it thousands of times, yet some of you keep writing as if you're reading completely different things. You get fanatical when it involves AMD, but vanish when it involves Nvidia. You exaggerate when AMD might be mentioned in Nvidia-related discussions as a worse example, but complain if any reference to Nvidia is made in AMD topics. In Nvidia threads, you demand that people who don't own an Nvidia graphics card refrain from commenting, yet here in a thread about an AMD product, half a dozen of you who traditionally attack AMD have gathered and monopolized the thread. Like an angry crowd demanding to see a hanging.
And of course, until you inform us about the card's price, you can't say whether it's good or bad. Generally, if it was a retail product in the year 2026, it's a bad choice—we've written this a hundred times—but for a system meant to run light or older games where the buyer is trying to save even $50 off the final price, it might be a perfectly fine option. Once you find out the price, come back so we can bash AMD together. I have no issue with that.
My take is that even the 8GB version is a trash product. But as I said before and repeat now—because some of you either don't understand, simply refuse to understand, or have another agenda altogether—the VFM (value for money) solutions released years ago are dead and gone, goodbye. Because I think after the GTX 1060 and the RX 580, there haven't been good "budget" solutions. Maybe I could throw in the RX 6600 when it was selling for under $200, and the RTX 3060 12GB when it occasionally sold for just over $200. There certainly hasn't been anything worthwhile for nearly a decade priced under $100, or even under $150. Companies simply don't offer cheap VFM options anymore. Maybe only Intel under current conditions, but even they are playing above $250 now.
Nvidia has been called out several times for cases even worse than the RTX 4080 12GB. It's just that because some of you consistently stay in high-end price brackets, you think criticism of Nvidia should only happen in the high-end tier.
We shouldn't even be giving you the time of day, especially when in a single post you've proven 10 times over that you either selectively read what others write or deliberately distort it. Who here is making excuses for the RX 9050 naming scheme? Just because we point out that OEMs practically demand that the naming of products they buy in BULK doesn't imply an older or inferior product? Man, if you can't comprehend what you're reading, go to any AI and ask: "Do OEMs request that product names for bulk purchases don't imply an older or inferior product?" Because no matter what we write here, you've probably decided in advance not to accept it. As we said, the name should have been RX 9040. Everyone agrees on that, regardless of preference.
For goodness' sake, some of you have turned this place into Wccftech.
Whatever your opinion on whataboutism, this should not be called a 9050. At least you're acknowledging that.
I wonder if there'll be a class action against AMD for this product though. The ingredients are there, too bad there isn't 3.5 GB but an actual 4 on it.
Mem bandwidth might still be enough for a case.
Still and regardless... storm in teacup.
Not a gen goes by with one or more of these ridiculous products, including highly questionable label, and its been known and documented that if a GPU carries a model number below x70, buyer beware. Its a fucking minefield and AMD just added another few KM of it. Nvidia though, might still be record holder with their slew of asymmetric-memory GPUs of the past. Each and every one underperformed or went obsolete way faster than you'd want or expect from said performance level GPU.
Its quite another level to be releasing this 'thing' in 2026 though and having ultra shite performance on launch when you throw 4-5 year old content at it. Lol.Yeah about that 'case'.... I'm not seeing it. If Nvidia wants to call their x60 an x80, hell, by all means go for it.
Similarly, if AMD wants to call this thing a 9050... whatever.
Customer due diligence is a thing, and we clearly all acknowledge this could have been given a better label, much like a few hundred other GPUs in the past. That's really all this is. The net performance and its price are secondary to all of this, and really don't matter imho.
I wonder if the ingredients are there for a class action on this one, if the 4gb's memory bandwidth related limitations are not made apparent enough to the customer, perhaps. At face value it's very deceptive, but if the bandwidth is well enough disclosed somewhere in what buyers get/are provided/is available, they'll get away with it.
But... if this AMD 9050 4GB is just slow across the board... that's still 4GB doing what it does across the entire capacity. Even if it does very little. I guess I'm glad I'm not a lawyer lol
That last bit was a poking a bit of fun how things seem to be going backwards yet model numbers are being deceiving. For example what's next, are we going to go back to using recycled GDDR5 in order to get a decent capacity of VRAM? for example: 9056 for 16GB model (with GDDR5)
I'm pretty sure no one at AMD is proud of this product, but it generates money. From what I know AMD hasn't actively marketed the card either. Strictly OEM. Goes to show how much trouble Hardware Unboxed had to go through to get one of these cards. And then label it a scam. If anything, those trying to sell you these cards should be accused of scamming.
Up until Kepler it was possible to edit VBIOS to disable partitions, unfortunately there is no way to do this with Maxwell GPUs. Disabling this last partition with 512 MB of VRAM would effectively reduce it to an actual 3.5 GB GPU, but it would return the consistency that the GTX 980 ultimately had being able to run the memory up to its limit without stuttering.
I have noted good things and criticized NVIDIA equally as AMD as well as other companies in general here and in other places as applicable, because at last I'm just buying the products for my own use or a few others usage, I'm not directly working with or for those companies.Yes, with the current market conditions in general mostly good budget options are only within used or older products. Intel does offer some good options for basic systems with actual low prices and low power consumption with things like the Arc A380, however that has the caveat of needing ResizableBar so older systems can't use that ones properly if support for ReBAR isn't avaliable for them.And that's rightful for people to do as well when applicable, also in my case I don't stay in the high-end brackets always, rather usually I don't. I personally don't see any major issue with the RTX 3050 6GB currently for what it is, a lower power version and small factor version of the normal 8GB RTX 3050 and while launch price was a bit higher it eventually got more proper, also currently the fastest slot powered consumer GPU without going to workstation oriented parts like the Arc Pro B50 or some of the RTX PRO lower parts that while faster are also more expensive. One that actually was more deceitful was the 8GB RTX 3060 with no price difference and when it originally appeared the 12GB model supply was drying up, funnily enough that one seems to have been discontinued by now but the 12GB RTX 3060 has returned albeit at a too high price for what it is.If you think there's something I have assumed wrongly or something I have wrongly noted or made a mistake in, you can feel free to correct without any problems if it really is the case so. However as for the OEMs demands; ultimately configuration, pairings and naming of this product is at choice of both OEMs and AMD. And ultimately I don't think I have said anything to offend you to that point of the first thing you said here either, albeit that's your choice at the end.
In any case at the end, this is just a product for AMD and OEMs to generate some quick money while taking advantage of current market conditions, a bad and deceitful one to common costumers of pre-assembled systems without any interesting or good trait to it in its current form, but that's what it is and there is with current market conditions sadly.
---------------------While a deceiful model in nature, I don't think enough and that far to guarantee an action class lawsuit probably.As long as the company does things within legal termas and the law they can do as they wish in fact and that's what there is indeed, Criticism of such do be deserved in some such cases though too.
Just read what people are actually writing instead of what others reply to them. Because it seems to me you got carried away by what others were replying to me and didn't actually read what I had written.
To recap: An OEM requested a cheap model from AMD. AMD took GPU dies that didn't pass yield checks at TSMC's factory to be used in existing models, disabled a few CUs here, disabled a couple of memory controllers there, and created a new model. They also populated only half the memory chips on the PCB, so the 4GB version is likely (guessing here) $50 cheaper to manufacture than the 8GB version. Should it have been named differently? We've been saying that from the start. Yes. Why isn't it named differently? PROBABLY because the OEM demanded it that way. Has this been happening? For years now. Decades, even. Do we as consumers care about their reasons? NO. We want model names that reflect what we are actually buying. Is that guaranteed? Almost never, not just GPUs, but about everything with a price tag on it.
The case of the RTX 3050 6GB is identical. A narrower data bus, less memory, fewer cores, lower GPU clock speed, and a power ceiling capped at the 75W provided by the PCIe slot compared to the 130W of the 8GB version (the card actually draws 70W, but never mind giving it a 5W bonus). Should that card have been named differently? Obviously. It should have been called the RTX 3040. Why didn't Nvidia do that? So it wouldn't look like an inferior tier model. Is the naming misleading? Yes, anyone buying it thinks they're only losing 2GB of VRAM. All that talk about "the fastest slot-powered consumer GPU" is just excuses and a textbook double standard. Even as an RTX 3040, it would still have been "the fastest slot-powered consumer GPU."Hold on, has anyone written anything to the contrary? Has anyone claimed that it was just a 4GB reduction and everything else stayed the same? Everything you're writing has already been written, and now you're coming in and writing it in a way as if you want to tell us that we don't accept that the card has undergone cuts elsewhere, or that we don't accept that the card should have had a different name.
All of this has been stated dozens of times, both by me and by other people whom some of you consider to be AMD fans. So can you explain to me why you're making this post with that tone? Why write in a manner implying that all of this hasn't already been stated, analyzed, and agreed upon a dozen times over since the existence of the 4GB version first became known?
You mentioned that 6/8 GB cards also run into frame buffer issues.The issue is clear, it's both frame buffer and halved cache/bus die resources. Hence it's necessary to reiterate, once again, that this isn't just a VRAM difference. Since that's all you addressed in your post, that's what I replied to.The OEM is not the decision maker at AMD. This is not a custom product AMD is making for someone else, to be marketed under their branding, like the console chips. This is an RX 9050. Or so AMD would like you to believe. Trying to pass the responsibility away from AMD only implies that if the AMD marketing leadership is so weak, as you're suggesting, that they let SI clients dictate the naming of AMD's own lineup. Even when it doesn't make sense, like here, that's another problem in its own right.
We ripped NVIDIA a new one when they tried the 4080 12 GB. Don't expect double standards when AMD pulls a fast one that's arguably more cheeky.