NVIDIA GeForce RTX 3050 6 GB

[ThomasK]

Shouldn't this be a GTX? or is it just misleading marketing?

RT performance with this card is a joke, at best.

 

[rrrrex]

No excuse for terrible naming, it should be RTX3040.

 

[N/A]

It doesn't even deserve the GTX, it's GT only. 4080 12 received enough pushback to fix it. And of course it fixed nothing beside 10% off. this is 3040 Ti at best. I miss the GT 1030 w 64 bit memory.

 

[Lew Zealand]

GeForce RTX™ 4060 OC Low Profile 8G Besonderheiten | Grafikkarten - GIGABYTE Germany

i do hope the headline is a joke "The Fastest Slot-Powered GPU"..

is a 4060 8GB with GDDR6@128 bit faster than this 3050 thing with 64bits memmory interface?

That is not a slot-powered card. From the link in your post:

Connectors 8 Pin*1

That 4060 needs PSU power and this 3060 6GB does not.

It doesn't even deserve the GTX, it's GT only. 4080 12 received enough pushback to fix it. And of course it fixed nothing beside 10% off. this is 3040 Ti at best. I miss the GT 1030 w 64 bit memory.

The 750/Ti, 1050/Ti, and 1650 were all GTXs, and even my lowly 745 is a damn GTX (lol I thought it was a GT!). So this at least gets that designation as it's the next one in the line of slot-power GPUs. And I guess it can do RT stuff to I suppose it gets the R…

I'd call it a DTX instead for DLSS as that's it's real useful feature.

GeForce DTX 3040 Ti. Perfect.

 

[Dr. Dro]

4x RT performance of 6500 XT, is there onyone who will turn on RT with this!

You'll find games like Metro Exodus are quite playable if you use a healthy amount of DLSS, I felt like the 4 GB VRAM was the biggest problem to having an OK experience with my laptop's 3050M, this card should still be a tad faster than that. It's a compromise that's perhaps not acceptable on a desktop, though.

Alright, there is often discussion about if a 4060ti or even 4070 can reasonably do RT to consider it a selling point at its price, but come on man, RT cannot be considered usable with this card. Between the already poor performance in raster and the 6GB of VRAM, it's not even worth noting.
Though it enables nvidia broadcast, so there's that I guess.

Back on the talk of the gpu,
The lack of AV1 encode means that you won't be able to use this in your jellyfin home server (or similar), so it loses on what could be an actually big selling point.

Womp womp.

I agree with you, but hey, sometimes it's the concept that counts. Knowing this is GA107, it's a good point of comparison against the 4060 that is also a 107 class chip. You'll be able to run jellyfin just fine on this though. I don't think AV1 codec support is a requirement for that.

 

[trsttte]

It doesn't even deserve the GTX, it's GT only. 4080 12 received enough pushback to fix it. And of course it fixed nothing beside 10% off. this is 3040 Ti at best. I miss the GT 1030 w 64 bit memory.

Not to mention the most important part: there was already a card called 3050 and it performs meaningfully better than this one. This is just trickery to fool unsuspecting consumers.

 

[sLowEnd]

Not to mention the most important part: there was already a card called 3050 and it performs meaningfully better than this one. This is just trickery to fool unsuspecting consumers.

It sucks, but precedent for this kind of naming goes pretty far back.

e.g. The GTX 460 768MB was cut down vs the 1GB variant, and so was the GTX 1060 3GB vs 6GB variant.

 

[Misterseb54]

Bof en terme de performance,autant mettre une RX 6600 qui l'a surpasse de très loin . Même ma gtx1660 super fait bien mieux . Alors il y a juste le tdp qui est intéressant mais pour du gaming autant prendre une carte qui a un tdp plus élevé. Par contre pour un pc multimédia, elle sera parfaite et très adapté dans des tour itx .

 

[Beermotor]

You'll find games like Metro Exodus are quite playable if you use a healthy amount of DLSS, I felt like the 4 GB VRAM was the biggest problem to having an OK experience with my laptop's 3050M, this card should still be a tad faster than that. It's a compromise that's perhaps not acceptable on a desktop, though.



I agree with you, but hey, sometimes it's the concept that counts. Knowing this is GA107, it's a good point of comparison against the 4060 that is also a 107 class chip. You'll be able to run jellyfin just fine on this though. I don't think AV1 codec support is a requirement for that.

The desktop board uses a much more cut-down GA107 than our laptop GPUs

Our laptops perform *much* better than this desktop card because they have more TMUs, (80 vs 72) ergo more CUDA cores and I believe they're also clocked higher.

Not to mention the most important part: there was already a card called 3050 and it performs meaningfully better than this one. This is just trickery to fool unsuspecting consumers.

Yep these might be the bad yields from earlier GA107 (laptop) production runs.

 

[Chrispy_]

Shouldn't this be a GTX? or is it just misleading marketing?

RT performance with this card is a joke, at best.

Iceberg Tech just did a video on this.
RT in a some games reached >30 fps at 1080p with DLSS performance (so 960x540 resolution).
Other games were 20fps slideshows.

"Acceptable RT performance" is a moving target, but in 2024 I think anything slower than a 4070 is going to struggle to maintain visual quality with RT in many games, because either the framerate will drop to unacceptable levels, or the visual clarity will drop to unacceptable levels from excessive upscaling.

 

[Dr. Dro]

The desktop board uses a much more cut-down GA107 than our laptop GPUs

Our laptops perform *much* better than this desktop card because they have more TMUs, (80 vs 72) ergo more CUDA cores and I believe they're also clocked higher.


Yep these might be the bad yields from earlier GA107 (laptop) production runs.

Do you have the Ti? Seems like it's actually a little better than the (original) 3050 mobile. Mine has this one, 80W variant

NVIDIA GeForce RTX 3050 Mobile Specs

NVIDIA GA107, 1343 MHz, 2048 Cores, 64 TMUs, 32 ROPs, 4096 MB GDDR6, 1500 MHz, 128 bit

www.techpowerup.com

Wider 128-bit bus (so slightly higher bandwidth) but lower memory capacity, also 16 instead of 20 SMs. Seems like each has a strength and a weakness in this case.

 

[theouto]

I agree with you, but hey, sometimes it's the concept that counts. Knowing this is GA107, it's a good point of comparison against the 4060 that is also a 107 class chip. You'll be able to run jellyfin just fine on this though. I don't think AV1 codec support is a requirement for that.

Oh it absolutely isn't, it has nvenc h.264, which is still quite respectable. So hey, single slot, plug and play jellyfin gpu? Not so bad, we can find uses for this card after all! AV1 would've been outstanding, but nvenc h.264 is not too shabby.

You'll find games like Metro Exodus are quite playable if you use a healthy amount of DLSS,

And I think it's limited to only metro exodus. Man that game is class at everything it does, runs well, looks well, has a very impressive and impactful RT implementation that runs better than all other RT implementations while still looking good without RT. A genuine marble, and a game that should be remembered and held up as a standard in todays age of poor performing and unremarkable looking games. (shame for the forced TAA)

This is not the only time I've gushed about metro exodus in this forum, and neither will it be the last.

 

[Macro Device]

I think anything slower than a 4070 is going to struggle to maintain visual quality with RT in many games

I'm only aware of two games with heavy RT, namely Cyberpunk 2077 and Alan Wake 2. The former proved playable (40+ FPS) with RT Reflections enabled on my RX 6700 XT (1440p, FSR: Balanced). 2080 Ti would've had no problem rendering this at 60+ FPS with DLSS on Quality. Path tracing, of course, requires something of 4070 Super onwards calibre. RT with only reflections enabled is totally fine with the same 2080 Ti in AW2 as well.

But yeah, RT performance is completely irrelevant in case of this "3050" as it's too slow for raster already.

 

[Beermotor]

Do you have the Ti? Seems like it's actually a little better than the (original) 3050 mobile. Mine has this one, 80W variant

NVIDIA GeForce RTX 3050 Mobile Specs

NVIDIA GA107, 1343 MHz, 2048 Cores, 64 TMUs, 32 ROPs, 4096 MB GDDR6, 1500 MHz, 128 bit

www.techpowerup.com

Wider 128-bit bus (so slightly higher bandwidth) but lower memory capacity, also 16 instead of 20 SMs. Seems like each has a strength and a weakness in this case.

Mine is the mid-2022 refresh version so that sounds about right.

NVIDIA GeForce RTX 3050 Mobile Refresh 6 GB Specs

NVIDIA GA107, 1492 MHz, 2560 Cores, 80 TMUs, 32 ROPs, 6144 MB GDDR6, 1750 MHz, 96 bit

www.techpowerup.com

 

[AusWolf]

Call it the 3030, make a low profile version, make it $120, and we'll call it a day.

 

[W1zzard]

The subheadings on the Conclusion page are super helpful. Is this a new formatting style? I only see it here and on the 4080 Super reviews.

Yes, this was added with RTX 4080S because I realized my conclusions get longer and longer, and attention spans of people is getting shorter and shorter

Would it make sense for, say, the Heat & Noise section of the conclusion to be on the Temperatures and Fan Noise page? Perhaps, but I prefer it this way with all the reviewer's insights in one place instead of scattered about.

I see your point, but it has almost become standard format for me to have the discussion on the conclusion page, while the previous pages mostly present data

Only TPU makes this detailed data available, and moreover it's easy to locate in every review going back at least five years.

Thanks!

Weird that the price is listed as a pro and then the price-to-performance is listed as a con. What's the point of being sub-200$ if it's a wasted 200$?

That you an actually buy something to play games with? and the money is definitely not "wasted", you can certainly enjoy games just fine

GeForce RTX™ 4060 OC Low Profile 8G Besonderheiten | Grafikkarten - GIGABYTE Germany

 

[Wasteland]

Most HTPC software is Linux-based so I'd still go with an AMD or Intel card, probably the latter if I was just doing video decode.

Re: Intel's video acceleration, I had the same thought at first. Then I remembered why I dismissed Arc for my media server--an idle power consumption of ~40 W is a tough sell for a 24/7 machine. It's a shame, because otherwise, an Arc A380 would seem to be the premiere choice for people who want to serve up varied/multiple video streams. Here's hoping that Intel works out the kinks next time.

(Worth pointing out that on top of being somewhat Linux-resistant, Nvidia's consumer offerings also have a cap on simultaneous hardware transcode jobs. In practice this probably isn't a big deal for most anyone hosting e.g. Jellyfin, but it is an area where Intel has an advantage.)

Still, you're right about the RTX 3050 6 GB. This card services what seems to be an extremely small niche at its current price point--far too expensive for basic display out and/or video decode duty, yet far too weak to be considered credible for gaming/compute. In order for this card to make sense, you have to care an awful lot about a specific combo, "max gaming with only slot power." I'm sure this card will sell well, because it's the cheapest retail offering with Nvidia branding, but I doubt that most who buy it will do so with a full understanding of what it is and what use case it best serves.

W1zzard's conclusion hit the mark: good for what it is, just too expensive.

 

[AusWolf]

Re: Intel's video acceleration, I had the same thought at first. Then I remembered why I dismissed Arc for my media server--an idle power consumption of ~40 W is a tough sell for a 24/7 machine. It's a shame, because otherwise, an Arc A380 would seem to be the premiere choice for people who want to serve up varied/multiple video streams. Here's hoping that Intel works out the kinks next time.

(Worth pointing out that on top of being somewhat Linux-resistant, Nvidia's consumer offerings also have a cap on simultaneous hardware transcode jobs. In practice this probably isn't a big deal for most anyone hosting e.g. Jellyfin, but it is an area where Intel has an advantage.)

It may be a controversial opinion, but for HTPC purposes, nothing beats an Intel Xe iGPU, in my opinion. It consumes way less power than any dGPU, and QuickSync Video has everything you need to watch your films.

 

[Wasteland]

It may be a controversial opinion, but for HTPC purposes, nothing beats an Intel Xe iGPU, in my opinion. It consumes way less power than any dGPU, and QuickSync Video has everything you need to watch your films.

Agreed. Intel IGPs are where it's at, ATM. As an added bonus, Intel CPUs idle at lower power than most current-ish Ryzen products, by quite a large margin.

I should have made the distinction between Intel's discrete GPUs and Quicksync generally more clear in my previous post. I should also make clear now that I'm not passing judgment on Ryzen's overall power efficiency, which has been highly praised by tech media and rightly so. It's just that Intel's idle power advantage looms larger when you're building a home server.

 

[Chrispy_]

I realized my conclusions get longer and longer, and attention spans of people is getting shorter and shorter

Thank you for accommodating longer conclusions instead of trying to win the affection of the TikTok crowd; I'm sure it's appreciated by many more than just me.

 

[Darmok N Jalad]

For $180, one could buy a used GPU of much better performance, and get a new PSU along with it to fix any concerns about power. The 5600XT is usually around $100, which means you might even come out ahead. The only way I could see this card being worth $180 is if was half-height, but as someone previously mentioned, that's probably not even possible to manufacture.

 

[wNotyarD]

For $180, one could buy a used GPU of much better performance, and get a new PSU along with it to fix any concerns about power. The 5600XT is usually around $100, which means you might even come out ahead. The only way I could see this card being worth $180 is if was half-height, but as someone previously mentioned, that's probably not even possible to manufacture.

It is possible.

MSI GeForce RTX™ 3050 LP 6G OC
Gigabyte GeForce RTX™ 3050 OC Low Profile 6G

 

[Deleted member 57642]

 

[Lew Zealand]

For $180, one could buy a used GPU of much better performance, and get a new PSU along with it to fix any concerns about power. The 5600XT is usually around $100, which means you might even come out ahead.

100%. I've done this in 2 different machines and the 5600 XT has been the leader in affordable price/performance on the used market for a while at ~$115 + $50 PSU. If you need to go even cheaper, the 1650 is a drop-in at about $90-100 used but at substantially lower performance. As well as total price.

The only way I could see this card being worth $180 is if was half-height, but as someone previously mentioned, that's probably not even possible to manufacture.

Both half height and slot-power only is this card's advantage (and there are some models), but still I think it needs to be $120-150. I'm gauging that on me being a person who'd like to buy one for certain drop-in GPU replacements but $180 is too steep.

 

[Fietser]

Maybe stocks running out on old cards.