NVIDIA GeForce RTX 5050 8 GB

[Dr. Dro]

There is no slot power version. It also requires an 8-pin connector. Only with a power limit of 60% of 80W could it qualify.

The original version of the 3050 didn't either, then they made the 6 GB model. It's quite possible we'll see slot power, HHHL 5050s eventually.

 

[ARF]

I'd still pick the 5050 for 250 (if I had to choose between those two cards)

This is a mistake - you always buy the card with more VRAM. 12 GB vs 8 GB is no brainer. Even if the current benchmarks lie.

 

[lexluthermiester]

There is no slot power version. It also requires an 8-pin connector. Only with a power limit of 60% of 80W could it qualify.

NVidia made two different versions of the 3050. One 8GB version that requires a PCIe power connection and a 6GB version that is slot powered. I'm looking for a slot powered version of the 5050. I wouldn't even mind if it was only 6GB. The card I'm using now in my SFF system is a 4GB 1050, so it would still be an upgrade.

Also @ARF, did I say something to make you angry? If so, please explain.

Even if the current benchmarks lie.

You are effectively calling W1z a liar. You need to stop. Your shtick is coming off like it's full on mindless drivel.

 

[silentbogo]

My biggest concern is TDP. It's an entry-entry-level card, but still needs a 6-pin to function.
Low-profile variants are going to be under question, even though I've just seen an LP version of 5060 on sale locally (has an 8[!!!]-pin power connector, which is totally bonkers, but apparently it exists)...

 

[lexluthermiester]

My biggest concern is TDP. It's an entry-entry-level card, but still needs a 6-pin to function.
Low-profile variants are going to be under question, even though I've just seen an LP version of 5060 on sale locally (has an 8[!!!]-pin power connector, which is totally bonkers, but apparently it exists)...

And I wouldn't mind having one, but I want to wait to see if they do a 5050 slot powered version. If they don't, the 5060 will likely be my choice.

 

[Dr. Dro]

This is a mistake - you always buy the card with more VRAM. 12 GB vs 8 GB is no brainer. Even if the current benchmarks lie.

Haha, no. Did you forget the part where the extra VRAM doesn't matter because the core is too slow? W1zz is offering sensible advice.

 

[ixi]

This should be a sub 200$ card...

Wanted to say 100€?

 

[:D:D]

While on the other Blackwell cards NVIDIA is blocking memory OC if it's higher than +375 MHz (+6000 MT/s), this is not the case with the RTX 5050, probably because it has GDDR6 memory instead of GDDR7.

Probably still +6000MT/s, GDDR6 would then be +750MHz. Can use API to check max and min allowable offsets.

 

[W1zzard]

GDDR6 would then be +750MHz

Which no GDDR6 can handle

Can use API to check max and min allowable offsets.

Nope, no such limits, even when the +375 limit is really active

 

[:D:D]

I was thinking of 14Gbps but it's 20Gbps.

Are you saying for instance nvmlDeviceGetClockOffsets doesn't work on that card?

Just for you, a snapshot of my old 1660 Supper with GDDR6 and mem limits of -2000MT/s to 6000MT/s

Which no GDDR6 can handle

Which basically means no real limit physically, not necessarily that it will work.

Also if we use MT/s for memory then +375MHz produces +6000MT/s for GDDR7 whilst +400MHz produces only 3200MT/s for GDDR6. So technically GDDR7 can be OC'd way higher than GDDR6 in this case.

 

[JustBenching]

Good to see the big evil monopoly offering cheap entry level cards. Well done nvidia for not ignoring the lower end of the market.

 

[W1zzard]

nvmlDeviceGetClockOffsets

Not using NVML, it's 64-bit only. But based on my other data I doubt it will show +375 as the max offset

 

[Krit]

Good to see the big evil monopoly offering cheap entry level cards. Well done nvidia for not ignoring the lower end of the market.

It's basically the same as RTX 4060 nothing new but maybe drivers will be worse this time. Ignoring is when you sell the same product for same price after two years don't sleep on that.

Where is technological improvement ? Where ?

 

[jaszy]

This should be a sub 200$ card...

In what world? Used 3060's sell quick for $220-250 on Ebay. Has 12GB VRAM, but the card is innately weaker and less power efficient.

The 5050 uses a full die 20/20 GB207 @ 150mm2. PCI 5.0 (even at x8) requires tighter SNR, which generally means increased PCB layer cost. Relatively, GB206 is only a little bigger at 181mm2. NVIDIA sells the full die 36/36 8GB variant for $380.

The 5050 really only needs to drop to $200-230 to have a competitive SM/$ ratio per Blackwell generation via 1080p. Otherwise, 5060, 9060XT, and B580 are all better options on both 1080p/1440p res.

$230 should put it more in line with the 5060. Anything closer to $200 would let it top this 1080p chart from a value perspective. Ironic.

Edit: Yes, the 5050 ends up worse at 1440p, but still sorta breaks even with 5060TI 8GB ($380) and 5070 12GB ($550) in relative frame per dollar, granted.. you obviously want 12GB+ of VRAM here.

And I wouldn't mind having one, but I want to wait to see if they do a 5050 slot powered version. If they don't, the 5060 will be my choice.

5060 is certainly a better price/perf option in the current Blackwell lineup. I doubt there will be a slot powered variant unless the card is significantly nerfed on clocks/board draw.

130W>75W is pretty big jump. Would end up energy inefficient to compete (due to a lack of raw SM units to offset).

Only thing competitive would be a higher SM thunderbolt style e-GPU via laptop spec. Ex: https://www.techpowerup.com/gpu-specs/geforce-rtx-5070-ti-mobile.c4238. I think ASUS sells one (ROG XG Mobile), but $$$

PS: This might be controversial to some people, but a 3GB IC GDDR7 variant (12 GB) of 5060 @ $300-330 would be a pretty nice card on a scale of relative value.

 

[_roman_]

Wanted to say 100€?

I wrote 70€ already. The argument it can not be made for 70€ is nonsense. There are so many consumer products with lower prices with more electronics in it. e.g smartphones have cpu, dram, storage (which a graphic card does not have), extra parts, ....

I do not understand why claiming 200 US dollar is a nice price, regardless with or without tax.

I think there can be a lot of money saved with a one slot and one fan design. This gigabyte design is just a money waste.

I wrote in teh past amd processors are sold as low with a certain price(check the cheapest am4 / am5 processors). Therefore let's be very anti consumer friendly and name 120€ with tax and shipping. Anything more is just a rip off.

graphic cards could be made for 35€ - 45€. I wrote it several times. this was the usual price. that also included a paper box with a fancy picture on top and proper description. I doubt nvidia lost money on those nvidia 710 and 730 i ordered several times for those desktop boxes. the argument the graphic card can not be made for 200 us dollar is not valid.

 

[persizi]

That card should be 75W, without 8-pin connector, single fan and preferably single slot. Now it is just a nonsense product.

 

[ARF]

You are effectively calling W1z a liar.

I didn't mean that explicitly. But that the current benchmarks don't show the future VRAM limitations which will make the 12 GB RTX 3060 card FASTER than RTX 5050

Also, I am ready to open a class - action law-suit against TPU, because obviously their reviews are extremely poorly executed and heavily pro-Nvidia biased.

Look at the tech setup. Who benches on the already very old and non-WHQL 25.3.2 BETA ?

 

[mechtech]

What’s with that vsync power use?!?

Has potential though for half height and short small cards.

@W1zzard
For the high power spikes I am assuming that’s with vsync off?

 

[ThomasK]

The price? The thermal putty? The sleeve bearing fans? The cheap plastic backplate? There's lots of potential things to criticize about this card, but I don't think Gigabyte providing an option of a big/quiet/cool card is one of those things.

With an adequately big chunk of aluminum and copper heat pipes, a dual fan option would suffice for keeping this 130w GPU cool and quiet. But probably, as you pointed out, Gigabyte is trying to compensate for the low quality of everything else, on top of the 25 dollar markup.

My opinion withstanding that this is mostly to make the card look better than it effectively is, attract the ignorant and make 'em pay for more for something unnecessary.

 

[Raiden85]

This is a $150 card at the most.

 

[dark_globe]

nothing has changed , the only card worth buying is 9060xt 16gb (there are some models priced below $400)
i consider this to be the only sub $400 "entry level" gaming card worth buying .

the next step is rtx 5070 which can already be find at $550 .

as for the sub entry level video cards priced at $250-400 with 8gb of VRAM
you are better off with a used rtx 2080Ti which you can get for the price of rtx 5050 (or less even)
and you will get a performance of the rtx 5060 with 11gb of VRAM ...

 

[jaszy]

I wrote 70€ already. The argument it can not be made for 70€ is nonsense. There are so many consumer products with lower prices with more electronics in it. e.g smartphones have cpu, dram, storage (which a graphic card does not have), extra parts, ....

I do not understand why claiming 200 US dollar is a nice price, regardless with or without tax.

I think there can be a lot of money saved with a one slot and one fan design. This gigabyte design is just a money waste.

I wrote in teh past amd processors are sold as low with a certain price(check the cheapest am4 / am5 processors). Therefore let's be very anti consumer friendly and name 120€ with tax and shipping. Anything more is just a rip off.

graphic cards could be made for 35€ - 45€. I wrote it several times. this was the usual price. that also included a paper box with a fancy picture on top and proper description. I doubt nvidia lost money on those nvidia 710 and 730 i ordered several times for those desktop boxes. the argument the graphic card can not be made for 200 us dollar is not valid.

Last get 4nm 8500G APU had MSRP of $179.00. 137mm2 via 6 core "4C", which is smaller than the 5050's GB207 full die as a monolithic config @ 149mm2.. w1zzard reviewed this specific CPU at $160 half a year into it's cycle.

AMD Ryzen 5 8500G Review - Zen 4 + Zen 4c Tested

The AMD Ryzen 5 8500G is a highly affordable Socket AM5 processor. It features the Phoenix 2 core, which combines the Zen 4 and Zen 4c architecture, for improved energy efficiency. In our review we test gaming performance with both integrated graphics and a discrete graphics card.

www.techpowerup.com

It's the only CPU you can really compare a GPU to since it's on a similar monolithic TSMC node. Everything else on AMD side is more or less irrelevant given i/o die and chiplet's skew pricing. Per Intel CPU's, anything "large" and monolithic is on Intel 10nm, which is closer to TSMC 7 in actual density.. so it really won't matter how big it is relative to TSMC 4.

The obvious difference between a CPU, is that a GPU actually requires EE design via VRM/Power and modern GPU's now require PCI 5.0, which adds cost due to layer count requirements, even at x8.

Board partners can't even hit MSRP on a 199mm2 9060XT without kickbacks from AMD. Most 16GB 9060XT's have a real MSRP of $420+ And this is only a 199mm2 die. Similar story to NVIDIA's 181mm2 5060 TI 16GB. PNY is the only vendor selling it at $430 MSRP in the US.

I think the 5050 is a little over priced, but I don't think NVIDIA or AIB's can drop below $200. Prob better off just not selling it.

A GT730 is a 87 mm2 die on TSMC 28nm.. Which is like 3k a wafer.. No offense, Some of you people live in a fantasy land.

 

[kiwi291]

For gaming it is a poor choice.
But for some applications, e.g. video surveillance (CCTV) as a card for four monitors, it can be a very nice proposition.
The RTX 50 series has a new NVDEC: 6th Gen. It will probably be about 20% faster than the NVDEC of the RTX 40 series, it looks really promising, the price is encouraging!

 

[Darmok N Jalad]

The small PCB and oversized cooler tells me a couple things.

1. This could have been manufactured to be much smaller than this, saving money in materials and packaging, which could have helped it reach a $200 price (more cards per pallet, less raw material needed).

2. Lots of cheap “gaming laptops” will feature this GPU, but it will probably be marketed as something more than a 5050.

 

[yfn_ratchet]

Lots of cheap “gaming laptops” will feature this GPU, but it will probably be marketed as something more than a 5050.

Eh, maybe? But 5050 Mobile has already been slated to use a maxed out GB207. If you wanna count it, though, RTX PRO 1000 Blackwell is probably going to fetch a far higher price for no real merit, lol.