Nvidia is making a small, power-saving tweak to its entry-level GPU. The chipmaker has refreshed the GeForce RTX 3050, which sits two places below the last-gen RTX 2060 on our GPU benchmarks hierarchy, lowering the budget card's TDP by 15W.
Nvidia had previously launched two variants of the GeForce RTX 3050 based on the GA106 silicon. The retail GeForce RTX 3050 comes with 2,560 CUDA cores, whereas the GeForce RTX 3050 OEM has 2,304 CUDA cores and slightly lower clock speeds. The third and new variant appears to use the GA107 silicon instead of the GA106 silicon while boasting a lower TDP.
The GA106 silicon has 30 Streaming Multiprocessors (SMs), amounting to 3,840 CUDA cores. It has more than enough firepower for a GeForce RTX 3050. On the other hand, the GA107 silicon is smaller and only houses 2,560. Therefore, it's a perfect fit for the GeForce RTX 3050. The switch to the GA107 silicon will allow Nvidia to reduce the production cost for a GeForce RTX 3050. Alternatively, the chipmaker may have a lot of leftover GA107 silicon that it needs to get rid of.
GeForce RTX 3050 GA107 Specifications
Header Cell - Column 0 | GeForce RTX 3050 GA107 | GeForce RTX 3050 | GeForce RTX 3050 OEM |
---|---|---|---|
GPU | GA107 | GA106 | GA106 |
CUDA Cores | 2,560 | 2,560 | 2,304 |
Tensor Cores | 80 | 80 | 72 |
RT Cores | 20 | 20 | 18 |
Base Clock (MHz) | 1,552 | 1,552 | 1,515 |
Boost Clock (MHz) | 1,777 | 1,777 | 1,755 |
FP32 Performance (TFLOPS) | 9.098 | 9.098 | 8.087 |
Memory | 8GB GDDR6 | 8GB GDDR6 | 8GB GDDR6 |
Memory Clock (Gbps) | 14 | 14 | 14 |
Memory Bus | 128 bit | 128 bit | 128 bit |
Memory Bandwidth (GBps) | 224 | 224 | 224 |
TDP (W) | 115 | 130 | 130 |
The GeForce RTX 3050 GA107 shares identical specifications tp the vanilla GeForce RTX 3050. It still has 2,560 CUDA cores, 80 Tensor cores, and 20 RT cores. Both graphics cards seemingly have the same clock speed; therefore, performance should be similar, if not equal. In addition, the memory subsystem remains intact. The GeForce RTX 3050 GA107 still wields 8GB of 14 Gbps GDDR6 memory. Complemented with a 128-bit memory interface, the graphics card can deliver a memory bandwidth that maxes out at 224 GBps. In any version, this card sits at the bottom of Nvidia's RTX stack and can't compete with the best graphics cards on the market.
The most notable difference between the GA107 and GA106 models is the TDP. The former has a 115W TDP, whereas the latter has a 130W rating. So we're looking at a 12% reduction due to the simple silicon swap. The TDP decrease isn't significant enough to warrant a change in power design; a single 6-pin power connector delivers up to 75W, and with another 75W from the PCI slot, the graphics card has 150W at its disposal. However, some manufacturers will take the opportunity to use a different power connector design.
MSI has listed the brand's new GeForce RTX 3050 Ventus 2X 8G OCV1, which sports the GA107 silicon. It has the same 1,807 MHz boost clock as the GeForce RTX 3050 Ventus 2X 8G OC and a 15W lower TDP. However, the latest OCV1 variant only relies on a 6-pin power connector compared to the regular model, which has an 8-pin power connector.
Unfortunately, we haven't found the GeForce RTX 3050 Ventus 2X 8G OCV1 in stock at any U.S. retailer for price comparison with the vanilla model. So it remains to be seen when the die swap will impact the GeForce RTX 3050's pricing.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Zhiye Liu is a news editor and memory reviewer at Tom’s Hardware. Although he loves everything that’s hardware, he has a soft spot for CPUs, GPUs, and RAM.
-
btmedic04 after the supposed $250 launch price that never happened, this feels like it's being done solely to increase profit margins as nvidia wont pass on the cost reduction of manufacture between GA106 and GA107 on to consumers. people need to stop supporting jensens leather jacket fetish with these predatory pricing models on gpus.Reply -
-Fran- Wait a dang minute... The OEM RTX 3050 is different from the DYI/AIB 3050???Reply
Why no one reported on that? Sure, the difference is very minuscule, but do not forget the mess Dell got itself into with the RTX3070 and the disabled CUDA cores via firmware/vBIOS, but nVidia has been selling, under the same name, two different specc'ed cards for OEMs and then consumers.
Why the hell do they always get a friggen' pass?
Regards. -
cfbcfb Welp, I guess I'll never look at nvidia's model numbers ever again?Reply
Just check the cuda cores, rt cores and clocks, and since that produces a thousand variants that may or may not perform in an expected manner, get the AMD upper midrange card and call it even?
Or has AMD been releasing cards with wildly varying specifications and performance too, and I missed that? -
InvalidError
The GA106 and GA107 RTX3050 have exactly the same amount of hardware enabled, performance should be very close to identical. I think the GA107 version is the one Nvidia really meant to launch and the GA106 was just a bench warmer to pad an otherwise abandoned market segment.cfbcfb said:Just check the cuda cores, rt cores and clocks, and since that produces a thousand variants that may or may not perform in an expected manner, get the AMD upper midrange card and call it even? -
artk2219 Thats is some crap that the OEM RTX 3050 is lower spec'd and no one has brought that up. I cant yell at just Nvidia for that as I remember AMD in recent histroy having two versions of the RX 580 (One of which is basically an upclocked RX 570) and RX 560 (896 and 1024 stream processor versions). Regardless, its a bad practice and we really should call them out more often when they do this crap.Reply -
RedBear87 -Fran- said:Why no one reported on that?
I'm not sure why you think that no one has reported on that/brought that up, Tom's Hardware's own reporting is linked in this article and it was reported elsewhere as well, I remember reading about it on videocardz, for instance. The interesting thing is that no one ever mentioned this OEM model again afterwards, it's been almost half a year since then, once these weird GPUs (GTX 1060 5GB anyone?) used to end up in internet cafes in Asia, but I'm not really sure if they're still relevant enough nowadays.artk2219 said:Thats is some crap that the OEM RTX 3050 is lower spec'd and no one has brought that up. -
King_V OEM variants of low end cards with even lower specs are nothing new or unusual.Reply
Odd would be if the OEM ones were equal to the retail ones. -
ManDaddio It's quite funny people miss the "OEM" mentioned. Meaning this is nothing new and people are not thinking what that means.Reply
Why are people now upset at this? This is not the first time an OEM product has lesser specs or performs worse.
If it specs as a RTX3050 OEM version then it is different.
People need to read descriptions and specs. They do it when buying TVs, cars, phones and many other things. Stop using ignorance as an excuse, people. -
InvalidError
Are you sure? I bet the bulk of people pick TVs by brand and window-shopping impression. Most phones being sold are probably whatever is "free" with two/three-years contracts unless it is a flagship phone where you are paying grossly inflated prices for brand as the main spec. If people were rational about phones, many off-brand devices are comparable to premium brands costing 3-5X as much.ManDaddio said:People need to read descriptions and specs. They do it when buying TVs, cars, phones and many other things. -
Ar558 I suppose a lower TDP is better, but the 3050 is such a bad value card that I'm not sure this really makes it more appealing. I guess it's better for those people who will buy one but anyone sane should buy something else.Reply