There's a new budget GPU in town, the 3GB version of the GeForce GTX 1050, and boy is this one a strange animal.
Recently we checked out the disaster that is the DDR4 version of the GT 1030, a horribly misleading product that shouldn't exist. We expect that to go down as one of the worst graphics card released in history.
We don't expect this GTX 1050 3GB version to be nowhere near as bad, in fact it might not even be bad at all. However we bring up the GT 1030 to memory because it's an unnecessary odd release and a bit misleading.
Prior to May 2018 there was only one version of the GTX 1050 (code-named GP107-300) and it came with a 2GB memory buffer. That's the same GPU we've seen on retail since October 2016 when we first tested it. However, on May Nvidia quietly updated the GTX 1050 page on their website to include a 3GB model that looked to slot-in between the existing 2GB 1050 and the 4GB 1050 Ti, though it's a bit more complicated than that.
The 3GB 1050 isn't merely a 1050 with an extra gigabyte of VRAM. Instead it features 768 CUDA cores but on a smaller 96-bit memory bus and this is due to a change in the backend configuration. Ok, so it's a GTX 1050 Ti with a gigabyte less VRAM... nope not that either.
In this table you can see that while it has the same CUDA core count as the 1050 Ti it has 25% less ROPs and that means it has 25% less than the original 1050 as well. Although you get the same 7 Gbps GDDR5 memory, the memory bus is now just 96-bit due to the reduction in ROPs which sees one of the 4 memory controllers disabled. The result is just 84 GB/s of memory bandwidth, down from the 112 GB/s of the 2GB 1050 and 4GB 1050 Ti.
Additionally, this also means the L2 cache capacity has been reduced down from 1MB to 768KB. Nvidia is clocking the cores higher to compensate, but we expect that in some cases this won't help overcome the 25% reduction in memory bandwidth, ROPs and cache.
So why is Nvidia selling a 3GB version of the GTX 1050 that likely won't perform like a 1050 with an extra gigabyte of VRAM?
Given the specs we expect that for the most part it will sit between the 2GB 1050 and 1050 Ti, but there will no doubt be instances where it's actually slower than even the 2GB 1050. The most likely explanation is that they had a stockpile of defective GP107 dies that had one faulty memory controller, for example. So Nvidia is salvaging those parts by creating the 3GB 1050. It's a bit odd to see this so late in the product cycle but that's not really the issue here. What we see as an issue once again is offering multiple GPU configurations under the same name.
At the very least the 3GB version should have been called the 1050 SE, or LE, or anything other than just the 1050 3GB. It's not unlike the 3GB and 6GB GTX 1060 situation and I'm expecting to find a similar result.
On hand for testing we have the EVGA GeForce GTX 1050 SC Gaming. It comes with a small 5% factory overclock and right now it's selling for $140 which is basically what you can expect to pay for a 2GB model, making it around $30 cheaper than the GTX 1050 Ti. Our test consists of 23 games, all of which have been tested at 1080p using various quality settings. However we're only going to discuss the highlights for 8 of the games and then jump into our performance breakdown, which looks at the whole picture.
Our test bed build around Corsair's Crystal Series 570X was used. Inside we have a Core i7-8700K clocked at 5 GHz with 32GB of DDR4-3200 memory. We realize this hardware configuration is overkill for these GPUs, but as usual the idea is to remove any potential system bottleneck that might mask GPU performance. Onto the good stuff...
Benchmarks
Starting with the Battlefield 1 results we see that the 3GB 1050 is 9% faster than the 2GB version and 16% slower than the 1050 Ti. So that's a great result for the 3GB version as it is indeed faster than the original 1050 in this title. That said it is still quite a bit slower than the 1050 Ti due to the downgrades I spoke of earlier.
Despite the positive performance just seen, when testing with Wreckfest we find a very different picture. Here the 3GB 1050 was actually slower than the original 1050, 13% slower in fact and that meant it was 21% slower than the 1050 Ti. A troubling result there so let's move on.
Now when testing with Rainbow Six Siege we see that the 3GB 1050 is 22% faster than the original 2GB model and just 11% slower than the 1050 Ti. So like Battlefield 1 this is a very good result for the new 3GB model.
Okay so we're certainly looking at very mixed results here, this time when testing with Far Cry 5 the 3GB 1050 was a few frames slower than the 2GB model, making it 5% slower. This also meant it was 15% slower than the 1050 Ti.
Testing with Fortnite shows similar performance between the 2GB and 3GB 1050 models making them both only slightly slower than the 1050 Ti.
We also see comparable performance when testing with Total War Saga: Thrones of Britannia, here the 3GB model was 3% faster than the original 2GB version.
Interestingly we see an 11% reduction in performance for the 3GB 1050 when playing ARMA 3 and as a result the 1% low result dropped below 60 fps. So another poor result for the new 3GB model.
Finally we're going to look at the Warhammer Vermintide 2 results and here we see a tie between the 2GB and 3GB versions of the 1050, making them 13% slower than the 1050 Ti.
On the next page we'll be looking at power consumption and the full game benchmark results across 23 games to get a fuller picture as well as our analysis of where the 3GB GeForce GTX 1050 stands.