Nvidia’s GeForce GTX 750 Ti appears once again in the news as we see some good leaked benchmarks and the Maxwell GPU’s specs unveiled. It’s performance is on par with what was expected judging from the early benchmarks. Please remember that we might see a small improvement in the future from driver updates. It’s performance overall will be roughly on par with other cards of its budget, but certainly won’t be a revolution in the lower Graphic Card price points.
Please sir, I can have more memory bandwidth and ROPS?
The Geforce GTX 750 Ti’s configuration is fairly strange. It sports a far greater number of CUDA cores than the GTX 650 Ti (which it is in line to replace), but sticks with only 16 ROPS. Similarly we see a bump in the number of TMU’s (to 80 for the GTX 750 Ti from the 650 Ti’s 64). This will help to increase fill rate, but the lack of ROPS and the low total amount of memory bandwidth (just 86.4GB/s for the Geforce GTX 750 Ti) is going to hurt performance.
We’ll be seeing a 2GB of GDDR5 RAM, which is clocked at 1350MHZ (5400MHZ effective). We don’t know how well the memory will overclock yet, but Geforce GTX 750 Ti on sports a 128 bit bus. This means even if we see the effective clock of the memory raise to 5800MHZ, we’ll only be seeing a total bandwidth of around 92GB/s, which isn’t very impressive. High clock speeds don’t scale very well with the GTX 750 Ti’s 128-bit bus.
The rest of the cards specs pretty much what you’d expect, GPU Boost 2.0 makes an appearance, of course this being a Nvidia card supports Hardware Physx, compute, CUDA, DX and OpenGL’s latest standards. GPU Boost is an automatic ‘overclocking’ function of the card. The drivers continue to monitor the cards temps and power consumption and raises frequencies of the GPU if it feels that there’s sufficient room to do so.
In terms of raw benchmarks, judging from the leaks the Geforce GTX 750 Ti gets destroyed by the GTX 650 Ti Boost, along with AMD’s R9 260X. If these benchmarks remain accurate, then the GTX 750 Ti will hold around a 7 – 10 percent performance advantage over the GTX 650 Ti. I suspect that much of this is down to the limitations on both memory bandwidth and lack of ROPS.
| Nvidia GeForce GTX 750 Ti Leaked Specifications | ||||
|---|---|---|---|---|
| GeForce GTX 750 Ti | GeForce GTX 650 Ti Boost | GeForce GTX 650 Ti | ||
| GPU Core | 28nm GM10x | 28nm GK106 | 28nm GK106 | |
| CUDA CORES | 960 Cuda Cores | 768 Cuda Cores | 768 Cuda Cores | |
| ROPS | 16 ROPS | 16 ROPS | 16 ROPS | |
| Texture Mapping Units | 80 TMUs | 64 TMUs | 64 TMUs | |
| GPU Clock SpeedNormal / Boost | 1098 / 1176 MHz | 980 / 1032 MHz | 928 MHz | |
| Memory Clock | 1350 MHz | 1502 MHz | 1350 MHz | |
| Amount of Video RAM | 2GB GDDR5 | 2GB GDDR5 | 1GB GDDR5 | |
| Memory Bus Width | 128-bit | 192-bit | 128-bit | |
| Total Bandwidth | 86.4 GB/s | 144 GB/s | 86.4 GB/s | |
It’s too early to form a judgement yet, as we’re yet to get many benchmarks with games (although there have been some). Furthermore, it’s possible that updated drivers will improve the performance of the architecture some, but it’s doubtful that we’ll be seeing figures which drastically improve on this performance. We’re still yet to have figures for power consumption, it’ll be interesting to see how the Geforce GTX 750 Ti does compared to the GTX 650 and AMD’s own.
For anyone interested in Nvidia’s Maxwell architecture, don’t be put off with these performance benchmarks. These aren’t an indication of what the higher end parts will perform, and with the limitations on Memory Bandwidth and lack of ROPS holding the Geforce GTX 750 Ti back, it makes it less fair. Having said that, if you need a card in the here and now, it’s very possible that Mazwell for the mid or lower range, won’t be a huge leap in performance.
