Report: GeForce GTX 750 and 750 Ti to Carry GM107 GPU

Status
Not open for further replies.
Thats amazing gtx 660 speed but half the power consumption and that is with out switching to 20nm process. The 800 series is going to be legendary.
 




Not GTX 660 speed, just in shader count and clocks (if legit). It's missing a significant amount of ROPs (1/3rd) and the memory is clocked lower to top off a 128 bit memory bus vs a 192 bit leading to a vastly smaller memory throughput. The power envelope of 75w vs 140w on the same process technology should tell you something. All the tweaking in the world isn't going to get that low and maintain the same performance as a 660.The other cards coming out (860, 870, and 880) could indeed be legendary if Project Denver is everything we hope it's cracked up to be.
 
It would be pretty impressive if the 750 Ti came with a 75W TDP, and not require a PCIe power connector. Leaked benchmarks suggest that it performs close to the 650 Ti Boost, although given the rumored specs I suspect the card is heavily memory bandwidth constrained. The 128-bit interface is supposedly clocked at just 5GHz, and the only reason I can think of for this is to keep the TDP at or below that magic 75W threshold. Wouldn't be surprised if overclocking the memory has a significant impact on performance. But in any case, if this is true the perf per W is better than I was expecting for 28nm Maxwell. If Maxwell can do this on 28nm, Nvidia might just hit their 2-3x perf per W estimates for 20nm.
 
Not GTX 660 speed, just in shader count and clocks (if legit). It's missing a significant amount of ROPs (1/3rd) and the memory is clocked lower to top off a 128 bit memory bus vs a 192 bit leading to a vastly smaller memory throughput. The power envelope of 75w vs 140w on the same process technology should tell you something. All the tweaking in the world isn't going to get that low and maintain the same performance as a 660.The other cards coming out (860, 870, and 880) could indeed be legendary if Project Denver is everything we hope it's cracked up to be.
It is even more amazing if you look at the part numbers. In kepler GK107 is mid range laptop gpu. with this card the power consumption(without going to 20nm) seems to be close to what you would find in a laptop. so if they keep the naming the same as well as the scaling interms of gk107 to gk110. The 110 had 7.5 time the cores so we could be looking at a GM110 with 7200 cores. That will be legendary.
 
..... what's so good about these? these are powerful graphics processors strangled by the 128bit memory bus. without 256bit bus (for gpus like these), you get less for money.these being on 28nm is a clear sign of nvidia being fab-blocked.
 


most gpu around this performance level have narrowed memory interface. sometimes gpu manufacturer do it on purpose so they can put it on the performance level they want
 
My ideaPad Y510 shows that my dual SLI GPU is the GK107...in a laptop... I guess I fail to understand how this is a good thing for the NON-MOBILE market. W/ K1 coming down the pipe this will change drastically for NVIDIA. I tried my damndest to find an AMD CPU + NVIDIA GPU in a laptop but they simply don't exist anymore so I had to swallon my pride and get the i7...As a person, I despise Intel and everything that they stand for. This is the first intel processor i've had to purchase in 15 years only so I could get NVIDIA..that's pretty lame.
 
Not GTX 660 speed, just in shader count and clocks (if legit). It's missing a significant amount of ROPs (1/3rd) and the memory is clocked lower to top off a 128 bit memory bus vs a 192 bit leading to a vastly smaller memory throughput. The power envelope of 75w vs 140w on the same process technology should tell you something. All the tweaking in the world isn't going to get that low and maintain the same performance as a 660.The other cards coming out (860, 870, and 880) could indeed be legendary if Project Denver is everything we hope it's cracked up to be.
It is even more amazing if you look at the part numbers. In kepler GK107 is mid range laptop gpu. with this card the power consumption(without going to 20nm) seems to be close to what you would find in a laptop. so if they keep the naming the same as well as the scaling interms of gk107 to gk110. The 110 had 7.5 time the cores so we could be looking at a GM110 with 7200 cores. That will be legendary.
Not really. If the power envelope is this low then the compute capabilities will be lackluster. The two tend to go hand in hand.I was hoping that Maxwell would cure Kepler's extremely poor compute capabilities.
 
..... what's so good about these? these are powerful graphics processors strangled by the 128bit memory bus. without 256bit bus (for gpus like these), you get less for money.these being on 28nm is a clear sign of nvidia being fab-blocked.
Nvidias cards have always had lower memory bus compared to AMD and still perform very well. memory bus aint everything. They could have some high speed cache on the gpu that makes memory bandwidth less of an issue.
 


well it is the price they have to pay to get better performance/watt. also i thought for compute parts GK110 are doing fine.
 


GK110 parts are anywhere from 1/2 to 1/10 the performance of their intended competitors in terms of Compute. But then again Kepler was a panicked direct answer to Tahiti XT. This might explain a lot.
 


why is that?
 
My ideaPad Y510 shows that my dual SLI GPU is the GK107...in a laptop... I guess I fail to understand how this is a good thing for the NON-MOBILE market. W/ K1 coming down the pipe this will change drastically for NVIDIA. I tried my damndest to find an AMD CPU + NVIDIA GPU in a laptop but they simply don't exist anymore so I had to swallon my pride and get the i7...As a person, I despise Intel and everything that they stand for. This is the first intel processor i've had to purchase in 15 years only so I could get NVIDIA..that's pretty lame.
i dont know why you would want a AMD CPU in a laptop with dedicated nvidia graphics. They are severely inferior in both performance per watt, and brute force. The only time you want a AMD laptop cpu is for the integrated graphics or because you dont have much money.
 
This doesn't seem to be a proper mid-range successor to the 650TiB at all. When the 660 came out, there were doubts about its narrow 192-bit bus; of course, for most people this proved to be fine but it meant that the 660 was not built for 1440p (same for the 650TiB which shares the same interface). People looked at GK106's relatively meager 24 ROPs and truly this was somewhat reflected in its performance.Bringing it down to 128-bit and 16 ROPs is simply disappointing, unless Nvidia shocks us all with a $100 price tag on the 750 Ti, or manages to clock it so high that the sheer clock makes up for the bus. I was hoping that all these rumors about 128-bit DDR5 on the 750 Ti would be isolated and quickly proven false; looks like I was wrong. It was so good to see the 760 equipped with a 256-bit bus that everyone (including myself) was looking forward to seeing a GK106 reincarnation with that trusty 256-bit bus of the 670, 760 and 680. Looks like it isn't going to happen and we're going to be in for a big disappointment perf-wise.
 
Status
Not open for further replies.