EVGA Introduces Its GeForce GTX 1050 3 GB Graphics Cards

[Raevenlord]

EVGA announced the market availability of a "vanilla" GTX 1050 3 GB graphics card (the Gaming 03G) and its SuperClocked variant (the SC Gaming 03G). They both carry the same 768 CUDA cores. Clocks for the Gaming 03G are NVIDIA reference (1392 MHz base, up to 1518 MHz boost, 3 GB of 7008 MHz GDDR5 memory across a 96-bit memory bus), and the SuperClocked version brings the core clocks up to 1455MHz and 1569MHz (base and boost, respectively).

The graphics cards come in a short form factor (144.78mm x 111.15mm) - a simple, single-fan cooling solution is more than enough to cool down the GPU, which only needs PCIe power to function correctly. Display outputs are set at 1x dual-link DVI-D, 1x DisplayPort 1.4, and 1x HDMI 2.0. The Gaming 03G version is available for $159.99, while the SuperClocked version comes in at $169.99. Both feature a three-year warranty.



View at TechPowerUp Main Site

 

[R00kie]

What a ripoff.

 

[bug]

Seriously, still dual slot?

 

[RejZoR]

96bit bus. Feels like we're back to GeForce 4 era...

 

[bug]

96bit bus. Feels like we're back to GeForce 4 era...

Memory bandwidth has to be correlated with GPU processing power. You can slap a wide bus onto a weak GPU, but that's just wasteful.
The overall product however, still feels odd. I mean, EVGA is something you expect enthusiasts to buy, whereas 1050 is meant for those that don't care (or need) much. But hey, if EVGA feels there's a market for these, who am I to argue?

 

[kastriot]

Not bad for 99$, even 1$ more would be waste.

 

[Final_Fighter]

the price is way to high. maybe if it were like $120 or less i might consider. i can get rx560 cards for under $100 new if i look around enough.

 

[Caring1]

Having the space for a 6 pin power socket on the PCB implies there is a more powerful board that requires it, or the PCB is carried over from another model IMO.
If that is the case then manufacturing costs should be lower, and therefore the market price.

 

[xorbe]

But my 4GB 1050 Ti SC was only $118 last year.

 

[CheapMeat]

This with the other news about the GT 1030 getting DDR4 without a name change (to say GT 1020) shows how scummy / low effort some of the products can be. I'm assuming they're trying to use or get rid of some materials in stock. Also, like someone else said, they should just be 1 slot, to even bother to make it worthwhile.

 

[bug]

Having the space for a 6 pin power socket on the PCB implies there is a more powerful board that requires it, or the PCB is carried over from another model IMO.
If that is the case then manufacturing costs should be lower, and therefore the market price.

But my 4GB 1050 Ti SC was only $118 last year.

This with the other news about the GT 1030 getting DDR4 shows how scummy / low effort some of the products can be. I'm assuming they're trying to use or get rid of some materials in stock. Also, like someone else said, they should just be 1 slot, to even bother to make it worthwhile.

Maybe high memory prices have something to do with this?

 

[jabbadap]

Having the space for a 6 pin power socket on the PCB implies there is a more powerful board that requires it, or the PCB is carried over from another model IMO.
If that is the case then manufacturing costs should be lower, and therefore the market price.

That is the same board as is in use of non-FTW gtx1050s and gtx1050tis(Might actually be Nvidia's reference board layout). But not even two fan SSC versions of them have that 6-pin connector soldered down. FTW:s uses different pcb and they have 6-pin connector soldered down(8-pin pads, but six pin soldered down).

 

[bug]

That is the same board as is in use of non-FTW gtx1050s and gtx1050tis(Might actually be Nvidia's reference board layout). But not even two fan SSC versions of them have that 6-pin connector soldered down. FTW:s uses different pcb and they have 6-pin connector soldered down(8-pin pads, but six pin soldered down).

If I'm not mistaken, EVGA's SSC line uses reference PCBs and FTW uses custom ones.

 

[Manu_PT]

Memory bandwidth has to be correlated with GPU processing power. You can slap a wide bus onto a weak GPU, but that's just wasteful.
The overall product however, still feels odd. I mean, EVGA is something you expect enthusiasts to buy, whereas 1050 is meant for those that don't care (or need) much. But hey, if EVGA feels there's a market for these, who am I to argue?

According to steamStats, this kind of gpu is what most ppl buy/have at home

 

[newtekie1]

You can slap a wide bus onto a weak GPU, but that's just wasteful.

Exactly, the 128-bit bus was definitely a waste on the normal GTX1050.

 

[bug]

Exactly, the 128-bit bus was definitely a waste on the normal GTX1050.

To tell the truth, I haven't read 1050 reviews, so Idk if it's memory starved or not. I was just saying, a narrower bus doesn't automatically mean a neutered card.

 

[Eric3988]

According to steamStats, this kind of gpu is what most ppl buy/have at home

https://store.steampowered.com/hwsurvey/Steam-Hardware-Software-Survey-Welcome-to-Steam

According to the survey, the 1060 is the most popular card in May. The 1050/TI make up slightly more percentage if you count them together. The 960 still holds a respectable 5% though.

 

[bug]

https://store.steampowered.com/hwsurvey/Steam-Hardware-Software-Survey-Welcome-to-Steam

According to the survey, the 1060 is the most popular card in May. The 1050/TI make up slightly more percentage if you count them together. The 960 still holds a respectable 5% though.

It's beside the point anyway. I was just saying, you're either in the market for an EVGA or a GTX 1050. Few people will need a quality card when buying 1050.

 

[newtekie1]

To tell the truth, I haven't read 1050 reviews, so Idk if it's memory starved or not. I was just saying, a narrower bus doesn't automatically mean a neutered card.

I have a GTX1050, I've never seen the memory bus load over 65% when playing games. The 128-bit bus just wasn't necessary. However, I don't know how the GTX1050Ti behaves, if the extra shaders means the 128-bit bus is more utilized.

I would be this new GTX1050 with 96-bit bus but with all the shaders unlocked, probably is limited by the memory bus. However, that is what nVidia wanted, so it wouldn't compete with the GTX1050Ti. I haven't seen any performance numbers on these new GTX1050s but I'm guessing their performance is actually going to be closer to a GTX1050Ti instead of a GTX1050 2GB.

 

[Ruru]

96bit bus. Feels like we're back to GeForce 4 era...

GF4 Ti had 128-bit bus, GF4 MX had 32/64/128-bit bus, but MX cards were entry-level anyway. IIRC this is the first card with 96-bit bus.

 

[Vya Domus]

I have a GTX1050, I've never seen the memory bus load over 65% when playing games.

That's the memory controller not the memory bus and it can obviously be designed (as is the case very often) to handle a lot more memory I/O operations than the bus can carry out.

 

[newtekie1]

That's the memory controller not the memory bus and it can obviously be designed (as is the case very often) to handle a lot more memory I/O operations than the bus can carry out.

The memory controllers are the memory bus. Each memory controller equals 32-bits of memory bus. Their load is very much directly related to memory bus usage.

 

[Vya Domus]

The memory controllers are the memory bus.

No they are not , the bus is one component , namely the wiring that connect the memory and the controller is another. Bus load is somewhat of an improper metric , it's either on or off for a specific amount of clock cycles while the controller reads/writes on the data lines but the act of reading and writing can occur much slower/faster than what the bus can allow.

You can totally change the bandwidth of a data transmissions by using a controller that say only reads the data lines each 2 cycles, therefor you can have a bus that's used "100%" but where the resulting bandwidth is half of what it can be because of the controller. And this can happen the other way around as well. Your thinking is flawed because you assume that the controllers are always as fast or slower than what they would need to be for a specific bus/clock combination.

An obvious sign that controller load is independent from the default bus/frequency configuration the fact that you can overclock the memory and it will all still function correctly. Sure the controller may only be wired for 32 data lines but that doesn't mean it can't be faster than that and therefor report lower utilization. Bus load =/= memory controller load otherwise why would there even be a distinction between the two in the first place , come on man.

 

[newtekie1]

No they are not , the bus is one component , namely the wiring that connect the memory and the controller is another. Bus load is somewhat of an improper metric , it's either on or off for a specific amount of clock cycles while the controller reads/writes on the data lines but the act of reading and writing can occur much slower/faster than what the bus can allow.

You can totally change the bandwidth of a data transmissions by using a controller that say only reads the data lines each 2 cycles, therefor you can have a bus that's used "100%" but where the resulting bandwidth is half of what it can be because of the controller. And this can happen the other way around as well. Your thinking is flawed because you assume that the controllers are always as fast or slower than what they would need to be for a specific bus/clock combination.

An obvious sign that controller load is independent from the default bus/frequency confugurationis the fact that you can overclock the memory and it will all still function correctly. Sure the controller may only be wired for 32 data lines but that doesn't mean it can't be faster than that and therefor report lower utilization. Bus load =/= memory controller load.

No, that isn't now it works. The bus is not one component. The 128-bit memory bus on the GTX1050Ti, for example, is actually 4 32-bit connections. Each 32-bit connection is wired to a 32-bit memory controller. So the GTX1050Ti has 4 memory controllers. The load on those memory controllers is a direct measurement of how much data is flowing over the memory bus. Each memory controller is designed to handled 32-bits of memory traffic, and not more. So the 4 memory controllers on the GTX1050Ti are designed to handle 128-bits of memory traffic and no more. So, if they are at 60% load, that means the memory traffic on the memory bus is only using 60% of the available bandwidth on the bus.

As for your overclocking, AFAIK, the memory controller and memory bus all get clocked to match the memory. That is why different GPUs that are paired with the exact same memory often reach different memory clock speeds. One of the memory controllers is becoming unstable before the actual memory is. And some memory controllers are built better than others and can handle the higher clock speeds.

 

[Readlight]

Why always card who has reasonable price always is the slowest ones? they will be used someday on 4k and nothing will work. maybe some 2D games.