Monday, May 21st 2018

NVIDIA GeForce GTX 1050 3GB Memory Bus-width Confirmed: A Major Trade-off

NVIDIA's entry-level GeForce GTX 1050 launched in a new 3 GB variant, earlier this month, with 50 percent more memory than the 2 GB which the original GTX 1050 launched with. But there's a major catch that's relegated to the fine-print of the card's specifications on NVIDIA website, and something most NVIDIA AIC (add-in-card) partners won't particularly blare on their product packaging anywhere near as loudly as the memory amount, and that's memory bus width. The 3 GB GTX 1050 has 50 percent more memory than the original GTX 1050, but a 25% narrower memory bus, at just 96-bit.

When you look at a GTX 1050 3 GB graphics card PCB, you'll likely only find three 8 Gb (1 GB) memory chips, with one set of memory chip traces blanked out. It's not even like NVIDIA compensated for the narrower memory bus with higher memory clocks. The chips run at the same 7 Gbps as the original's, yielding just 84 GB/s memory bandwidth, compared to the original's 112 GB/s. The CUDA core count of the GTX 1050 3 GB is the same as the GTX 1050 Ti, with 768 CUDA cores, which twitch their thumbs as data is moved between the GPU and memory over Pony Express. Besides more CUDA cores, the GPU clocks are marginally higher, with 1392 MHz base and 1518 MHz GPU Boost, compared to 1354/1455 MHz of the original. NVIDIA, which recently sermonized the industry on "making products easier for consumers to identify" with its stillborn GPP, is once again caught concealing a major specification. To find it, you'll need to visit the product page of the GTX 1050, scroll all the way down to the specs sheet, and click on "view full specs" to reveal the memory bus width.
Add your own comment

43 Comments on NVIDIA GeForce GTX 1050 3GB Memory Bus-width Confirmed: A Major Trade-off

#2
dj-electric
Price and performance will decide if there's a market for the GTX 1050 3GB, and nothing else. It can have a 64bit memory controller for all i care.
People should stop obsessing over paper specs, this isn't 2008 anymore.
Posted on Reply
#3
cucker tarlson
Lol @Nvidia, they've skimped on faster ddr5 that could've at least plastered most of the gap.
Posted on Reply
#4
qubit
Overclocked quantum bit
One of the reasons I upgraded from the 780 Ti was because 3GB RAM just wasn't enough with some games, even at 1080p. It certainly had enough GPU power left over. The other reason was DX12 capability.

It should have been 4GB for this new 1050 or discontinue the card, in my opinion.
Posted on Reply
#5
jabbadap
"cucker tarlson said:
Lol @Nvidia, they've skimped on faster ddr5 that could've at least plastered most of the gap.
That would be essentially competing with gtx1050ti, which they probably don't want to do. It's obviously meant to be cheaper than ti, with faster and more expensive memories there won't be price gap to put it.
Posted on Reply
#6
TheinsanegamerN
"qubit said:
One of the reasons I upgraded from the 780 Ti was because 3GB RAM just wasn't enough with some games, even at 1080p. It certainly had enough GPU power left over. The other reason was DX12 capability.

It should have been 4GB for this new 1050 or discontinue the card, in my opinion.
The 1050 is not a 780 ti. It is a low end card, and those who buy them are not super concerned with pushing graphical settings that need 4+ GB. At 1080p, more then 4GB used is exceedingly rare, and requires more GPU power then even the 1060 can provide.
Posted on Reply
#7
qubit
Overclocked quantum bit
"TheinsanegamerN said:
The 1050 is not a 780 ti. It is a low end card, and those who buy them are not super concerned with pushing graphical settings that need 4+ GB.
It's a much newer generation than the 780 Ti, so are you sure it has lower performance? Even if it does, 3GB still seems to be a bit tight for today's games to me.
Posted on Reply
#8
Vya Domus
"dj-electric said:

People should stop obsessing over paper specs, this isn't 2008 anymore.
Interestingly enough , people cannot read said paper specs and end up being "scammed".

"dj-electric said:

It can have a 64bit memory controller for all i care.
It's also interesting how most of the time when Nvidia does it most people try to undermine their actions.
Posted on Reply
#9
efikkan
Trading some memory bandwidth for more computational performance is fine, there is no way GTX 1050 at 1733 GFlop/s needs 112 GB/s, that's more bandwidth per GFlop/s than a GTX 1080 Ti.

But I don't get why Nvidia is refreshing their low-end lineup. There is simply no need for this amount of options among low-end hardware, with GT 1030, GTX 1050 and GTX 1050 Ti all having 2-3 variations of memory and/or core configuration, then multiply this with all the AIBs making each of these with different coolers and "OC versions", like if a 3% overclock on a GTX 1050 is going to matter. Two models in total is more than enough to cover the entire low end. And don't get me started on the GTX 1060 variants, what are we up to now, 5?

This is not a Nvidia thing, AMD is just as guilty in this. And vendors, if you want to make it easier for the consumers, make one version of each product.
Posted on Reply
#10
ssdpro
NVIDIA concealed that memory bus width so bad - it is amazing how well they "concealed" it on the public specifications page. Only the most hardened investigative consumer would know to look at the product specifications buried deep inside something as complicated and little known as "the internet".
Posted on Reply
#11
the54thvoid
Non-story. Really, headline news. :kookoo:
Posted on Reply
#12
jabbadap
"efikkan said:
Trading some memory bandwidth for more computational performance is fine, there is no way GTX 1050 at 1733 GFlop/s needs 112 GB/s, that's more bandwidth per GFlop/s than a GTX 1080 Ti.

But I don't get why Nvidia is refreshing their low-end lineup. There is simply no need for this amount of options among low-end hardware, with GT 1030, GTX 1050 and GTX 1050 Ti all having 2-3 variations of memory and/or core configuration, then multiply this with all the AIBs making each of these with different coolers and "OC versions", like if a 3% overclock on a GTX 1050 is going to matter. Two models in total is more than enough to cover the entire low end. And don't get me started on the GTX 1060 variants, what are we up to now, 5?

This is not a Nvidia thing, AMD is just as guilty in this. And vendors, if you want to make it easier for the consumers, make one version of each product.
Working dies with broken rops/memory channel. Throw them a way or try to get some profit from them. Other than that I don't think next gen lower end graphics cards are any where close of release. What it takes? Some half a year after high end for getting low end to market. And nvidia "might" release something on summer.
Posted on Reply
#13
Archmage1809
I have looking for 1060 5GB vs 3GB and 6GB benchmark (only in Chinese, but the graph itself is enough).
And in some task, the 5GB even worse than 3GB.
There are no point releasing this kind of memory bandwidth cut down in exchange for CUDA core.
Since those card price point is very close. $100 and $130. What is this MSRP? $115?

P/s: yeah, I know it nowhere near the MSRP, that for reference only
Posted on Reply
#14
efikkan
"jabbadap said:
Working dies with broken rops/memory channel. Throw them a way or try to get some profit from them. Other than that I don't think next gen lower end graphics cards are any where close of release. What it takes? Some half a year after high end for getting low end to market. And nvidia "might" release something on summer.
With the exception of the early test run of Maxwell, low-end chips usually comes much later than mid-range and high-end.

When it comes to the next generation; Nvidia can release it anytime now…
Posted on Reply
#15
Chloe Price
Oh great. Cut down the already narrow membus, and few shaders more.

I'm sure that this is the card what we're going to see in OEM PCs.
Posted on Reply
#16
cucker tarlson
"qubit said:
It's a much newer generation than the 780 Ti, so are you sure it has lower performance? Even if it does, 3GB still seems to be a bit tight for today's games to me.
780Ti will destroy 1050. Even in 2018 games.
Posted on Reply
#17
Imsochobo
"dj-electric said:
Price and performance will decide if there's a market for the GTX 1050 3GB, and nothing else. It can have a 64bit memory controller for all i care.
People should stop obsessing over paper specs, this isn't 2008 anymore.
1060 6gb vs 1060 3gb and you think it performs the same just more memory?
nope, 10% ish difference.
1030 ddr4 vs gddr and it's like going from barely playable to hey it can run windows just like my intel IGP can.

RX560 issue is not to be neglected, it does matter.

Lets say you bought a 1080TI just to find out it performs like a 1080 would it matter? YES it does!
Posted on Reply
#18
DeathtoGnomes
What you have here is Nvidia picking up chips off the ground and trying to make a bigger profit. Sorta like charging $1.98 for a 6 pound can of coffee you know will taste like shit.
Posted on Reply
#19
campb292
"ssdpro said:
NVIDIA concealed that memory bus width so bad - it is amazing how well they "concealed" it on the public specifications page. Only the most hardened investigative consumer would know to look at the product specifications buried deep inside something as complicated and little known as "the internet".
^------ OMG this. I don't know what is going on with the commentary at the end of this article.
Posted on Reply
#20
theGryphon
It's been there for some time now. Late news :rolleyes:
Posted on Reply
#21
xorbe
I guess this is a one-off card to service gaming cafe titles that need a bit more vram but not much horsepower.
Posted on Reply
#22
FordGT90Concept
"I go fast!1!11!1!"
Someone has to start the slow clap. Again.
Posted on Reply
#23
dj-electric
"Imsochobo said:
1060 6gb vs 1060 3gb and you think it performs the same just more memory?
nope, 10% ish difference.
1. I never said it 6GB and 3GB versions perform the same
2. You got baited on that example. The 3GB version of the GTX 1060 has 11% less CUDA cores (1152) compered to the 6GB version (1280)

You proven exactly what i was talking about. One shouldn't care about paper specs, but actual performance numbers.
Posted on Reply
#24
TheinsanegamerN
"qubit said:
It's a much newer generation than the 780 Ti, so are you sure it has lower performance? Even if it does, 3GB still seems to be a bit tight for today's games to me.
The 1050 is not some ubermech GPU. The 780ti is somewhere between 40-50% faster on average then a 1050 (sorry for bad site, but finding direct comparisons is a PITA), and the 780ti is no longer optimized in drivers, while the 1050 is. The 1050 is more comparable to a 680 in raw performance.

http://gpu.userbenchmark.com/Compare/Nvidia-GTX-780-Ti-vs-Nvidia-GTX-1050/2165vs3650

With that kind of power, you are not going to be pushing things like ultra details/MSAA/2K/4K. At 1080p high settings, 3GB if VRAM is more then enough for most modern games, with the exception of poorly coded ports. Back when I had my GTX 770 2GB in 2016, I was pushing ultra/max MSAA at 1080p in nearly every game and not maxing out the VRAM buffer, except for a few specific games like wolfenstein and Forza.

The buyer of 1050 type GPUs will most likely either be playing at even lower resolutions, playing older or less demanding titles like indie games or esports titles, or doesnt mind playing on medium/high instead of ultra. These use cases dont need tons of VRAM to function, 3GB will be more then sufficient for the GPU in question, especially at 96 bit, where memory bandwidth will hamper performance long before buffer size will.
Posted on Reply
#25
DeathtoGnomes
"FordGT90Concept said:
Someone has to start the slow clap. Again.
:clap:
Posted on Reply
Add your own comment