RTX 3060 12 GB vs. 3060 ti

[DeeJay]

Hey flight simmers,

 

I intend to buy a new graphic card soon. At the moment, I have a Palit RTX 2060 Super (8GB), and I'm running a i7 9700 @ 4,4 GHz and 16 GB DDR4 RAM. The cpu runs nice and quiet. My problem with the graphic card is not even so much performance, but temperatures and noise. I can't stand the whining of the 2 fans running at 2500 rpm at 85°C, even more in MSFS (which I hardly fly yet), less so in P3d v4, but bad enough there. And if I'm gonna upgrade anyways, I would look forward for some better performance, too, as I'm running an ultrawide monitor at a 3440 x 1440 resolution now, and that forced me to make some compromises with the settings in P3d.

 

Now, I've found the RTX 3060 with 12 GB VRAM, for example here: https://www.alternate.de/EVGA/GeForce-RTX-3060-XC-GAMING-Grafikkarte/html/product/1727976

But in various forums I've read recommendations for the RTX 3060 ti as the better choice, but this one only has 8 GB VRAM.

 

Now, my question is: How important is VRAM and what makes the ti "better" despite having less VRAM?

 

I could see a biiiig improvement in loading times at my last upgrade (GTX 970 4GB => RTX 2060 super 8GB), and I concluded that this would be the benefit of having a lot more VRAM. Furthermore, I've read how important VRAM would be for P3d v5 - which I don't have yet, but who knows... I guess MSFS wouldn't complain either about 12 GB VRAM, right?

 

I know, a 3070 ti would give me both, but apart from the much higher price, I fear this would be overkill compared to my CPU anyway (??).

 

Thanks in advance.

 

Dominik

Edited May 26, 2022 by DeeJay

[renault]

Hi Dominik

 

Basically I had virtually the same system as you until I upgraded my GPU from a RTX2060 super to a RTX3070ti.

 

It sounds to me that your GPU is really pushed to its capabilities  running at 3440 x 1440 . On my own system at 2560 x 1440 p 

with fps locked at 30 fps (this is important for this card)  I was running at about 80 - 90 % utilization with temps at 70-73C.

This was with all settings in MS2020 at ultra.  Vram usage was around 7-7.3 total system allocated memory.

 

I can't comment on P3d as I never have owned it.

 

With the 3070ti, Vram usage is the same, GPU utilization is about 40% and temps are gpu 55-58C, Memory 64-68C and hot spot junction temperature

is about 70 - 72C.  The card is very quiet.

 

I'm sure that many others will have opinions on this matter and hopefully they will also comment.

 

If I were in your shoes I would not buy a new gpu right now but do the following

 

1) If you haven't done this , lock your fps at 30 using either Nvidia Control panel or the vsync setting in MS2020

 

2) Go into MS2020 and set scaling to 75%.  This will run your system at 2580 x 1080p and the built in upscaler in MS2020 will

upscale to your monitor resolution,  If the image quality is not quite to your liking, then increase the scaling by 5% and try again .

You should be able to get a visual fidelity and balanced system load which is acceptable.

 

Your 2060 will work much better under these conditions until SU10 ....

 

3) With SU10 the Nvidia DLSS (Deep learning Super Sampling) system will be incorporated into MS2020 .

 

     There are numerous articles by Nvidia and others about DLSS and the quality it gives, while significantly reducing the system load on both

     the cpu and gpu. 

 

     Once you have had a chance to try this and to see how much better your system runs MS2020 with it, I think then you can make a much better informed decision

    than you can at this point in time.   It has the potential for such a significant improvement for most users  that I personally think it is impossible to decide upon a

    GPU  purchase , until you have a chance to use DLSS  for a while with your current system. System behaviour will be very different with DLSS  compared with

   what you are achieving with SU9 at the moment.

 

The following is my understanding of the background to your question

             "what makes the ti "better" despite having less VRAM? 

 

Cheers 

Pete

 

In order to understand the answer to why the 3060 ti is a better card overall we need to delve into the 
workings of a gpu a little bit.

 

The following discussion is specific to Nvidia GPU's since Nvidia uses different memory bus specification (memmory bus width) on different cards.

 It is my understanding that AMD uses a single memory bus width of 384 bits for all their cards.

 

The memory bus width on a GPU more or less defines the number of Vram memory chips that can be used on a graphics card.

 

The memory bus width refers to the number of bits of data per clock cycle that can be sent or received from the GPU.
So for example a card with a 100 bit memory bus width means that it can transfer 100 bits of data in a single clock cycle.

 

Memory bandwidth is the rate that data can be stored or retrieved  from memory by a processor.

Memory bandwidth is a combination of both the memory bus size and the actual speed that the memory runs at.
Memory bandwidth refers to the ability to move data such as textures in and out of the Vram as they are processed by 
the GPU.

 

The 3060 has a 192 bit memory bus which is slower than the 256 bit memory bus that the 3060ti has. In other words, the 3060 in a single clock

cycle can transfer 192 bits of data into or out of Vram, whereas the 3060ti in 1 clock cycle would be able to transfer 256 bits of data.

 

A Vram memory chip has a bandwidth of 32 bits. DDR6 memory chips usually come in multiples of 1 or 2 GB.

 

So a 3060 with a 192 bit memory bus would use 6 lanes of 32 bit bandwidth which is 6x32 = 192 bits.

In other words, due to the bus memory bandwidth, the 3060 has room for 6 memory chips.  

 

Nvidia basically designs for  a minimum memory of 8 GB for its RTX series cards.
So this means that  a 3060 could have six, 1 GB DDR6 memory chips which would be 6GB in total.

 

However, since this is less than the 8 GB design spec, they used six 2GB DDR6 memory chips to adhere to
the memory design spec, which means the card ended up with a total of 12 GB.  The 12 GB vram fills the 6 slots.

 

The 3060ti has a memory bus width of 256 bits. This means that it would have 8 lanes (8 x 32 = 256 bits)

and therefore would have room for 8 memory chips.  Using 1 GB memory chips then gives you a total of 8GB total
Vram which meets their RTX design specification.

 

The 3060 has a higher memory speed (1875MHz)  than the 3060 ti memory speed of 1750Mhz. This is based on non overclocked design specifications.

To calculate the memory bandwidth we multiply the memory speed x memory bus rate.

 

For the 3060 this would give a memory bandwidth of 1875Mhz x 192 bits = 360 GB/second
For the 3060ti this would give a memory bandwith of 1750Mhz x 256 bits = 448 GB/second

 

This means that the 3060ti can process textures etc faster in its memory than the basic 3060, even though the 3060 comes with more memory

and the memory speed is actually slightly faster. In other words the 3060ti is still the faster card because with its higher memory bus speed it is able to move more information (448 GB/sec)  in/out of memory than the 3060 (360 GB/sec).

 

Just for comparison ...

 

RTX 3070 Memory bus width = 256 bits (Supports 8 chips @32 bit/chip)
                Vram = 8 GB @ 1GB DDR6 per chip 

 

RTX 3070ti Memory bus width = 256 bits (Supports 8 chips @32 bit/chip)
                Vram = 8 GB @ 1GB DDR6X per chip 

 

RTX 3080  Memory bus width = 320 bits (Supports 10 chips @ 32 bit/chip)
                Vram = 10 GB @ 1GB per chip

 

RTX 3080ti  Memory bus width = 384 bits (Supports 12 chips @ 32 bit/chip)
                Vram = 12 GB @ 1GB   chip 
 

RTX 3090 Memory bus width = 384 bits (Supports 12 chips @ 32 bit/chip)
                Vram = 24 GB @ 2GB  per chip

 

[wain71]

I have the 3060ti paired with a 9600k, mildly OC'ed and I have to say on both MSFS & XP I am happy with the smoothness and great visuals I am seeing, I was going for the 3060 with more vram but the ti won out in the end, like you say it's still an expensive market out there and we all have a budget of sorts...

great bit of explaining there Pete, always helpful to get a bit of help in plain English as this sometimes gets lost in all the techy jargon

[DeeJay]

@Pete: Thanks so much. This must be one of the best and most conclusive posts I've read in the past years!!

 

I might give that SU10 thing a try, but I see myself ending up with a well cooled 3060 ti.

 

I'm perfectly happy with the visuals of MSFS running with 1080p, and honestly I hardly see a difference there to 1440p, just the performance difference is very noticeable.

However, running P3d in 1080p res with the 1440p monitor gives a very "milky", washed and ugly image, strange enough ONLY in full screen mode. That's why I'm running it in 1440p, which still looks perfect and performs OK, I can't say really very good.

But, yeah, whatever I did so far, memory usage is always at max, and even if I clock down the GPU, the fans howl awfully.

Edited May 27, 2022 by DeeJay

[W2DR]

6 hours ago, DeeJay said:

@Pete: Thanks so much. This must be one of the best and most conclusive posts I've read in the past years!!

I sure agree with that. It needs to be made a sticky..............

[PiotrMKG]

On 5/27/2022 at 6:33 AM, renault said:

RTX 3090 Memory bus width = 384 bits (Supports 12 chips @ 32 bit/chip)
                Vram = 24 GB @ 2GB  per chip

3090 has 24 chips (12 per PCB side) with effective clock 19,5Gbps which translates to bandwidth of 936.2 GB/s

3090ti has 12 chips@2GB with effective clock 21Gbps which translates to bandwidth of 1008GB/s

[Doug Sawatzky]

I find vram is more important than FPS these days, which is why I am still enjoying my 1080ti with 11gb of vram.