Inferior Editing Experience on RTX 2070 Super vs RTX 2060 KO

NateDZD

I sold my EVGA RTX 2060 KO Ultra to a friend who was building his first PC, and I replaced it with a Gigabyte RTX 2070 Super (Windforce OC 3X). While the gaming performance of the card is markedly better than the 2060, I did notice that when editing video, I have a noticeable lack of stability when scrubbing through the timeline or previewing transitions (freezing, stuttering, and occasional crashes). Hardware accelerated rendering is still very fast, but the overall editing experience is pretty bad compared to the 2060.

Does anyone know what might be causing this? I didn't have any of these issues with the 2060 KO, but the 2070 Super has been a bit difficult to edit with. Since it's primarily a gaming system (I game around 4-5 days a week for about 2 hours per day, and I edit one video a week for about 3 hours per video), I stuck with the gaming drivers from Nvidia. Would the Nvidia Studio Driver be a better option, or could that adversely effect gaming performance? I do have a tendency to play some graphics-heavy titles like Shadow of the Tomb Raider and Resident Evil 3, so I'm not sure if the Studio Driver is the right fit for me, but I guess I'd like to know if it's a driver issue or something else.

PowerSpec_MikeW

The EVGA RTX 2060 KO Ultra is unique in a way. There are videos online that have dissembled the card and basically instead of being a typical 2060 GPU core,  it's actually a 2080 core. Most likely this is a case where the GPU failed testing and could not be sold as 2080 core and was offered to EVGA at to be sold as lower end GPU at a reduced clock frequency. VRAM will be reduced as well. So it will perform worse in games, but potentially having more CUDA, Tensor and RT Cores will increase the workstation performance.

NateDZD

TSMikeW said:

The EVGA RTX 2060 KO Ultra is unique in a way. There are videos online that have dissembled the card and basically instead of being a typical 2060 GPU core,  it's actually a 2080 core. Most likely this is a case where the GPU failed testing and could not be sold as 2080 core and was offered to EVGA at to be sold as lower end GPU at a reduced clock frequency. VRAM will be reduced as well. So it will perform worse in games, but potentially having more CUDA, Tensor and RT Cores will increase the workstation performance.

I just uttered an audible "whoa" akin to Keanu Reeves circa 1999.

I seem to recall the salesperson mentioning this to me when I initially picked up the 2060 KO Ultra. I didn't put much thought into it as I kinda took it as an upsell tactic (an obviously successful one since I went ahead and purchased it). I was originally looking at the RX 5600 XT, which I think was around $290, and the EVGA card was only $20 more, so it was an easy upsell. I guess it was more than just an upsell tactic then, eh?

Do you think there's any way to offset this limitation in the 2070 Super? I'm willing to do a little bit of upgrading now, but I'm looking to do a slightly bigger upgrade when Ryzen 4th gen comes out later this year, so if any suggested solutions involve hardware purchases, I'd like to keep it below $150.

It's also not world-ending since I'm not editing video more than once a week or so (and the raw footage I receive is never any bigger than 1080p), so I'm perfectly fine with just living with it. I just have to remember to find a nice Ctrl-S rhythm when I'm editing. I'm honestly very happy with the 2070 Super for gaming, and ultimately that's what my build is for. But now the pandemic has forced a number of other use cases onto this rig. (And now I'm wondering if I can expense some of my PC upgrades...  🤔😉)

TSTonyV

The 2070 Super really shouldn't be worse in video editing than the 2060 KO. While the 2060 KO does have the same type of die as a 2080, the 2070 Super also has the same type of die as a 2080 but the 2070 Super has better specs otherwise than the 2060. Realistically, the 2070 Super is incredibly similar to the 2080. 

The studio and game ready drivers that NVIDIA releases are typically extremely similar, sometimes identical. The difference is that game ready drivers receive more frequent updates to keep up with optimization in new games, whereas the studio drivers see less frequent updates because professional applications don't see updates that often. Your game performance shouldn't be drastically affected by the studio drivers. I would try the most recent studio driver just to see if there's any noticeable difference or improvement, and if not, try rolling back your game drivers. 

Unfortunately I don't know video editing software so I can't really suggest any settings changes or tweaking within those, but I think drivers are a good start. Double check your performance mode under 3D settings in the Nvidia control panel and make sure the card is set to max performance or the adaptive/balanced option just in case. It most likely already is, but it's worth checking. 

PowerSpec_MikeW

To add more to this. It would appear the 2060 KO and 2080 both shared TU104 chip variants. While the original 2070 used a TU106, which is a smaller die size.  The 2070 Super uses another TU104 chip variant that differs from both the 2060 KO and the original 2080. The 2080 Super uses yet another variant of the TU104 chip. 

While the 2060 KO should have the extra cores available that i mentioned, they should be disabled and even if they weren't, this is a product that likely failed validation, so they may be disabled for a good reason. On paper the 2070 Super should have better specs, more accessible core, and higher clockspeed. There were benchmarks that saw the RTX 2060 KO outperforming the original RTX 2070, despite the specs on paper being superior, but only in workstation performance. 

Now, looking at the TU106 and TU104 to compare the original RTX 2070 to the 2060. The only thing I could find architecturally online that accounts for the improvements in workstation performance is that the TU104 has a greater raster engine ratio. It's certainly possible something has changed with the die that we're not privy to between the TU104-400, 400A, 410, and 150.  I doubt we ever will be.

It would be really interesting to benchmark these cards on the same system for a real world comparison.

As for troubleshooting performance tweaking. Verify you did a clean install of the drivers with DDU when you installed the new card. If not, do so. What video editing software are you using?

NateDZD

TSMikeW said:

As for troubleshooting performance tweaking. Verify you did a clean install of the drivers with DDU when you installed the new card. If not, do so. What video editing software are you using?

Thanks for the tip! I just cleaned things up with DDU, and I'm updating the drivers now. I did notice when I ran Driver Easy before running DDU, the 2060 driver was still showing up in the menu as a "Missing Hardware."

I mostly use Filmora 9, but I've been an Adobe Premiere user for a number of years, and if I can get some extra money, I may end up using it again.

PowerSpec_MikeW

@NateDZD I would show hidden and remove any disconnected versions of the card listed there. I apologize for the delayed response, missed your reply.

NateDZD

@TSMikeW Thanks! I actually had already done that, and it did provide a noticeable increase in stability. It's still a tiny bit wonky, but overall I feel much better editing than I did before running DDU and clearing out any remnants of the 2060.