GPU Options

I'm still new to Resolve and not yet very familiar with performance across hardware configs. What are you guys thinking might offer the better performance?

1. A single GTX 1080, 2560 CUDA cores, 256 bit bus, 8GB memory

2. Two 1060's, 1280 CUDA cores, 192 bit bus, 6GB memory

Pricing for both options is in the same neighborhood. I'm presuming with option 2, one card would drive the GUI, the second would be used for color, effects, etc.

Thoughts?

In Resolve (unlike GPU rendering engines like Octane or Redshift), you're always better off with a single powerful GPU than two less powerful ones.

If you can squeeze it, I would go with a 1080ti if I was you, I know it's at least $300 more expensive than the 1080, but you'll be glad you go one.

Alternatively, you could look to see if you can grab a used Titan X with 12gb of memory, those GPU's are still king of the hill.

Kays, does more GPU memory always help performance? I ask because with the computer itself, RAM above what's needed doesn't help at all. So will 12GB of GPU RAM always have better performance than 6 or 4 GB, or is it like with computer systems?

Thanks much for the advice!

Hi.

I am even more new to resolve than Jim. What I read is that one GPU for GUI Stopped in ver 12 of Resolve. In ver 14 is the Recommendation to use the same GPU for both processing and GUI.

But there is a lot of rumors on the internet about new and faster next gen nVidea Geforce Graphics cards. I will suggest that you try to google GTX 2080.

Regards Carsten.

In regards to pricing, it's a bad time to shop for a GPU.

In short, the cryptocurrency miners and hardware makers go for the same chips as the suppliers for gamers (and Resolve users). This is reflected in all such GPUs being sold, on average, 150% of their MSRP, with the high-end cards, like 1080ti, sometimes nearly double their price.

- Further reading

But there is a lot of rumors on the internet about new and faster next gen nVidea Geforce Graphics cards. I will suggest that you try to google GTX 2080.

There's a general idea that Nvidia may dedicate a higher-powered chip (Ampere, GV 104) to a certain GPU (2080), looking for the cryptocurrency crowd to latch onto it.

Meanwhile, the same generation would include the Turing chip (GV 102) card, the 2080ti.


Nvidia is expected to unveil their new consumer GPUs in the coming months, which is set to make this interesting.

I'd say hold is fair play at the moment

In regards to pricing, it's a bad time to shop for a GPU.

Yeah, the 1070 is selling for a good chunk more now than what I paid a year ago.

In Resolve, you're always better off with a single powerful GPU than two less powerful ones.

Thanks.

In regards to pricing, it's a bad time to shop for a GPU.

Yeah, the 1070 is selling for a good chunk more now than what I paid a year ago.

If you have a 1070, run Resolve with that.

Update to the 2000s with R15, if you feel like it.

In Resolve (unlike GPU rendering engines like Octane or Redshift), you're always better off with a single powerful GPU than two less powerful ones.

Kays, your statement totally contradicts my understanding. I’m curious as to why you claim this, and I’m very invested in this because I’m picking parts for some Resolve workstations now.

Take, for example, these benchmarking results of the TITAN V from Puget Systems.

https://www.pugetsystems.com/labs/artic ... 12GB-1108/

They conclude, “it is very likely that two or three GTX 1080 Ti cards will out-perform a single Titan V at a lower cost.”


Sent from my iPhone using Tapatalk

[quote="Take, for example, these benchmarking results of the TITAN V from Puget Systems.

https://www.pugetsystems.com/labs/artic ... 12GB-1108/

They conclude, “it is very likely that two or three GTX 1080 Ti cards will out-perform a single Titan V at a lower cost.”


Sent from my iPhone using Tapatalk[/quote]

2 GTX 1080 Ti cards is 16GB of RAM vs the 12GB of the Titan. One needs to compare total RAM and GPU performance, not just how many cards.

Sorry, wrong. Resolve is using VRAM according to the lowest amount in any card in use.
Two 8 GB cards don't improve the VRAM to 16 GB.

2 GTX 1080 Ti cards is 16GB of RAM vs the 12GB of the Titan. One needs to compare total RAM and GPU performance, not just how many cards.

Sorry but no : 2 x 8GB of RAM result in 8GB used by Resolve.
What is hard to tell is the real gain in global perf and smoothness (GUI, playback, computing, ...) of 2 cards compared to a single card, which really exists but hard to specifically define when at work.
But for instance one 6 GB + one 8 GB card result in 6 GB used in Resolve...

I am using a 1070 since last year when it came out and i am not planing on changing it any time soon. The card is great. After all, how much more time will you win in rendering? 3min, 5min? If you have a slower cpu, upgrade that. Maybe a 7820x, 8700, whatever has more than 4 cores. Don't know that you budget is. maybe higher since you are thinking over a 1080ti.

Hi,

Simplifying to the maximum:

You have to see the calculations with GPU of some different. The goal is to parallelize calculations.
So 2 GTX 1080 Ti cards is (2x8) = 16GB VRAM "addressable space" if we make an optimized balancing of the threads between the 2 GPU (s).

In a perfect world it's better than a gtx1080ti with 11 GB VRAM. In reality not so simple.

In summary for GPU Computing: In the processing of an image: more kernel, more blocks so more threads that perform the calculations in parallel.


First of all, you have to see the GPU (s) as a multi-dimensional grid made up of independent calculation blocks.

A block represents a matrix of multi-dimensional threads. The developer chooses the dimensions he needs.

In the GPU (s):
A block contains 'n' threads. Each thread executes an instance of a kernel and has coordinates in that block to identify the thread.

On the other hand, at a given moment, before relaunching new calculations (for example for a new 'IN image'), it is necessary to recover the results of all threads and to consolidate them in a 'OUT image'. This re synchronization is then impacted by the slowest GPU or exchange of information from (s) GPU (s) to the software that is in RAM via the information bus. It is an intra bottleneck GPU (s) / BUS / CPU (s) / RAM /etc...

For the balancing between GPU (s): More GPU (s), more parallel calculations but provided that the GPU (s) are homogeneous. Ideally the same capabilities of calculations and addressability and speed of exchange in information bus.

That is why it is recommended to have the same GPUs / speed Bus in a multi-GPU configuration and ideally in PCI-E slots with the same bit rate (X16).

The smallest GPU in computing capacity is then used as the common denominator (and for the GPU GUI is reduced by the need for GUI)

It's not the developer who decides everything: it is the hardware via interrupts and manages the rendering sync to the software that retrieves the image in memory in what is nothing other than an array of floats.

@JIM

1. A single GTX 1080, 2560 CUDA cores, 256 bit bus, 8GB memory
2. Two 1060's, 1280 CUDA cores, 192 bit bus, 6GB memory

Without hesitation the option a GTX 1080:

- as many cores as 2 gtx1060
- important and we do not talk about it often but the bus rate will be bigger the more it will be possible for the hardware to transfer the bits in the bus (exchanges between RAM / CPU / GPU / BRIDGE / etc ...)

One GPU:
less power: less PSU
less heat

Hope this help.

Also, Resolve is very scalable for one's needs. While some users do need a top-of-the-line GPU, others can just as well go with an older card. By dedicating additional system memory to Resolve, using SSDs for the OS and scratch disk, along with optimized media and reasonable working resolution, one can level the stress put on to the workstation. And that new playback engine in R14 really is efficient.

We have a workstation with GTX 760, and it still runs like a charm with HD as a working format, dealing with complex FXs and overlays without a hiccup. It's quite impressive, actually, as the card seems faster than it did with R12.5. Point is, you can get a ridiculously priced 1080ti, if you know you need it. But Resolve is rather good in assessing the GPU potential during initial startup. If all looks good with the resolutions you need, the update can be pushed back.

Given that CUDA support determines the lifespan of a Resolve GTX, we'll be looking at the update soon enough to keep up with the software. But there's no particular reason to dismiss older cards outright.

Kays, does more GPU memory always help performance? I ask because with the computer itself, RAM above what's needed doesn't help at all. So will 12GB of GPU RAM always have better performance than 6 or 4 GB, or is it like with computer systems?

Thanks much for the advice!

The answer is....it depends on what you're doing. For instance, if you're mainly working in HD or 2K, the extra RAM won't have as much of an impact.

I can tell you from my personal experience that going from a 980ti (6gb of RAM) to a 1080 (8gb of RAM) made no discernible difference when working with 4k footage. At some point I was running both cards together and there was a slight improvement in overall performance when using particularly heavy nodes and effects (like NR, or OFX), but it wasn't twice as fast or anything. I ultimately decided to get rid of the 980ti as the 1080 had similar performance and more memory. My plan was to add a second 1080 in order to keep the system balanced...and that's when the prices for GPU's started skyrocketing.

My new plan is to replace my 1080 with a 1080ti or a Titan X as soon as I find a reasonable deal (i.e. below $1000), but not because of the additional 3-4Gb of memory, but rather for the increased performance from CUDA cores.

It really stinks that we're currently undergoing what is a price bubble for GPU's. Right now it's a pretty awful time to be in the market for a GPU, and unfortunately thinks look even bleaker for the near future (and don't think for a second that the 20xx generation won't also suffer from the same overpriced hype).

I can tell you from my personal experience that going from a 980ti (6gb of RAM) to a 1080 (8gb of RAM) made no discernible difference when working with 4k footage.

That's interesting, thanks for sharing.

Did you test them with 12.5 or 14?

It really stinks that we're currently undergoing what is a price bubble for GPU's. Right now it's a pretty awful time to be in the market for a GPU, and unfortunately thinks look even bleaker for the near future (and don't think for a second that the 20xx generation won't also suffer from the same overpriced hype).

Blame cryptocurrency mining for off-setting the market. This is actually a growing pain in the hardware industry, as gamers are waking up and pushing back their system upgrades (mobo's, CPU's, etc.) because the GPU costs a stupid amount of money. Nvidia and AMD are thus getting external pressure to solve the issue and to segregate the miners from gamers.

In any case, wait and see is currently the best bet.

If you have a 1070, run Resolve with that.

That's for my gaming rig. It't not going anywhere. (At least, not without buying something equal or better to replace it with.)

If you have a slower cpu, upgrade that.

That will be my first step. My current GPU is a GTX 960. I figure it could also use an upgrade.

So 2 GTX 1080 Ti cards is...

...waaaaaaay over budget. A single 1080 might be doable...later this year.

and don't think for a second that the 20xx generation won't also suffer from the same overpriced hype.

Hi.

The 20xx vil use next gen vRam called GDDR6. And Samsung started full production of this new faster vRam back in January. So lets hope there will be plenty of supply.

Regards Carsten.