RTX 3090 worth it?

4090 has ECC.

He did write 3090ti too???

https://techgage.com/article/nvidia-geforce-rtx-4090-the-new-rendering-champion/

Seems like 4090 is 2x 3090 with almost the same power consumption.

Its more than a month to the RX 7900XT <— without ECC?

EposVox has done some testing on 4090 , Topaz Labs VEAI, GAI and PAI.

Here is the result,

So the RTX 4090 is 12% faster than the a770 in Video Enhance AI but costs 4.57x more?

RTX 4090 takes 3 PCI slots in your case, you could just buy 3 a770 units. For 65% of the cost of the RTX 4090 you would get 2.7x the workload

Is this right?

1 Like

Going to take the guy’s graph with a big grain of salt. He’s on prerelease drivers, a VEAI that hasn’t been updated for 40-series, and who knows what other settings he used that may have bottlenecked the GPU. Because in my experience, the sec/frame margin between 3090 and 3080 is wider than what his graph expresses.

As usual you can’t assume youtubers know what they’re doing and, while it’s an interesting datapoint, ultimately have to do your own testing in the configuration and with the presets you actually use.

Are you sure that EposVox actually tested something and shows correct numbers on VEAI graph? Jut tried to convert 1000 720p frames to 4k using 3090 and result was 0.58s for frame.
But more questionable is 0.511 for RTX 3060…
I asked my grandson to use his PC with RTX 3070Ti for a few minutes and VEAI 2.6.4 shows 0.82s for each frame on same 720p source. Suppose that in case of 3060, number must be more then 1s?

What CPU did you use during the test ? :thinking:
I believe EposVox was using 12900k during the test.
You may also click on the link and ask him directly in the comment section.

I got RTX 3090 ( Colorful Vulcan OC-V 24GB LHR), Ryzen 9 5900X on ASUS PRIME B550M-A motherboard and Zalman ZM800-TXII V2 power supply
And my grandson runs MSI RTX 3070Ti with Ryzen 7 2700 on GIGABYTE B450M S2H motherboard with 650W power supply (it is my old PC)

:nerd_face:

@suraj

FYI: If you are looking for a Gigabyte RTX 4090 Overclockable video card, today October 12th. (Currently at list price, free shipping.) They have them in stock for (US only) Gigabyte’s new on-line store.

I just bought mine!!!

:cowboy_hat_face: :nerd_face: :grinning_face_with_smiling_eyes:

1 Like

I’m not sure about the A770 (i would need a Whitepaper for its achitecture to see whats going on).

But the numbers for 3080 vs 2080 are legit, is 29 vs 22 teraflops.

And PCI-E 4 plays a role.

Rx 7900X will have PCI-E 4 & 132 Teraflops fp16. ← so maybe a good piece faster than 4090.

Update: Seems like A770 has a tensor like (Matrix Operations) Core.

https://www.intel.com/content/www/us/en/products/docs/arc-discrete-graphics/xe-hpg-microarchitecture.html

I heard similar from RDNA3.

Update 2:

https://videocardz.com/newz/amd-adds-wmma-wave-matrix-multiply-accumulate-support-to-gfx11-rdna3-architecture-amds-tensor-core

1 Like

Hmmm,

I wonder if the SDKs provided by the makers of this hardware will support all these cards or just the more popular ones on the market. - I know that programmers like to keep what’s called 'scope creep to a minimum, so they may only want to write and debug code that will run with the most widely used hardware. - In that case the 3090, 4090 and Intel (advanced) video stuff has a greater chance of being considered for optimization.

1 Like

Thats why Apples hardware does work realy fast for its raw performance.

You don’t have to optimize it for a lot of different hardware.

I heard that in gaming you optimise for the weakest hardware of an generation.

Personally, I am not much of an Apple fan. But, having the same company that manufactures the stuff and provides the OS, drivers and SDKs in control of it all does give them a certain advantage when optimizing.

I know their history and I know the evolution of the IBM PC to what it has morphed into today. - It may be more chaotic, but I’m a fan of structured code, knowledgeable systems programming and the freedom of choice it provides.

I’m also not a gamer, so I have no idea what their programming philosophy is. But that sounds reasonable. - Why deny an eager gamer a chance to BUY and play your game just because their system is a bit too old or underpowered?
:heavy_dollar_sign::scream: :nerd_face: :disappointed_relieved: :heavy_dollar_sign:

aaaannnnnd?

So slow that you can’t show it? :wink:

The other thread did show only 22% speedup (best case, vs 3090), Proteus 720p to 1080p.

Well,
When it gets here, we’ll find out. I’m sure it will be faster than my 3090, which has been moving faster lately, I think that Topaz has gotten into the Nvidia SDKs.

I’m hoping they’ll implement the additional stuff available to take advantage of the 4090’s capabilities, too.

One of the problems I see in the speed comparisons posted so far, is that they don’t mention whether the benchmarks provide driver support for all the various GPUs, or just use “generic” calls.

It’s starting to look to me like it depends more on the CPU than the GPU, at a certain level.

For ADA and AMP to work properly everything has to be floating point.

I would like to see everything loaded into gpu memory, then the performance should be outstanding, like rendering.

I agree. And, the Nvidia 3000 and 4000 series have more than adequate memory and a wide data path to the GPU. I would think their driver SDK has been designed to optimize how the various cores and onboard memory are used.

Having a fast line between the system CPU and the video card resources is also very important. On 11th, - 12th gen Intel chipsets the first PCIe x4 slot feeds directly to the CPU. the same PCIe x4 (in my Z590 machine with an i9-11100K CPU in it) also services my first NVMe drive. - It’s pretty fast.

I have a fair number of monitoring utilities on my system, and occasionally active a few just to see how much work the various components are doing. It wasn’t all that long ago when VEAI only used a few % of the Nvidia’s capabilities. Now that they offer Nvidia-specific drivers for some CODECS and their containers, things have sped up significantly.

That said, the thing that will truly make VEAI’s enhancement capabilities really pop will be the adoption of a few lossless codecs.

In fact, compressing slows down processing a lot, that’s why I always use tiffs for pictures (and of course for the quality).

The primary need for lossless video is when you are editing or enhancing. It should never need to be compressed until it gets finalized for distribution on whatever media it is intended for.

What VEAI needs is a lossless CODEC that is optimized to work on modern high-capability hardware. The Nvidia GeForce RTX GPUs are only one example. (But good ones. )

These results can be skewed by the compatibility of the benchmark being used. If the benchmark doesn’t use a driver that can utilize the explicit capabilities of the GPUs and just runs a “generic” standard graphics driver on all of them, the results are meaningless for most practical purposes.

1 Like