Question What to get? Upgrade time!!!

[harrycallum]

Hardware upgrade time. Or at last considering. All I know is I require more vRAM so that means A6000, the newer Ada Lovelace A6000, or perhaps the RTX 5500-5000 since 24GB RAM would be sufficient.
I just know my current 12GB is a bottleneck.
I do mostly Unreal cinematics and Houdini simulations. In Unreal I'm often crashing due to vRAM constraints.
Considering the prosumer level RTX 4090 but it seems like quite the power hog!
I'm attracted to the higher end Quadro cards for reliability, efficiency and smaller form factor considerations.

And ideally want to avoid having to rebuild an entire new PC around a GPU upgrade. Currently have an older ROG Zenith Xtreme x399 motherboard, dual 12GB (Titan Volta) cards and EVGA 1200w power supply.

When it comes to reviewing hardware specs I quickly feel a migraine coming on so just 2 questions
Which GPU should I target? This probably means looking at a lot of benchmark comparisons and making a choice based on price vs performance. Speaking of which benchmarks can be confusing This shows the A5000 way outperforming the A6000 for typical gaming RTX A5000 vs Quadro RTX A6000 - Technical City
What is the best way of determing motherboard compatibility?
Thanks for any feedback!

NVIDIA RTX & Quadro Desktop Workstations

 

[Tech Junky]

Without knowing the system build it's hard to answer.

If it's Intel the bottleneck is going to be the PCIE lanes due to routing them off the DMI for the 2nd card as it's sharing bandwidth with everything else unless it's a higher end board / CPU like Xeon.

AMD on the other hand allows for more flexible PCIE use across more slots w/o using a controller to pass the signals back to the CPU.

I've noticed this through looking at potential PCI card for expanding NVME storage. When doing this on Intel it requires an expensive card ($600) with a PLX switch on the card to get multiple drives working and limits the throughput to ~Gen3 3.5GB/s speeds vs a cheap $100 card for AMD systems that gives full 4x4x4x4x bifurcation on the slot vs using the PLX controller.

The Quadro series is more serious when it comes to compute power than a gaming oriented GPU.

ROG ZENITH EXTREME | Motherboards | ROG United States

The ASUS PRIME X299-DELUXE is the prime choice for prosumers and content creators with 1-click overclocking from 5-Way Optimization, ASUS Pro Clock Technology, and multiple GPU support featuring SafeSlot protection. X299, HEDT, ASUS, Prime, Deluxe, ATX, Motherboard, Intel, socket, LGA2066, AURA...

rog.asus.com

Looking at the specs of the MOBO it appears the limitation might be the PCIE being at Gen3 Moving to the TRX/WRX would open more bandwidth to Gen4 Going over to the non-TR with a X570 would offer most of the same features but, the drawback would be SATA lanes / ports if you're big into data hoarding with a bunch of drives. If you're going to do a rebuild though then moving up to the current X670 would be the best best for your money for holding onto it for a long period of time.

Ultimately it comes down how much you value what you're doing and whether or not it's worth the $$$$ to do it.

 

[Stuka87]

Your PSU should work fine for a 4090. And that 4090 while being a power hog, will use less power than the two Titans you have now (Titan V is 250W each).

Well, unless you want two 4090's. Then its going to be tight. Especially with a thread ripper.

 

[igor_kavinski]

Speaking of which benchmarks can be confusing This shows the A5000 way outperforming the A6000 for typical gaming RTX A5000 vs Quadro RTX A6000 - Technical City

You are better off using techpowerup GPU database for comparisons.

NVIDIA RTX A6000 Specs

NVIDIA GA102, 1800 MHz, 10752 Cores, 336 TMUs, 112 ROPs, 49152 MB GDDR6, 2000 MHz, 384 bit

www.techpowerup.com

NVIDIA RTX A5000 Specs

NVIDIA GA102, 1695 MHz, 8192 Cores, 256 TMUs, 96 ROPs, 24576 MB GDDR6, 2000 MHz, 384 bit

www.techpowerup.com

A6000 is faster than 4070 Ti and A5000 is 25% slower than 4070 Ti.

 

[thecoolnessrune]

The big thing to keep in mind here when looking at the RTX A6000 is that NVIDIA in their infinite wisdom has multiple generations that are similarly named and are very different GPUs. The RTX A6000 is a 48GB card with an Ampere GPU (GA102) and a 300W TBP. The RTX 6000 Ada Generation (yes that's its name, noting the lack of "A" next to 6000). is a 48GB card with the Ada GPU (AD102) and a 300W TBP. And yes, there's also an old RTX 6000 (without Ada in the name) as a 24GB card with a Turing GPU (TU102).

The old Turing RTX 6000 is based on a TITAN RTX, the RTX A6000 is based on a RTX 3090, and the RTX 6000 Ada Generation is based on a RTX 4090. The biggest difference between the consumer and workstation GPU is that the workstation GPUs are capped fairly firm at 300W TBP and have to carry the extra power budget of 48GB of ECC RAM. So a 4090 will outperform an RTX 6000 Ada Generation in workloads like Unreal with relative ease because it has a 450W TGP.

If your workload is comprised mostly of just Unreal, then you can probably save a lot of money and get a 4090. Feel free to reduce the power budget to get it to operate more efficiently. The RTX 6000 Ada Generation will be much more important for very large data sets and ray tracing, but it is a lot more money for what will always* be less performance due to the power budget.

*Unless you're using professional features like double precision that are locked to lower performance on the consumer cards.

NVIDIA RTX 6000 Ada vs RTX A6000 for Content Creation

The NVIDIA RTX 6000 Ada is the latest addition to the NVIDIA's professional family of GPUs. With cutting-edge hardware and the latest Ada Lovelace architecture and 48GB of VRAM, this GPU should be terrific for a wide range of content creation workflows.

www.pugetsystems.com

 

[Aapje]

If 24 GB is enough, just get a 4090 and power limit it to your liking. That's better value than the Pro line.

The Pro line already being power limited and having worse cooling just reduces your options. You can always increase the power limit again of the 4090 if you have a really important job that needs to be done yesterday, but you can't run the Pro cards that high.

I don't know how significant the impact is of PCIe 3 for rendering, but I suspect it's not that bad.