When the Gpu's prices will go down ?

[Zarax]

The irony is that the 4090 manages to be both overkill and useless at the same time.
It's likely faster in raster than any CPU can manage right now, yet they overdid it by having a new "RT ultra plus" setting in CP2077 where it doesn't reach 60fps, meaning that "true RT" is still at least one generation away.
Ironically it just persuaded me that we're at least a couple gens from having meaningful RT rather that to fork the cash...

 

[Noxia]

Ignore the 6950 XT....I picked up a MSI 6900 XT for £629 about two weeks ago and the drivers I have think it's a 6950 XT and running at those clocks.

Yeah with the voucher codes for the calisto project/dead island it can bring you just under £600.

 

[Grimfusion]

The irony is that the 4090 manages to be both overkill and useless at the same time.
It's likely faster in raster than any CPU can manage right now, yet they overdid it by having a new "RT ultra plus" setting in CP2077 where it doesn't reach 60fps, meaning that "true RT" is still at least one generation away.
Ironically it just persuaded me that we're at least a couple gens from having meaningful RT rather that to fork the cash...

Just wait until the 4090TI comes out. Rumours of it appearing in 3DMark testing at the moment.

 

[Zarax]

Just wait until the 4090TI comes out. Rumours of it appearing in 3DMark testing at the moment.

I'm actually waiting for my 5800x3d +6800xt rig to be shipped, Nvidia can try again to get my money in about 5 years

 

[TrojanWhore]

Exchange rate is going the right way, $1.18 currently

 

[Noxia]

Judging by pricing, you'd think retailers are watching two different exchange rates considering the difference between them

 

[RobinsonJ]

What the hell is going on with graphics card prices right now? I go to NVIDIA website. What's available is incorrectly priced, way above MSRP, MSRP is stupidly priced - (I could buy a decent second hand car for the cost of most of these cards), some previous generation cards are almost as expensive as new 40 series (Palit GeForce 3090 is £1936, MSI 4090, £1919). It's a complete cluster****. The market seems completely broken and/or this confusion is deliberate.

 

[Trypta]

What the hell is going on with graphics card prices right now? I go to NVIDIA website. What's available is incorrectly priced, way above MSRP, MSRP is stupidly priced - (I could buy a decent second hand car for the cost of most of these cards), some previous generation cards are almost as expensive as new 40 series (Palit GeForce 3090 is £1936, MSI 4090, £1919). It's a complete cluster****. The market seems completely broken and/or this confusion is deliberate.

It's Nvidia in their arrogance refusing to lower prices in an attempt to sustain growth after the boom of gpu sales during the crypto surge. They care for their investors, not consumers.

You're correct; it is very much deliberate, and it would be best none of us entertained it either.

 

[newuser]

I am a creator in 3D and so for me Nvidia and ray tracing and Cuda seems the correct and only way to go for me. i have recently considered going AMD with the new cards announced being cheaper but i think for what i need i don’t have much option than to go green, untill AMD gets more industry support.

The 12gb 3060 is a popular card for rendering if your projects fit in the 12gb of vram. A 3060 will render at about the same speed as a 2080ti. That's assuming your main constraint is vram and that you need cuda.

If you're offloading most of the work onto the GPU then a 3060, a recent mid range cpu, a functional mboard and 32gb of DDR4 isn't a bad place to be.

 

[ninjag]

The 12gb 3060 is a popular card for rendering if your projects fit in the 12gb of vram. A 3060 will render at about the same speed as a 2080ti. That's assuming your main constraint is vram and that you need cuda.

If you're offloading most of the work onto the GPU then a 3060, a recent mid range cpu, a functional mboard and 32gb of DDR4 isn't a bad place to be.

I was looking at the 3060 compared to the 3060Ti and was wondering how the 12 GB of VRAM at 198 Bit bus width compares to 8 GB VRAM at 256 Bit bus width, can the 8 GB make up for the lesser capacity by being able to transfer quicker or will the extra capacity of the 3060 being overwhelmingly better?

I have noticed the 3060Ti runs rather cool at 27°C in explorer, could that be the large die used? Even under heavy load it still stays under 60°C.

 

[RobinsonJ]

The 12gb 3060 is a popular card for rendering if your projects fit in the 12gb of vram. A 3060 will render at about the same speed as a 2080ti. That's assuming your main constraint is vram and that you need cuda.

If you're offloading most of the work onto the GPU then a 3060, a recent mid range cpu, a functional mboard and 32gb of DDR4 isn't a bad place to be.

Oh, really? I do rendering with iRay and Blender. I was looking to boost it (I have a 2070 at the moment, which is dated by now). I didn't know I could get a 3060 12GB had the same render performance as a 2080Ti. I thought it was about 2070 performance.

 

[JediFragger]

They care for their investors, not consumers.

Tbf to Nvidia I'd ditch you bunch of toxic ****s faster than a case of herpes if I could

Investors got money y0!

 

[Angilion]

I was looking at the 3060 compared to the 3060Ti and was wondering how the 12 GB of VRAM at 198 Bit bus width compares to 8 GB VRAM at 256 Bit bus width, can the 8 GB make up for the lesser capacity by being able to transfer quicker or will the extra capacity of the 3060 being overwhelmingly better?
[..]

That would depend on the workload, specifically how much VRAM the workload uses at the same time. On the whole, it'll be like this: 0-8GB and the 3060Ti will be faster. 8-12GB and the 3060 will be faster. 12GB+ and the 3060Ti will be faster. Having insufficient VRAM for the job heavily reduces performance. But it'll vary to some extent depending on how often and for how long the job requirements exceed the available VRAM. If it's 8-12GB for a small enough part of the job time, the higher performance of the 3060Ti for the rest of the job might more than offset the lower performance during that small part of the job time.

That's my impression, anyway. I'm far from an expert on rendering workloads.

 

[RobinsonJ]

I read on the Daz forums that you bench 4.531 iterations per second with RTX 2070, and 10.648 iterations per second with 3060 Ti. That really surprises me. But 3060 has second gen RT cores so perhaps it's possible.

 

[Chuk_Chuk]

That would depend on the workload, specifically how much VRAM the workload uses at the same time. On the whole, it'll be like this: 0-8GB and the 3060Ti will be faster. 8-12GB and the 3060 will be faster. 12GB+ and the 3060Ti will be faster. Having insufficient VRAM for the job heavily reduces performance. But it'll vary to some extent depending on how often and for how long the job requirements exceed the available VRAM. If it's 8-12GB for a small enough part of the job time, the higher performance of the 3060Ti for the rest of the job might more than offset the lower performance during that small part of the job time.

That's my impression, anyway. I'm far from an expert on rendering workloads.

With regards to your statement on job time, for rendering if you exceed VRAM it will be till the end of the render. VRAM has little to no fluctuation during the sampling process of the render, which is the main bulk of time. That doesn’t necessarily mean you are constantly pinging those items that are off of VRAM.

With how rendering works my assumption is that the model/texture that is bumped off VRAM to RAM will be pinged whenever a ray of light needs to query that model/texture information.

However I know cycles (I’m assuming other software are the same) will sometimes throw up an error message of you attempt to render a scene that does not fit in VRAM.

 

[Chuk_Chuk]

I read on the Daz forums that you bench 4.531 iterations per second with RTX 2070, and 10.648 iterations per second with 3060 Ti. That really surprises me. But 3060 has second gen RT cores so perhaps it's possible.

Have you compared it on the blender open data benchmarks website just to double check those numbers?

 

[RobinsonJ]

Have you compared it on the blender open data benchmarks website just to double check those numbers?

Vanilla NVIDIA GeForce RTX 3060 is 2426.48 and vanilla NVIDIA GeForce RTX 2070 is 2244.25. That's CUDA and Optix on Windows ticked. So the 3060 is a bit faster for blender but of course has the 12GB so you can detail up your environments more than you can with the 2070.

 

[Grimfusion]

I'm actually waiting for my 5800x3d +6800xt rig to be shipped, Nvidia can try again to get my money in about 5 years

I was wondering if I should get the 5800X3d but I went mad and got the Ryzen 7 7700X and I have to say that I'm impressed. It cost a bit but at least it's future proof. Relative I guess if you are going to use the 5800X3d for 5 years.

 

[RobinsonJ]

OK I think I get it now, after more reading. Prices are ****** up because NVIDIA and partners still have a lot of 30 series cards to get rid of, and rather than drop the price on those they're reducing 40 series supply (that's why they're "sold out" at MSRP and can otherwise only be purchased either day 1 when they sell out, or at some massively stupid mark-up). They want us to buy previous generation cards at current generation prices. Only way to do that is restrict supply. Suppliers are reporting 4080 cards supply will be 20% LOWER than 4090 supply was, if you can believe it.

So I would have to wait until next year, probably summer next year, to get a decent, affordable 40 series card.

 

[Zarax]

I was wondering if I should get the 5800X3d but I went mad and got the Ryzen 7 7700X and I have to say that I'm impressed. It cost a bit but at least it's future proof. Relative I guess if you are going to use the 5800X3d for 5 years.

In over 30 years I never upgraded the CPU without changing everything else so future proofing in that sense means little to me.
Worst case, if 8 cores start becoming restrictive, I'll sidegrade to a 5950X which is likely to be quite cheap in 5 years or so...