• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

OCUK RTX4060TI review thread

Thanks for clearing that up I thought I was losing my mind or Nvidia were moreso. :)

E: I agree which is why I was surprised they knocked out a 12Gb 3060

Because it would have been a 6GB card otherwise,and Nvidia really wanted to it be above £300. It was also slower than the RX5700XT 8GB too.
 
Last edited:
I wouldn't say it's a load of ****, it's a niche for a certain type of creative and unfortunately Nvidia wipes the floor with AMD & Intel for compatibility

FIgugi6.png

7xxxx AMD series do better but still lag behind their green equivalent

Stable Diffution depends on a suit of AI extentions that older GPU's don't have. even RDNA 3 only has a limited set of those instructions, probably there for Image generation later on, the full set is reserved for CDNA which is a different architecture, Ampere and Lovelace do, its built in to the Tensor Cores.
 
Last edited:
I just read the article that slide came from ^^^^ Toms expressed surprise at Intel ARC lack of performance, citing that "But they have FP16" so does RNDA2, Its running on FP16 because it doesn't have INT8 or INT4 which is much faster than FP16 for AI, RDNA2 doesn't have those either and i think RDNA3 only has INT8.

CDNA has INT4.

Jessus ###### christ...
 
Last edited:
I just read the article that slide came from ^^^^ Toms expressed surprise at Intel ARC lack of performance, citing that "But they have FP16" so does RNDA2, Its running on FP16 because it doesn't have INT8 or INT4 which is much faster than FP16 for AI, RDNA2 doesn't have those either and i think RDNA3 only has INT8.

CDNA has INT4.

Jessus ###### christ...
Nerd terms aside though, Nvidia is just better, the chart was just illustrative of that fact
 
Because despite all the deniers they get people to upgrade quicker. Here a bunch of RTX3060TI owners bought an RTX4070,which is £100s more expensive. Then when you need 16GB or more,make sure the cheapest card is expensive and people will be "forced" to upgrade for a higher price within a few years.

Well I'm upgrading from a 1660 Super so almost anything from last couple of generations is going to be an improvement. I tend to upgrade every 3 or so generations.
 
Well I'm upgrading from a 1660 Super so almost anything from last couple of generations is going to be an improvement. I tend to upgrade every 3 or so generations.

If you look at the last few generations around the $300~$400 mark. The GTX970 was around 40% faster than the GTX770. The GTX1070 was around 60% faster than a GTX970(twice the VRAM). The RTX2060 6GB was 20% faster than the GTX1070. The RTX2060 Super was just under 40% faster than the GTX1070. The RTX3060TI was around 40% faster than the RTX2060 Super. The RTX4060TI is around 10% faster than the RTX3060TI.

The RTX4060TI is going to cost the same as the RTX3070 and be slower in most games which use under 8GB of VRAM.

The RTX4000 series is the worst generational improvement in the sub £600 area for the better part of a decade. It looks worse than Turing V1 in some ways! :(

But it even gets worse when you look at the chips used.

3JG8a0l.png


Compared to the top die,the RTX4070 is actually the real equivalent of your GTX1660TI. The RTX4060TI is really a 50 class die.

I would wait for the RTX5000 series,or an RTX4000 Super refresh. If you can't wait now,I would just get an RTX3060 12GB for £279(it might even drop further in price):

The RTX4060TI 16GB is going to be £200 more. Plus like the RX7600,it is underprovisioned in terms of cache(both have 32MB),so this is why they both have issues at higher resolutions and with some games. Plus unless you have a PCI-E 4.0 system,expect a few extra percent performance to be lost.

I would not be paying much over £250 for an 8GB dGPU in 2023.
 
Last edited:
The RTX4060TI is around 10% faster than the RTX3060TI

That's really the crux of it, put simply, its a 3060Ti OC for the same $399 3060Ti money.

So who is this for? If you bought a 2060 Super for $399 and you haven't replaced it with a 3060Ti by now why would you do that, now?

I have a 2070 Super, i paid £450 for it, the same brand and model 4070Ti is £890, a 2X increase in price, why would i do that?

I'm not an unreasonable man, i get that costs have gone up, inflation, so i get that things cost more, but 2X?
 
Last edited:
I own a 3060ti, largely because I was badly in need of an upgrade and managed to get one for a "good price" in the market at the time of purchase.

Which is to say I still felt a little ripped off, the 4060ti is a flat out insult.

Hard pass, I'm happy to wait a generation or two until the GPU world gains some sanity.

If I was buying now it'd be a 6700XT, after cashback they can be had for under £300.
 
Last edited:
Compared to the top die,the RTX4070 is actually the real equivalent of your GTX1660TI. The RTX4060TI is really a 50 class die.

I would wait for the RTX5000 series,or an RTX4000 Super refresh. If you can't wait now,I would just get an RTX3060 12GB for £279(it might even drop further in price):

The RTX4060TI 16GB is going to be £200 more. Plus like the RX7600,it is underprovisioned in terms of cache(both have 32MB),so this is why they both have issues at higher resolutions and with some games. Plus unless you have a PCI-E 4.0 system,expect a few extra percent performance to be lost.

I would not be paying much over £250 for an 8GB dGPU in 2023.

Thanks for the analysis and chart, it really helps put things in perspective. I've been leaning towards a 12GB 3060 for some time, but thought I'd wait and see what the midrange 40xx series turned out like - I really didn't expect such expensive and underspec'd cards.
 
Thanks for the analysis and chart, it really helps put things in perspective. I've been leaning towards a 12GB 3060 for some time, but thought I'd wait and see what the midrange 40xx series turned out like - I really didn't expect such expensive and underspec'd cards.

If you're not stuck on Nvidia, the 6700XT is actually a little faster than a 3060ti while coming with a cashback scheme that puts it under the £300 mark:

My basket at OcUK:

Total: £346.98 (includes delivery: £7.99)


It's not listed on OCUK but it should still be valid, someone else might be more help there as I don't have the proper info fully at hand. It also comes with TLOU 2, which you could either keep or sell for £20 (the price the last code went for on the MM) to bring the price down further.


Edit: My bad! It's actually the Asus card which comes with the cashback deal:


See this article: https://www.eurogamer.net/digitalfo...is-amazing-amazon-deal-on-the-asus-rx-6700-xt

Little bit of a faff, but you could net one for around the £275 mark when factoring in the sale of the game code.
 
Last edited:
Thanks for the analysis and chart, it really helps put things in perspective. I've been leaning towards a 12GB 3060 for some time, but thought I'd wait and see what the midrange 40xx series turned out like - I really didn't expect such expensive and underspec'd cards.
The RTX4060TI should have only had a 16GB version at the current price of the RTX4060 8GB.

The RX6700XT with cashback is probably the best pure gaming upgrade in terms of price/performance over your GTX1660 Super.


However,as you want to run Stable Diffusion then the RTX3060 12GB makes the best sense as a compromise. The RTX3060TI is also available for £300 now,but only has 8GB of VRAM(I think you said more than 8GB is useful).

Edit!!

Fw0KjSraYAEA0Pz
 
Last edited:
I don't know a lot about running AI models on GPU's, but if that's a goal and you genuinely are considering a 4060ti? I would consider the memory bandwidth as much as the amount of available memory.

Even when investing in the 16gb model (which to my mind would be insanity in any use case) that's a 128 bit bus with 288GB/s bandwidth.

I can't say for certain that it would be a concern, but it does come to mind as a bit of a potential red flag. I can't see the 16gb model being able to properly utilise all of its memory with those specifications in general tbh.
 
Last edited:
8GB is fine but you might be better off with the 16GB variant, depends on how serious you are about AI Art, just having enough can become not enough.

That is if you're going 4060Ti.
 
Last edited:
I don't know a lot about running AI models on GPU's, but if that's a goal and you genuinely are considering a 4060ti? I would consider the memory bandwidth as much as the amount of available memory.

Even when investing in the 16gb model (which to my mind would be insanity in any use case) that's a 128 bit bus with 288GB/s bandwidth.

I can't say for certain that it would be a concern, but it does come to mind as a bit of a potential red flag. I can't see the 16gb model being able to properly utilise all of its memory with those specifications in general tbh.

Yeah, the memory bandwidth is a factor. But from the digging I've been able to do on forums and Discord, it's a less important factor for Stable Diffusion performance than amount of Vram. So a 4060 Ti 16GB should outperform a 12 GB 3060 despite the lower memory bandwidth on the 4060 Ti. But £200 worth of extra performance? Nope.
 
Yeah, the memory bandwidth is a factor. But from the digging I've been able to do on forums and Discord, it's a less important factor for Stable Diffusion performance than amount of Vram. So a 4060 Ti 16GB should outperform a 12 GB 3060 despite the lower memory bandwidth on the 4060 Ti. But £200 worth of extra performance? Nope.

I thought it might play a factor, I'm shocked they're asking so much more for the memory bump tbh.
 
Back
Top Bottom