It looks like the 'real' /affordable RDNA3 + next gen NV desktop launch won't launch until September. Thoughts?

[TNA]

I don't see ratchet and clank on PC and ff16 is a ps5 exclusive.

Regards to ratchet and clank, it looks and runs amazing and looks better then cyberpunk oh and it's a better game then it.

FF16 will be on PC soon enough and with Sony bringing exclusives to pc won't be long Ratchet and Clank will be on PC too.

 

[timorous]

There is a 7.6% clock difference between the MBA card and the Merc 310, this with a 5.4% difference in performance, that's a 7/10 scaling, bang on what one would expect.

The memory clock, 30Mhz, 0.01% is nothing.

The Taichi and the Nitro+ are in the same ball park with 1 or less that 1 FPS difference with similar core clocks.

2780/2750 is 1.09% not 0.01%

2750 ram is limiting at 2.9Ghz or so clockspeeds.

This is why the Nitro + is slightly ahead of the taichi white and the Merc is slightly ahead of the Nitro+.

If the TUF clocked over 2.9Ghz it would probably top the chart.

 

[humbug]

2780/2750 is 1.09% not 0.01%

2750 ram is limiting at 2.9Ghz or so clockspeeds.

This is why the Nitro + is slightly ahead of the taichi white and the Merc is slightly ahead of the Nitro+.

If the TUF clocked over 2.9Ghz it would probably top the chart.

That's one very select example.

The Taichi is 106.2 FPS vs 109.6 FPS despite the Taichi having higher memory clocks, what the Nitro+ does have is higher core clocks, so again no memory bandwidth bottleneck.

Taichi White again stock memory, high core, high frame rates.

The Merc, the fastest of them all, over 2900Mhz, lower memory clocks than the Taichi.

 

[g67575]

You would think the 4090 is 50% faster than the 7900XTX.

The 4090 on average is 22% faster at 4K, source TPU, 356 Watts 7900XTX vs 411 watts 4090.

Indeed.

But I think it would probably make more sense to wait until they can use an optimised TSMC 3nm process to scale up the compute units beyond 100.

I wonder if the vcache versions of RDNA3 exist /are in development? I can see that providing a worthwhile improvement.

 

[Joxeon]

You would think the 4090 is 50% faster than the 7900XTX.

The 4090 on average is 22% faster at 4K, source TPU, 356 Watts 7900XTX vs 411 watts 4090.

TPUs 4090 data is flawed as they used a 5800x for the 4090 testing.

While nowhere near 50% the reality is it’s closer to 30% when tested with a current gen CPU.

 

[CAT-THE-FIFTH]

Well lets see how the RX7600 8GB is priced? Leaks indicating it might be around $300!

 

[g67575]

£300 for a RX7600 doesn't seem too outlandish.

Especially when considering that it could fall to the price of the RX 6600/6650 XT in 2023.

 

[Joxeon]

£300 for a RX7600 doesn't seem too outlandish.

Especially when considering that it could fall to the price of the RX 6600/6650 XT in 2023.

$300 is an insult especially when you consider intel are selling a card with double the size TSMC 6nm die and double the VRAM + bus for around the same price.

 

[Ross Thomson]

Looking at the specs 32cu, full navi 33 die and the 165w TBP it looks like the 7600 is a replacement for the 6650 XT rather than the 6600. 33 more watts than the 6600.

 

[Ross Thomson]

$300 is an insult especially when you consider intel are selling a card with double the size TSMC 6nm die and double the VRAM + bus for around the same price.

But the have to because no one is buying it. 16GB beam doesn’t made a card good

 

[Grim5]

It's also worth noting Intel are probably making very little or no profit selling such a large GPU on a near cutting edge process node and with 16gb vram at that price where as AMD wants 50% profit

 

[g67575]

$300 is an insult especially when you consider intel are selling a card with double the size TSMC 6nm die and double the VRAM + bus for around the same price.

Not true is it? The Arc A770 16GB costs ~£400 at the moment. The performance is similar to the RX 6600 XT.

 

[g67575]

Looking at the specs 32cu, full navi 33 die and the 165w TBP it looks like the 7600 is a replacement for the 6650 XT rather than the 6600. 33 more watts than the 6600.

yup.

On a separate note - It ought to be clear to people by now, that the Navi33 and AD106 GPUs aren't remotely exciting, and this should not be a surprise, they are low end and cheap to mass produce.

The only interesting feature they both support is frame generation (with FSR3 still to come).

On paper, the main performance difference appears to be the approx. 2x FP32 processing power of the Navi33 GPU, compared to Navi23.

The RTX 4060 TI 8 or 16GB card is based on the AD106 GPU, notably not AD104, which would have been the same approach that they took with the RTX 3060 TI (based on GA104).

Therefore, it cannot really be considered to be a successor to the RTX 3060 TI, despite it's name.

 

[Joxeon]

Not true is it? The Arc A770 16GB costs ~£400 at the moment. The performance is similar to the RX 6600 XT.

Only because it’s pretty much out of stock everywhere currently but was £330 a couple of weeks ago, regardless the 8gb version is easily available at £280 and packs double the die size, bus and requires a more complex PCB than a 7600 due to this and the higher TDP.

 

[SpudMaster]

Regarding Arc, well all know Intel haven't chosen to sell it at its current price out of the goodness of their hearts. They are forced to because no one wants it. They are probably just breaking even.

 

[Raelgun]

FF16 will be on PC soon enough and with Sony bringing exclusives to pc won't be long Ratchet and Clank will be on PC too.

Feel free to wait years

You also realise ratchet and clank has been out for years already right?

 

[timorous]

That's one very select example.

The Taichi is 106.2 FPS vs 109.6 FPS despite the Taichi having higher memory clocks, what the Nitro+ does have is higher core clocks, so again no memory bandwidth bottleneck.

Taichi White again stock memory, high core, high frame rates.

The Merc, the fastest of them all, over 2900Mhz, lower memory clocks than the Taichi.

Yet going from the TUF to the Taichi White only gives a 0.6% performance gain for a 2.8% clockspeed gain. Looks rather memory bound to me when TUF to Merc gives a 1.5% gain for a 2.4% clockspeed gain. The extra memory clock of the Merc over the Taichi white seems to be helping quite a bit.

Point is 2.85Ghz core and 2.8Ghz memory are fairly balanced so to see good gains above that you need to OC both the core and the memory beyond that to see the best results and is why the Merc is the best in the test. It has the best balance of core clock and memory clock where as the TUF has a great memory clock but weaker core clock and the Taichi White / Nitro + have a stronger core clock but weaker memory clock.

Also the standard Taichi to TUF with the same memory clock is showing linear clockspeed to performance scaling.

 

[humbug]

Only because it’s pretty much out of stock everywhere currently but was £330 a couple of weeks ago, regardless the 8gb version is easily available at £280 and packs double the die size, bus and requires a more complex PCB than a 7600 due to this and the higher TDP.

Intel launched that card at £450 ffs.... it was as' if not more expensive at the time than an RTX 3060Ti and slower than a £300 RX 6600XT.

Its was that until about 2 months ago... Intel are not the great saviour, very obviously not, we are cheering on someone who is worse than Nvidia and treating the only people who have tried to be different as an inconvenient thorn in the side to an agenda. WTF is wrong with us????

 

[humbug]

The amount of cheery glossing over of the problems with that card and pushing of it by almost all media outlets was ####### astonishing to me, and at these prices, if the far cheaper and far better RX 6600XT was ever mentioned at all by any of these people it was like a small print footnote legally required but really ####### inconvenient.

Mind-blowing, and so many people got suckered by it, some still are.

We deserve all the crap that is going on now, they look, they see, they think we stupid.

 

[Jono8]

It is bizarre how Intel thought they could walk into the GPU market and ask for more money than equivalent performing Nvidia/AMD cards.

Did no one in their management think "Hm, guys we might have to bring amazing value to the market for us to be taken at all seriously at first".