What do you think of the 4070Ti?

[g67575]

RT performance isn't much better than the 7900XT either.... they are both bad for the money. we are paying £900 for mid level cards now, people buying these need a sanity check, people paying phone number prices for halo cards, i get that.... these are humble mass market level cards, its a Ford Focus not a Bentley. £900.

I don't agree. It's people still buying the RTX 3000 series that are the main problem, in terms of keeping demand high for these cheaper to produce cards.

The demand for the RTX 3000 series led to the (MSRP) prices of the next generation of Ampere.

 

[g67575]

Not really Gibbo's fault is it?

When the 7900s went of sale and were selling out some people became interested in how well they were selling and how well the 4080s were selling and asking Gibbo for details. Gibbo lets people know how well these are selling and people get upset...

People are being a bit dumb unfortunately, cause of the unfortunate market conditions.

 

[CAT-THE-FIFTH]

Why are people complaining about the 4070 TIs 4K performance?

72 FPS minimums in most games. A little behind the competition (7900 XT), which I'd say is a bit better at 4K, but both can handle games at 4K. The minimums are 56% higher than the RTX 3070, an 50% higher than the RTX 4070 TI.

The ugly thing is the prices.

Because the dGPU is the size of the GA106 in the RTX3060 and the ratio of shaders is more like an RTX3060TI.

 

[Shewie]

Will 12GB be enough for games with HD texture packs?

For now yes, but it is not a future-proofing GPU.

 

[g67575]

Because the dGPU is the size of the GA106 in the RTX3060 and the ratio of shaders is more like an RTX3060TI.

It's pretty irrelevant. Transistor count is higher (35.8 billion for AD104, vs 28.3 billion for GA102 /RTX 3090 TI).

 

[CAT-THE-FIFTH]

It's pretty irrelevant. Transistor count is higher (35.8 billion for AD104, vs GA102 /RTX 3090 TI).

Stop excuse making. Over the last 20 years plenty of midrange series chips have had more transistors and you people been mass indoctrinated by marketing to make something which was pretty normal to be some big deal.

As for someone who has been into PC building for 20 years,it's shocking how people have had their expectations managed to such a low level,they have seemingly forgotten how much we had mainstream chips match or exceed past flagships. This is Apple level expectation management.

The RTX4070TI isn't a 70 series card but a rebadged 60/60TI series card. The GPU is under 300MM2 too like a GA106. People keep forgetting Nvidia released an AD103,so the AD104 actually is now the same position as the GA106 in the lineup and the AD103 is basically where the GA104 was positioned.

But more importantly look at the amount of shaders and memory bandwidth relative to the top end dGPU in the range,ie,the RTX4090.

But even ignoring die size,just look at the shader count:

NVIDIA GeForce RTX 4090 Specs

NVIDIA AD102, 2520 MHz, 16384 Cores, 512 TMUs, 176 ROPs, 24576 MB GDDR6X, 1313 MHz, 384 bit

www.techpowerup.com

NVIDIA GeForce RTX 4070 Ti Specs

NVIDIA AD104, 2610 MHz, 7680 Cores, 240 TMUs, 80 ROPs, 12288 MB GDDR6X, 1313 MHz, 192 bit

www.techpowerup.com

The RTX4070TI has 47% of the shader count of the RTX4090,half the VRAM capacity and 50% of the memory bandwith.

Now look at the RTX3090TI vs the RTX3070TI:

NVIDIA GeForce RTX 3090 Ti Specs

NVIDIA GA102, 1860 MHz, 10752 Cores, 336 TMUs, 112 ROPs, 24576 MB GDDR6X, 1313 MHz, 384 bit

www.techpowerup.com

NVIDIA GeForce RTX 3070 Ti Specs

NVIDIA GA104, 1770 MHz, 6144 Cores, 192 TMUs, 96 ROPs, 8192 MB GDDR6X, 1188 MHz, 256 bit

www.techpowerup.com

The RTX3070TI has 57% of the shader count of the RTX3090TI and 60.3% of the memory bandwidth of the RTX3090TI.

Now look at how it is compared to the RTX3060TI:

NVIDIA GeForce RTX 3060 Ti Specs

NVIDIA GA104, 1665 MHz, 4864 Cores, 152 TMUs, 80 ROPs, 8192 MB GDDR6, 1750 MHz, 256 bit

www.techpowerup.com

It has 45.2% of the shader count of the RTX3090TI and 44% of the memory bandwidth.

The RTX4070TI relative to the RTX4090,is basically what the RTX3060TI was relative to the RTX3090TI for the previous generation,with a die size of the RTX3060/GA106 series.

 

[g67575]

Anyway, the price should be around £790 for the RTX 4070 TI (but no more than this), considering the performance improvement of 50% (1% lows) vs the RTX 3070 TI.

Or £733, considering the performance improvement of 56% (1% lows) vs the RTX 3070.

Ideally, it would be priced at £700.

 

[brostradamus]

Stop excuse making. Over the last 20 years plenty of midrange series chips have had more transistors and you people been mass indoctrinated by marketing to make something which was pretty normal to be some big deal.

As for someone who has been into PC building for 20 years,it's shocking how people have had their expectations managed to such a low level,they have seemingly forgotten how much we had mainstream chips match or exceed past flagships. This is Apple level expectation management.

The RTX4070TI isn't a 70 series card but a rebadged 60/60TI series card. The GPU is under 300MM2 too like a GA106. People keep forgetting Nvidia released an AD103,so the AD104 actually is now the same position as the GA106 in the lineup and the AD103 is basically where the GA104 was positioned.

But more importantly look at the amount of shaders and memory bandwidth relative to the top end dGPU in the range,ie,the RTX4090.

But even ignoring die size,just look at the shader count:

NVIDIA GeForce RTX 4090 Specs

NVIDIA AD102, 2520 MHz, 16384 Cores, 512 TMUs, 176 ROPs, 24576 MB GDDR6X, 1313 MHz, 384 bit

www.techpowerup.com

NVIDIA GeForce RTX 4070 Ti Specs

NVIDIA AD104, 2610 MHz, 7680 Cores, 240 TMUs, 80 ROPs, 12288 MB GDDR6X, 1313 MHz, 192 bit

www.techpowerup.com

The RTX4070TI has 47% of the shader count of the RTX4090,half the VRAM capacity and 50% of the memory bandwith.

Now look at the RTX3090TI vs the RTX3070TI:

NVIDIA GeForce RTX 3090 Ti Specs

NVIDIA GA102, 1860 MHz, 10752 Cores, 336 TMUs, 112 ROPs, 24576 MB GDDR6X, 1313 MHz, 384 bit

www.techpowerup.com

NVIDIA GeForce RTX 3070 Ti Specs

NVIDIA GA104, 1770 MHz, 6144 Cores, 192 TMUs, 96 ROPs, 8192 MB GDDR6X, 1188 MHz, 256 bit

www.techpowerup.com

The RTX3070TI has 57% of the shader count of the RTX3090TI and 60.3% of the memory bandwidth of the RTX3090TI.

Now look at how it is compared to the RTX3060TI:

NVIDIA GeForce RTX 3060 Ti Specs

NVIDIA GA104, 1665 MHz, 4864 Cores, 152 TMUs, 80 ROPs, 8192 MB GDDR6, 1750 MHz, 256 bit

www.techpowerup.com

It has 45.2% of the shader count of the RTX3090TI and 44% of the memory bandwidth.

The RTX4070TI relative to the RTX4090,is basically what the RTX3060TI was relative to the RTX3090TI for the previous generation,with a die size of the RTX3060/GA106 series.

have you ever wondered if 4090 is actually a new tier, even amd seem to have been caught off guard.. its 80 billion transistors
fundamentally these comparisons dont make a lot of sense, because engineering has to figure out the optimal tradeoff between perf, power, cost.. singling out individual parameters like die size, memory controller address size etc. doenst paint a good picture, what you should be instead talking abt is fps/watt or such kind of metrics when comparing graphic cards within a price tier
now the market being the way it is i am not sure if the higher price is just due to higher cost or lack of competition

 

[uscool]

Anyway, the price should be around £790 for the RTX 4070 TI (but no more than this), considering the performance improvement of 50% (1% lows) vs the RTX 3070 TI.

Ideally, it would be priced at £700.

Still overpriced for product claiming to be 70 tier card I don't go by let's decide how much performance from last gen and increase price

 

[g67575]

So much ranting, keep going on about different die names as if they matter

They do, but only in the context of a single generation. Each gen is different.

 

[CAT-THE-FIFTH]

So much ranting, keep going on about different die names as if they matter

They do, but only in the context of a single generation. Each gen is different.

Your the one is desperately ranting. This is a technical forum and I am making technical arguments.

Maybe you need to spend less time on social media and stop have belief in companies. Unless consumers wise upto their games they will be played as mugs.

 

[varkanoid]

Anyway, the price should be around £790 for the RTX 4070 TI (but no more than this), considering the performance improvement of 50% (1% lows) vs the RTX 3070 TI.

Or £733, considering the performance improvement of 56% (1% lows) vs the RTX 3070.

Ideally, it would be priced at £700.

when you are talking ideals it be better off at £400.

 

[uscool]

So much ranting, keep going on about different die names as if they matter

They do, but only in the context of a single generation. Each gen is different.

Yeah but pricing stayed similar now they suddenly 2x ??

CPUs have had big improvements and the pricing has stayed similar?

 

[g67575]

Anyway, performance improvement is fairly good at around 50-56%.

Again, 50% improvement (1% lows) comparing the RTX 3080 10GB to RTX 4080 16GB (60 FPS 1% low vs 90FPS 1% low).

It's the prices that are unaffordable, I am not at all happy with these.

Some clearly can afford the prices of both the RTX 3000 series and 4000 series, if people keep buying, prices will stay high.

Eventually, most people just won't have the cash to spend, and prices should come down...

 

[g67575]

when you are talking ideals it be better off at £400.

Eventually. Not realistic, given the current prices of the RTX 3080.

If you want to blame someone, blame the people paying top whack for 'new' RTX 3080s /3090s (even now).

 

[brostradamus]

Anyway, performance improvement is fairly good at around 50-56%.

Again, 50% improvement (1% lows) comparing the RTX 3080 10GB to RTX 4080 16GB (60 FPS 1% low vs 90FPS 1% low).

It's the prices that are unaffordable, I am not at all happy with these.

Some clearly can afford the prices of both the RTX 3000 series and 4000 series, if people keep buying, prices will stay high.

Eventually, most people just won't have the cash to spend, and prices should come down...

i would look out for q4 financial results.. i am getting a sense that both companies have budgeted for lower SI sales volumes due to the economic situation, and they dont want to compete- eventually they just treat you as a statistic, maybe the laptop deals will be better because OEMs can dictate better terms ..perhaps that could satisfy the value equation

 

[CAT-THE-FIFTH]

have you ever wondered if 4090 is actually a new tier, even amd seem to have been caught off guard.. its 80 billion transistors
fundamentally these comparisons dont make a lot of sense, because engineering has to figure out the optimal tradeoff between perf, power, cost.. singling out individual parameters like die size, memory controller address size etc. doenst paint a good picture, what you should be instead talking abt is fps/watt or such kind of metrics when comparing graphic cards within a price tier
now the market being the way it is i am not sure if the higher price is just due to higher cost or lack of competition

It really isn't because its another 600MM2 die product which Nvidia has done for over a decade.You have had lots of similar types of transistor jumps before,especially when TSMC has failed and companies had to stay on one node for extended periods. It's just because Samsung 8NM was basically a 10NM class node and it looks big. Wait for the next Nvidia generation and see less of a transistor jump.

Remember Kepler? Nvidia went straight from 40NM to 28NM. ATI had problems,because they tried to use TSMC 32NM which failed so ended up causing Nvidia to win over the HD6970 because they simply went large.

Then you had the extended period on TSMC 28NM,because TSMC 20NM failed. So you had Nvidia go from TSMC 28NM to TSMC 14NM Finfet,which was a massive jump. Nvidia didn't even go 500MM2 for its top gaming dGPU and they made massive margins on Pascal.

The difference is Nvidia wasn't so greedy that generation,so the advantages of the massive jump in node technology was given back to the consumer even though prices did rise. Remember,the GTX1070 was a decent amount more in USD than the GTX970 but at least Nvidia met consumers halfway.

This time they took that advantage,and didn't care. Look at how they keep comparing the RTX4000 series with overpriced cards like the RTX3070TI and RTX3080TI? Or the RTX3090? These were poor value.

Also the sad reality AMD is in on the game. You saw it how I reacted negatively to how they jacked pricing up with Zen3 and how they tried it with Zen4. With the RX7900 series,at least they kept the maximum price the same as the prior generation,but the problem is the RX6900XT and RX6950XT were poorly priced at launched relative to the RX6800 series,and the RX7900XT is really the RX7800XT at best. At least,AMD reduced the price of the RX6900XT and other RDNA2 dGPUs below £650,but Nvidia didn't do that and people still keep buying the RTX3000 at dumb prices too.

Either way,if people want better pricing or at least more appropriate positioning they should not buy these products.

Yeah but pricing stayed similar now they suddenly 2x ??

He is trying to be edgy. He defended the high Zen4 pricing,but ultimately was proven wrong when AMD dropped all the prices relatively quickly!

 

[kalniel]

No that's mixed use registers. Amd has implemented 2 threads per CU. So you can have 2 separate instruction queues

Two Wave32 'instructions' for FMA yes, same as Nvidia implemented in Ampere.

 

[Hedge]

Each generation should be better than the previous, but each generation should occupy similar price points accounting for inflation.

It shouldn't be "this new one is better than the old one so it's only right for you to pay double"

I still can't get over digital foundry using a ps5 comparison against a 3k pc.

 

[brostradamus]

Two Wave32 'instructions' for FMA yes, same as Nvidia implemented in Ampere.

i believe wave32/64 is an internal amd only specification.. nvidia is a superscalar architecture do they need to batch/queue instructions like that?