Nvidia GeForce RTX 3070 Ti Founders Edition

[Terbinator]

I wonder what the thought process is behind using GDDR6X for no apparent performance gain but a lot more wattage?

At a super stretch maybe it is easier to source 6X due to Nvidia being the only customer? Even then, why not have it clocked lower or take lower quality samples to keep the thermals in check and increase your inventory.

Would love to see a comparison of a 70TI with RAM clocked same as a 70.

[CAT-THE-FIFTH]

I wonder what the thought process is behind using GDDR6X for no apparent performance gain but a lot more wattage?

At a super stretch maybe it is easier to source 6X due to Nvidia being the only customer? Even then, why not have it clocked lower or take lower quality samples to keep the thermals in check and increase your inventory.

Would love to see a comparison of a 70TI with RAM clocked same as a 70.

Looking at the power consumption,the RTX3070TI is more a parts dump IMHO. I would like to see the voltages of the GPU and GDDR6X it uses compared to the RTX3070(GPU) and RTX3080(VRAM). It wouldn't surprise me it got the rejected GA104 GPUs which could not make it to mobile,which need more voltage,and the rejected GDDR6X modules which are not good enough for the other cards. Then Nvidia searched the parts bin and chucked on a cooler.

[kompukare]

I wonder what the thought process is behind using GDDR6X for no apparent performance gain but a lot more wattage?

At a super stretch maybe it is easier to source 6X due to Nvidia being the only customer? Even then, why not have it clocked lower or take lower quality samples to keep the thermals in check and increase your inventory.

Would love to see a comparison of a 70TI with RAM clocked same as a 70.

On some hardware monitors software you have package power and core power for CPU (because the CPU vendors expose this).

Wonder if there is any easy way to get similar figures for GPUs?

Maybe if VRAM and core are on their own VRMs and the VRM controller exposes this? Haven't looked, but I don't recall seening any breakdowns.

[CAT-THE-FIFTH]

On some hardware monitors software you have package power and core power for CPU (because the CPU vendors expose this).

Wonder if there is any easy way to get similar figures for GPUs?

Maybe if VRAM and core are on their own VRMs and the VRM controller exposes this? Haven't looked, but I don't recall seening any breakdowns.

TPU does a teardown of all the cards they get:
https://www.techpowerup.com/review/n...edition/3.html
https://www.techpowerup.com/review/n...edition/5.html

Igor'S Lab does them too but with more detail:
https://www.igorslab.de/en/nvidia-ge...ttle-hunger/3/
https://www.igorslab.de/en/nvidia-ge...uddly-small/2/

[Terbinator]

Looking at the power consumption,the RTX3070TI is more a parts dump IMHO. I would like to see the voltages of the GPU and GDDR6X it uses compared to the RTX3070(GPU) and RTX3080(VRAM). It wouldn't surprise me it got the rejected GA104 GPUs which could not make it to mobile,which need more voltage,and the rejected GDDR6X modules which are not good enough for the other cards. Then Nvidia searched the parts bin and chucked on a cooler.

Hm, good point. This gens GTX 465?

[kompukare]

TPU does a teardown of all the cards they get:
https://www.techpowerup.com/review/n...edition/3.html
https://www.techpowerup.com/review/n...edition/5.html

Igor'S Lab does them too but with more detail:
https://www.igorslab.de/en/nvidia-ge...ttle-hunger/3/
https://www.igorslab.de/en/nvidia-ge...uddly-small/2/

Yes, not sure if W1zzard would be up to it, but Igor might be a candidate to hook up equipment to try to measure what the GPU and VRAM draw in terms of power.

What I actually meant though, is whether any hardware monitoring tool can read the different planes like the package power or CPU. Like those digital PSUs with the USB connections and data-logging, having it in software would a lot easier if perhaps less accurate.

Looking at the power consumption,the RTX3070TI is more a parts dump IMHO. I would like to see the voltages of the GPU and GDDR6X it uses compared to the RTX3070(GPU) and RTX3080(VRAM). It wouldn't surprise me it got the rejected GA104 GPUs which could not make it to mobile,which need more voltage,and the rejected GDDR6X modules which are not good enough for the other cards. Then Nvidia searched the parts bin and chucked on a cooler.

If the quantities for the 3000s are actually quite high (despite of the shortages), they should have plenty of parts which didn't make earlier cuts. Nvidia being as stingy as they are is not always a bad thing as selling all the dregs prevents them from going to landfill (purposefully skimping VRAM on the other hand encourages landfill waste).

[CAT-THE-FIFTH]

Hm, good point. This gens GTX 465?

I have a feeling it is,and its not like Nvidia can't sell it easily!

Yes, not sure if W1zzard would be up to it, but Igor might be a candidate to hook up equipment to try to measure what the GPU and VRAM draw in terms of power.

What I actually meant though, is whether any hardware monitoring tool can read the different planes like the package power or CPU. Like those digital PSUs with the USB connections and data-logging, having it in software would a lot easier if perhaps less accurate.

Igor said he only saved 20W undervolting the GPU,and 20W would be saved using GDDR6. Hence,I think a lot of the extra power consumption might be the GPU part. The clockspeeds of the RTX3070 and RTX3070TI are very similar.

It also reminds me of the RX6700XT - power consumption also looked relatively high when compared to the RX6800. Even though I think a lot of it is down to the much higher clockspeeds,although I also suspect AMD was binning better examples for their laptops too.