Quote from: Glegster on June 15, 2022, 21:40:07Oddly i can get a 3060 under 450 usd... your price quotes for all the gpu's are a little high

I can get a 3060 for around $350 without VAT but that doesn't mean it's the same everywhere. And importing might not be attractive (taxes, tariffs, hassle, warranty, shipping costs).

Quote from: 3000seriesPriceAreTooMuch on June 13, 2022, 10:52:56The price of 3000 series at my country are still too high!!!!

At the cheapset
3060      700$
3060 ti   850$
3070     1147$
 
3080     1780$

3090     3300$

Oddly i can get a 3060 under 450 usd... your price quotes for all the gpu's are a little high

Quote from: RobertJasiek on June 15, 2022, 10:11:07Hence, there should be consumer GPUs for both demands: low TDP / moderate noise versus as fast as possible.

And we can assume there will be a 4060 or 4050 and perhaps even something lower if they really do inflate TDPs across the board. 3060 is already at 170 W (Ti at 200).

Enterprise sector is heading there because of liquid cooling. Once you make the jump, it makes sense to up the density to get the most out of the cooling hardware. If you can get the same performance with half the number of cold plates, tubing and joints, it's a no-brainer. Cutting floorspace is a nice bonus.

An enterprise can have a server room, for which noise and heat of an individual GPU count little but everything is measured in total for all GPUs together and average price per performance counts. An endconsumer with one GPU also wants to have fun while sitting next to it so noise and heat of the individual GPU matter. An unrepresentative query among endconsumers says that most prefer at most 200 or 300W but some 5% would buy the fastest consumer GPU despite 500~800W. Hence, there should be consumer GPUs for both demands: low TDP / moderate noise versus as fast as possible.

Quote from: RobertJasiek on June 14, 2022, 07:45:24I am at a loss to understand why apparently Nvidia chases AMD to let 4080 have ca. 420W and 4070 ca. 300W according to currently rumours instead of simply decreasing TDP and / or moderately increasing speed.

I don't know what's going on in the GeForce market. In the server market, there is demand for density. So, package power will go up. If you make a die larger (or you combine multiple dies), you will need more power. And of course, it's a valid strategy to pump power in in order to push frequency up. It's up to the market to decide whether that makes sense, where is the limit. If there are people who will accept such power draws (and we know there are some; remember quad SLI?), the market is there and it's valid that Nvidia might compete for it. The only question is whether it's big enough. Increasing your power budget is one way towards higher performance, if more performance is what you want. If your budget is fixed, then performance gains only come from efficiency gains. Sometimes, it's enough; other times, you can stall.

I don't think we should go there for gaming. It's just wasteful. Super-powerful processors are just an excuse to optimize less. But the enterprise sector is heading there.

For me, the electricity bill is first (maintaining 400W would also be too expensive) and at most 37 dB is second. I will undervolt and -clock for sure but the default TDP must be low enough to start from. Efficiency could enable both and at least 3080 speed.

From Samsung 8nm to TSMC 4nm is supposed to increase efficiency by roughly 1.6x. Even just porting 3080 to the smaller node could greatly increase speed at 320W, reduce 320W to roughly 200W or a enable a modest combination of both. I am at a loss to understand why apparently Nvidia chases AMD to let 4080 have ca. 420W and 4070 ca. 300W according to currently rumours instead of simply decreasing TDP and / or moderately increasing speed.

I would happily buy an RTX 3080 Super that would simply port 3080 to TSMC 4 nm, maintain 3080 speed and have roughly 200W. Maybe there will at least be a 4070 Laptop with close to 3080 Desktop speed at, say, 110W.

Quote from: RobertJasiek on June 13, 2022, 18:39:02Nvidia's massive TDP increments might spoil it again. 4070 with about 300W would still be ok but some rumours say 330 to even 400W - clearly too much.

What really matters is efficiency. If a big jump in energy demand is accompanied with corresponding jump in performance which you don't need, just buy something lower down the ladder. So, even a 500 W 4070 could be acceptable. It would just be too powerful for you. If push comes to shove, reduce boost, avoid overclocking, tune it for efficiency.

Quote from: Hardware Geek on June 13, 2022, 16:09:35I do hope that prices in the RTX 3000 series drop below MSRP and availability becomes much better. That's about the limit of what I'm willing to have for power draw. I expect both of the next generation to have much higher power draws. I'm just not interested in having a computer that uses that much power. I'll wait until the next couple of generations come out and upgrade again when I can get noticeable improvements in picture with the same power or less.

I wish I was in your position but I still need to buy my first modern dGPU-Computer while 3080 has the minimum speed good for my applications. The TDP of 3080 is at the upper limit I tolerate so 3080 around its launch would have been good for me. We know what happened then. Now, RTX 4000 is my next hope for my first modern dGPU but Nvidia's massive TDP increments might spoil it again. 4070 with about 300W would still be ok but some rumours say 330 to even 400W - clearly too much. Nvidia needs to understand that quite a few consumers have their maximum TDP limits and won't ever accept ovens. I appreciate CUDA and tensor cores but Nvidia seems to try hard to push people to AMD. RX 7700 might have a reasonable TDP.

I do hope that prices in the RTX 3000 series drop below MSRP and availability becomes much better. That's about the limit of what I'm willing to have for power draw. I expect both of the next generation to have much higher power draws. I'm just not interested in having a computer that uses that much power. I'll wait until the next couple of generations come out and upgrade again when I can get noticeable improvements in picture with the same power or less.

Wow, too bad, but which $ is that? USD? That would be terrible indeed.

The price of 3000 series at my country are still too high!!!!

At the cheapset
3060      700$
3060 ti   850$
3070     1147$
 
3080     1780$

3090     3300$

What exactly is to stop retailers from marking up 40-series? Why should it be any different? Prices have been dropping like a rock and we're hopefully out of the woods. But as long as supply is limited, 40-series can sell for prices even higher than 30-series, with no regard for MSRP. Especially at the top end where many people have more money than patience.

Great!  the delay will give both AMD and Intel a chance for increased market share, Don't buy 3000 series yet, wait for the 4000 series to be released (will cause massive price drop) + crypto crash will eventually make 3000 series very cheap

Now RTX 3000 models should be at 80 ~ 85% of their initial MSRPs and LHR with restricted functionality lower than that. In Germany, RTX 3080 10 GB LHR are at 121 ~ 133%. The relatively most expensive such models sell less than 1 per day per major retailer. If more RTX 3000 shall be sold, currently offered greed prices must be replaced by the should-be prices. Otherwise people will wait for RTX 4000 or RX 7000. I have waited for 23 months now and will wait another 8 months or more if Nvidia, AMD et al continue with their greed.

That logic works as long as they have higher margins on 30-series. At this point, everybody knows that 40-series is coming soon (as in this year). If you buy a 30-series anyway, you probably don't want to wait. The development of 40-series is already paid for. If they simply wanted to milk the market, they could launch 40-series at correspondingly higher prices alongside 30-series. Which could make sense if there is a market and they don't compete for manufacturing capacities. Only willingness of people to pay sets limits.