Nvidia pushes ray-traced gaming ahead with new GeForce RTX 3000 GPUs

[Maxiking]

I think TDP figures are ridiculously high. If these were AMD products, comments would be so different. Seems Nvidia was so worried about RDNA2 that they did some overkill?

No one cares about TDP if there is no competition.

AMD gets laughed at because their RX590 gpu and Vega were consuming more than 1080ti while delivering only the sub 1070gtx or 1080GTX performance. They were hot, loud and slow.

The final recommendation. Stop watching More law is dead and his "leaks". Architecture development takes years, this chip has been in development for years and only now manufactured.

 

[Lew Zealand]

As Steve mentioned in a recent Q&A video, gamers don't give a crap about TDP. Almost all, including myself, will happily pay for appropriate performance and TDP is the most minor of concerns. Will my PS power it? If yes, then we're done. Even 200W vs 300W doesn't matter.

These new cards seem to be coming out at reasonable prices (I expected ~$100 more per tier) and I'll probably be in line for a 3080, but I'll wait until Navi 2 just to compare.

 

[sorten]

I must admit that the 3070 looks like a good value for the performance, but maybe just because prices were so high last generation.

My 1660 Ti is holding up just fine, so I'm going to wait to read about the 3060 and AMD's RDNA2 lineup.

 

[Lionvibez]

Enjoy!

 

[neeyik]

Such an odd looking die layout, especially compared to Nvidia's previous chips:



Turing:


Pascal:

All of the L2 cache has been pushed to the edge of the chip

Edit: The specs for the chip are bonkers - each SM has 128 FP32 units (or 64 units with dual FP32 issue)! 10496 CUDA cores is, frankly, ridiculous.

So there are 82 SM units in total, but it looks like the full chip has 84 of them - 2 being disabled. There's scope for a 3090 Super then

Edit 2: I think they're dual issue FP32 units, as this would tie in with the leaks stating 5248 cores:

5248 unit x 2 FP32 instructions per clock = 10496 'cores'

Edit 3: 12 MiB of L2 cache, by the looks of things too.

Can't wait to read the whitepaper for this one!

 

[Julio Franco]

My 2080 now officially sucks

 

[Shadowboxer]

Not that impressive when you consider Nvidia is responsible for that outrageous $1,200 price tag to begin with (and let's be clear it was $1,300. You can't even get bargain 2080 Tis for $1,200). Just a heads up, the prices given here are likely MSRP and not actual prices you are going to see. Card typically start at founders edition pricing, which is $100 over MSRP. Not including the typical AIB markup you seen depending on the model.

Given that turing provided zero improvement in price to performance at launch (it was actually worse if you look at the 2080 ($800) vs the 1080 Ti ($700), finally having an improvement in that category is less impressive and more relieved the GPU market is actually moving forward again. What some PC elitists don't seem to understand is that the bottom half of the PC gaming market drives everything. Without those numbers games wouldn't be getting ported and companies wouldn't be willing to sink as much money into hardware and software for the platform.

Nvidia could cut its prices in half and you still wouldn’t be happy! Oh wait, they have just done that.

Yeah, I get it, you want a 2080ti for $250. Personally I’d like a brand new Rolex for £30!

Oh and we are well aware that the bottom half drives the industry, the bottom line. Not sure why you’re making this point. However, as tech enthusiasts we want to be as far away from that bottom line as possible. Nvidia just reduced the cost of that hugely.

 

[Shadowboxer]

The 3090 having 24GB of VRAM is actually tremendous value for anyone who needed that much VRAM. Before the 3090 you needed to buy a Quadro M6000 for more than double the money to get that size buffer. And that quadro is likely no where near as quick as the 3090 either.

 

[neeyik]

The 3090 having 24GB of VRAM is actually tremendous value for anyone who needed that much VRAM.

Well, it's made the Titan RTX redundant overnight. It's $1000 cheaper and has double the FP throughput - you'd be kicking yourself if you'd just bought a few for a compute/ML workstation.

 

[Shadowboxer]

Well, it's made the Titan RTX redundant overnight. It's $1000 cheaper and has double the FP throughput - you'd be kicking yourself if you'd just bought a few for a compute/ML workstation.

I just looked it up, I didn’t realise the Titan RTX already had 24GB. I stand corrected, you didn’t need an M6000.

Yeah a lot of bad purchases are showing up now. My mate purchased a 2080 super 2 weeks ago for £650 despite me telling him to wait. Also anyone who started the year with a £700 purchase of a Radeon VII would be crying. I’m hoping Zen 3 can be a big jump like this because I am holding off on buying a needed new CPU right now.

 

[Vulcanproject]

I must say, I am pleasantly surprised and dumbfounded. Who could've foreseen that nVidia would be delivering great performance (in the RTX 3070) at a reasonable price point ($500)?

Me. I'll spare everyone my ticker tape parade but it was pretty clear a long while ago this would be a step leap and had to be delivered at reasonable price points because of the next gen consoles.

3070 needed to be about as fast as 2080Ti and as predicted it' s basically thereabouts.

This should start to deliver proper Ray Tracing performance to the masses. Reality is the midrange cards will be the ones that sell in big numbers.

Based on this the inevitable future RTX3060 should be delivering useful 1440p performance, and probably 4K with DLSS for hopefully sub $400. Maybe 350.

 

[neeyik]

Also anyone who started the year with a £700 purchase of a Radeon VII would be crying

Depends on the intended use - it's FP64 throughput is still well worth that kind of money. Yes, the 3080 has far superior FP32, but it still trails in double precision.

 

[BadThad]

Yep, probably even lower ! Greed is the blood that nvidia runs on.

I'm sure if you had your own business with billions invested, you'd sell at cost to avoid fools calling you greedy. Right?

 

[Nobina]

2080 Ti owners on suicide watch.

 

[Zinixo]

So It looks like water-cooling the 3070 will be a challenge.The power connector is right where the water inputs/out usually run.

 

[Soaptrail]

Nvidia could cut its prices in half and you still wouldn’t be happy! Oh wait, they have just done that.

Yeah, I get it, you want a 2080ti for $250. Personally I’d like a brand new Rolex for £30!

Oh and we are well aware that the bottom half drives the industry, the bottom line. Not sure why you’re making this point. However, as tech enthusiasts we want to be as far away from that bottom line as possible. Nvidia just reduced the cost of that hugely.

I want a GPU for under $300 but no mention of a 3060 or when I might be able to buy one.

 

[Bamda]

"The same could be said about the new RTX 3070 that according to Nvidia, nearly matches the $1,200 RTX 2080 Ti in raw performance."

The most important line of the entire article. Finally a 2080 Ti for $500.

 

[Lionvibez]

The 3090 having 24GB of VRAM is actually tremendous value for anyone who needed that much VRAM. Before the 3090 you needed to buy a Quadro M6000 for more than double the money to get that size buffer. And that quadro is likely no where near as quick as the 3090 either.

24GB of Vram for a gaming card could have been 16GB with a cost reduction and no one would notice.

Anyone in the market for a Quardo will most likely have certain requirements that will not allow them to choose a desktop gpu.And yes the Geforce will be faster at gaming but no one in a professional environment will care about games. The Quardo will be faster in its intended market and that is by design by NV. They don't want desktop gpu's cutting into that very expensive to own Quardo market.

 

[neeyik]

24GB of Vram for a gaming card could have been 16GB with a cost reduction and no one would notice.

The way that VRAM quantity and memory controller counts are linked together prevents this from happening. The 3090, retaining its memory bus width, can only be 12 or 24 GB; to have any different would require a narrower bus (and thus reducing the bandwidth).

 

[Hunt77]

It's a publicly traded company. It has to show financial growth every 3 months in their quarterly reports or its stock will fall or remain stagnant.

Are you an Nvidia shareholder? I am and I've been very very happy with the returns from Nvidia's stock. It was $25 in 2016, and it's $550 today. It's a phenomenal company and one of the best managed in the entire technology industry. You should consider buying a few shares, Robinhood allows you to do it commission free and off your phone.

I don't feel sorry for the 3090's price or Nvidia's markup, this is targeted at people who have money, and a $300 price difference from the previous generation is nothing to these people. Nvidia knows this, and keeps testing the best price for the upper 5% of their video cards (the 2080tis, the 3090s etc). They sell like hotcakes so why lower the price of their best performing class of GPUs?

Most people are not concerned about the price of the high end item but what it will do to the product stack price all the way down the line. Sure that is great for investors but most of us are not investors and at the current stock price it is hard to buy enough to see much ROI. Which leaves most of us having to anti up for higher priced cards than we want. I guess that I can finally go to consoles if I must at least you get the GPU included in the price!

 

[neeyik]

Pulling the various specs together, here's how each of the new Ampere cards compares (purely in theoretical values) against the former 'bests':

 

[Jroc187]

It wasn't very clear from the stream. Is that 2x with or without ray-tracing turned on? I'm going to assume they are comparing with it on because the 2080ti has about 26TFLOPs and the 3080 has 30TFLOPs.
Hopefully the faster memory of the 3080 and other optimizations helps it in games without raytracing turned on to push past 20% better FPS. Compared to the 2080, it does seem to have 50% more TFLOPs which is good.

it is close to 2x the performance in some games that were tested that don't use raytracing, acording to digital foundry. this card is the real deal

 

[ZackL04]

Where do I pre order a 3070?

 

[kiwigraeme]

As a middle of the road consumer - I think Nvidia stabilizing their prices is good thing - someone buys a RX3060 after awhile for a $50 discount and a couple of AAA games thrown in wouldn't be too unhappy. They could of gone all Apple ( however Nvidia need to sell to every one )
+ our resident whale is happy to pay $2000 for his 3090max or whatever

 

[Lounds]

I hope AMD undercuts them by quite a lot. RX5700 was such a tiny die, if they have that with Ray tracing for £250 it could be the new ATI 4000 series moment.