3DMark Benchmarks for GeForce RTX 3070 surface

Seems like my RTX2080ti (TU102, not TU102A) is about 20% faster in Port Royale, running about 46fps average.

 

I don't doubt this is legit. Look at the 3090 compared to the 3080. As the cores scale up the performance is not scaling lineally. Not sure why but it looks like something like a 3070 Ti with cores in between the 3070 and 3080 may be the sweet spot.

 

It will be interesting to see. AMD's rumored 80 CU card should have enough extra cores for most workloads. It might be like Nvidia's first generation where the lower end cards don't fare that well at RT but the highest end sku is pretty decent.

 

Faster than nVidia and at lower power, from what I'm hearing...

 

First time in a long time we've seen some possibly decent competition.

Was really looking forward to the 3070, releases on my birthday even, but I'm in no rush, really want to see how AMD can compete. Even if it performs beneath the top of last gen but still priced competitively.

 

Yeah, I went with best case when I said 30%, as I calculated a few of them and realised the TimeSpy was near the top of the pile at 28% - I rounded up to 30%! Given that this is a rumour/leak, and given 3080 has 50% more cores I also felt 30% might be more realistic for a performance difference vs the lower figures. We'll see, maybe it will only be 20-30% faster.

I really don't have a problem with my pre-order, I got in before the final price rises as I knew it was only gonna go up in price, and I knew it was the card I wanted as AMD is not an option with my GSync monitor....it's right at the highest limit that I would spend on a graphics card though, so I'm not so pleased about that. I also chose the Asus TUF version (non OC), which turned out to be a wise choice as it's one of the best ones and certainly for performance/build/value/noise/temperature, was influenced by the Guru3d review (yep I know it was a review of the OC version).

With previous generations, like Kepler for instance, I think I remember doing some performance calculations on core count and it scaled almost linearly for them with increased core count, as long as there's no power limitations....I figured it would be the same with this Ampere generation as long as the 3080 wasn't power limited. At launch though we see the 3080 is power limited in a lot of scenarios, but vBIOS with "unlimited power" didn't net much of a performance increase for the 3080, so I don't think the 3080 is significantly power limited. Is 3090 more power limited than the 3080, maybe that's why 3090 performance didn't scale linearly with increased cores vs 3080. Although thinking about it, it was a bit of a red herring that the 3080 has 50% more cores than the 3070 yet didn't cost 50% more than the 3070....it could have been a bit too good to be true to have the 3080 being better performance value than the 3070 - that was one of the main reasons I chose 3080 over 3070, but hell...even if it only turns out to be 20-30% faster than the 3070 I'll still take it as my decision as I won't be upgrading this card for a very long time. And I'm certainly very happy with the performance level of the 3080 in absolute terms, that doesn't change.

 

The image is real, if the results are real who knows but I doubt Igor is going to be uploading stuff without it being atleast semi accurate and from other leakers on twitter seems to hold true

 

The timespy numbers are suspect. The 3080 is significantly faster than a 2080ti in timespy but his graph makes it look like 10%.

 

The performance leaks seem to be inline with with the architecture I suppose, by really going for that rasterized performance crown they'll take a larger RT performance hit.

 

Ah. Graphics scores. Thanks. I missed that (clearly)

 

I hear you I wasn't trying to say the 3080 or 3090 are bad cards but there is certainly a bottleneck as the performance isn't scaling up with cores. This does leave a little more wiggle room for AMD to be competitive at the high end that is if AMD's cards scale up with cores and frequency. I'm happy Nvidia didn't completely knock it out of the park because we all really need AMD competitive again. I very much want to see them leapfrogging each other every generation.

 

We're not gonna know for certain until the launch of AMD about whether NVidia knocked it out the park, but unlikely they have. But, but, but, NVidia did kinda knock it out the park in terms of performance improvement over Turing for the same money, due to the large percentage increase in performance over the prior cards, eg comparing 3080 vs 2080, so that's still a knock out which is why people went mad for this generation of NVidia cards.....but yeah seems like AMD are gonna be on par or pretty close to what NVidia has presented with this 3000 series generation.

 

the 3070 is the marker of nvidia's success or failure with Ampere. the uArch is elegant, and if i may add the availability of the 3070 is tied to the yield rate of the full fat Ampere. i suspect there is already plenty of (poorly) binned Ampere gpus waiting for a 3070 designation given the supply issues w/ 3090/80.
in any case, this was always going to be expensive and nvidia thought it could deal with the demand (tho' this maybe too far in creating it?).

 

Interesting.. my 2080Ti is doing 42fps in Port Royal test and 44,5fps after OC.
NVidia said 3070 is slightly faster than 2080Ti but will see.

 

There is no relation what so ever between GA100 and GA104.

 

Neat.
So is more than one shipment of the 3070s planned or are they also going to be available exclusively via scalpers on eBay like the 3080s and 3090s?

Inb4 "iT's nOt HArD tO gEt ONe yOu JusT hAvE tO kEeP hiTtInG f5 24/7 tiLl u SeE iT"

 

Yeah' I'm not picking a side I just don't hold either near and dear to my heart. I'm not extremely impressed with Nvidia this round but it is considerably better than the 2xxx generation which was terrible if you factored in cost. The GTX 980 Ti was 75% more performance per dollar over the GTX 780, the GTX 1080 Ti was 77% more performance per dollar over the GTX 980 Ti, the 2080 Ti was a -32% performance per dollar compared to the GTX 1080 Ti, and finally the 3090 is 21% more performance per dollar over the 2080 Ti. So its better just we are all paying a very large premium due to AMD's lack of competition compared to the GTX 980 Ti and GTX 1080 Ti days.

 

soo the entire GPU can work as RT hardware? That might be a better approach in the long run, lots of flexibility on how to dynamically allocate resources, unless nvidia's dedicated RT core is faster than AMD's entire chip for RT tasks

 

Benchmarks posted on the forum of big navi show it trading blows with 3080, but with RT enabled it trades with 2080ti, beating it by a small margin.

 

If ever...

It is clear for me from all the reviews, benchmarks, etc that these Ampere cards are so damn fast that you can't even catch them in stores.