Question nVidia 3060Ti Reviews Thread

[BFG10K]

NVIDIA GeForce RTX 3060 Ti Founders Edition Review

NVIDIA's GeForce RTX 3060 Ti comes at incredible pricing of $399, yet beats the GeForce RTX 2080 Super in rasterization, and even the RX 6800 in raytracing. In our RTX 3060 Ti review, we found that NVIDIA picked a fantastic cooler that's extremely quiet, yet fits everywhere because of its...

www.techpowerup.com

Nvidia GeForce RTX 3060 Ti im Test

Mit der GeForce RTX 3060 Ti bietet Nvidia für auf dem Papier 400 Euro die Leistung einer RTX 2080 Super. Erste Custom-Designs überzeugen.

www.computerbase.de

Performance is ok but I'm not really interested in yet another 8GB card.

 

[Tup3x]

Started queuing for Ventus 3X OC. At 499 € (inc. 24% VAT) it is rather nice package. More VRAM wouldn't hurt but considering my current playing needs it should be just fine.

 

[aleader]

You guys are all in USA or what?Because 400USD MSRP is BS outside USA.

I've 'seen' them at MSRP on Best Buy in Canada (with exchange), but everything is/was always out of stock, and you won't see those prices again until next fall.

 

[aleader]

Seriously, though... would you upgrade any piece of computer hardware for just a 20% performance boost? I feel like I'm wasting my time and money if my new part isn't at least twice as fast as the old one.

Judging by some of the ridiculous comments on youtube and other forums, there are a lot of people that will sell both their kidneys to go from a 2080 super to a 3080! Nuts. I'm with you, and I usually set my bar at double before I upgrade. In this case it was my 1070 at 1440p, and the 3060ti is just a hair less than double, the 3070 is a hair more. Having said that, I caved and bought a 5700XT so I can wait for Black Friday next year when the chaos is truly over That will still give me an almost 60% boost at 1440p...enough to turn MSAA on in DCS World.

 

[Mopetar]

That's really like saying they want to go from a 3060 Ti to a 3080. It's a fairly big step up (about an 80% increase in shaders), particularly if they want to game at 4K, which is where the 3080 starts to flex its muscles. The 3060 Ti can do 4K in many titles which is pretty impressive for a $400 card, but there are plenty of games where the average FPS is going to dip below 60 and depending on what the 1% lows look like the experience may not be the best. If they want to run games with ray tracing then Ampere is a big step up over Turing as well.

A 3080 is 35%/43%/56% faster (using TPUs numbers) than a 2080 SUPER in 1080p/1440p/4K gaming, so there's obviously a noticeable jump in performance from upgrading. For that you're only being asked to pay the same $700 the 2080 SUPER originally ran you. It's normal to expect that with each generation you get a bump in performance relative to the price previously paid. However, I don't know if it's normally a 56% increase. If you're looking at RT performance specifically, it's closer to a 75-80% jump.

Now part of why the numbers look so good is because Turing wasn't very good in the value/dollar department, but if they could snag a 3080, that 2080 SUPER can probably be sold for close to its original value right now just because of the shortages of everything else at the moment.

 

[papapapapapapapababy]

250w tdp. for a crappy gpu. how hard can you suck nvidia?

 

[tviceman]

Solid card at $400.

It seems to me that Ampere as an architecture has significant scaling issues.

The RTX 3070 has 21% more cores at a slightly higher clock speed, yet is only 14% faster in a best case (4k) scenario.
The RTX 3080 has 80% more cores at a slightly higher clock speed, yet can only muster 50% faster at 4k.
And the RTX 3090 with 116% more cores again running slightly faster can only pump out 64% faster frame rates at 4k.

Ouch.

 

[Bouowmx]

For the EVGA FTW3 Ultra, default power limit is 240 W, can be increased to 260 W. About middle of the pack of partner cards.


Ignoring the increased power, the following tests will assume 200 W reference limit.
Stock frequency is 1950-1980 MHz. I can undervolt to 937 mV to run at 80% power without decreasing frequency.


Going all the way down to the boost frequency specification (1665 MHz), I can undervolt to 768 mV to run at 55% power. Nice!

This configuration can be cooled with just 300 rpm fan speed.

This configuration gets 89% of stock performance in Unigine Valley.
Add a simple memory OC to get 92% of stock.
Although preset is Custom, the settings are the same as Extreme HD, except resolution is increased to 2560x1440.

Render:	Direct3D11
Mode:	2560x1440 8xAA fullscreen
Preset	Custom
Quality	Ultra

Configuration	Stock	Undervolt (1665 MHz)	Undervolt (1665 MHz) + Memory OC (2000 MHz)
FPS	91.0	80.9	84.0
Score	3808	3384	3516
Min FPS	42.8	41.8	40.3
Max FPS	165.7	143.9	147.8

 

[Tup3x]

Since there were truckloads of MSI Gaming X Trios available, I ended up getting one and cancelled the Ventus 3X. Should be cool and quite - perfect for overclocking.

 

[tviceman]

For the EVGA FTW3 Ultra, default power limit is 240 W, can be increased to 260 W. About middle of the pack of partner cards.
View attachment 34977

Ignoring the increased power, the following tests will assume 200 W reference limit.
Stock frequency is 1950-1980 MHz. I can undervolt to 937 mV to run at 80% power without decreasing frequency.
View attachment 34978

Going all the way down to the boost frequency specification (1665 MHz), I can undervolt to 768 mV to run at 55% power. Nice!
View attachment 34979
This configuration can be cooled with just 300 rpm fan speed.
View attachment 34980
This configuration gets 89% of stock performance in Unigine Valley.
Add a simple memory OC to get 92% of stock.
Although preset is Custom, the settings are the same as Extreme HD, except resolution is increased to 2560x1440.

Render:	Direct3D11
Mode:	2560x1440 8xAA fullscreen
Preset	Custom
Quality	Ultra

Configuration	Stock	Undervolt (1665 MHz)	Undervolt (1665 MHz) + Memory OC (2000 MHz)
FPS	91.0	80.9	84.0
Score	3808	3384	3516
Min FPS	42.8	41.8	40.3
Max FPS	165.7	143.9	147.8

80% power without a performance drop is great.

It's a shame we've gone from overclocking 15+% with Fermi & Maxwell to undervolting just to shave off heat.

 

[DJinPrime]

Solid card at $400.

It seems to me that Ampere as an architecture has significant scaling issues.

The RTX 3070 has 21% more cores at a slightly higher clock speed, yet is only 14% faster in a best case (4k) scenario.
The RTX 3080 has 80% more cores at a slightly higher clock speed, yet can only muster 50% faster at 4k.
And the RTX 3090 with 116% more cores again running slightly faster can only pump out 64% faster frame rates at 4k.

Ouch.

It's not that simple. And it's not even that other part of the GPU are not equally scaled up (like memory). The problem is how software work. Maybe it's because of the limitation of how we think (we don't do more than 1 task at a time very well), the way we currently program is still fairly serial. You do this, then do that, and then do this, etc. Even when we can parallelize things, do task 1, task 2, task 3 concurrently. Often, after those task you come back to a serial path. Worse, you can't continue until all 3 tasks are finished in some situations. And within each task, it's still a series of instructions. That's why it's very hard to use multiple cores CPU, we just don't know how. GPU task are much more parallelizable (run same shader code on each pixels, ray calculations, etc), so that's why you see thousands of simple graphic processing units. But still, we can't completely get away from serial tasks. That's why SLI was such a problem, you can't truly divide the tasks all the way through. Something have to divide the tasks, and something has to stitch everything back together at the end.

With Ampere, the number of cores gone up significantly, so tasks that are highly parallelized will be very fast. That's why you see a higher benefit with 4k. At lower resolution, the workload is much lower (fewer pixels to run shader on, etc), so the benefit of all those extra cores will not be as much. The serial part is constant, the parallel part finish much faster so, the benefit overall is less. AMD use fewer but faster cores, that's why they seem to perform better at lower resolution and lose at 4k. This is the situation with PS5 and Xbox X, even though they both have a relatively same rdna2 gpu, the PS5 perform as well (if not better) than the xbox with fewer but faster cores.

So, it's not that Ampere don't scale well. If the task is highly parallelizable, then it will scale. That's why games that's well coded and optimized scales really well, while others don't. It's the software that's not scaling. It's up to the software side to catch up, because the hardware side is hitting a limit on how much faster they can go per core. Going wider is the only way to go. It's highly likely that as time go by, you will see Ampere performing much better than Turing, and each level of Ampere cards will start to separate more. It's not that NV is doing something weird with their driver, it's because developers have learned how to do parallel programming better.

 

[tviceman]

It's not that simple.

But it is "that simple." With games currently available, Ampere is experiencing clear scaling issues. Turing, Pascal, Maxwell, and even Kepler all had noticeably better scaling with more cores and bigger chips.

If it's a matter of getting developers to code specifically to Ampere's strengths, then generally speaking that will never happen. And Ampere will thus continue to scale poorly.

 

[DJinPrime]

But it is "that simple." With games currently available, Ampere is experiencing clear scaling issues. Turing, Pascal, Maxwell, and even Kepler all had noticeably better scaling with more cores and bigger chips.

If it's a matter of getting developers to code specifically to Ampere's strengths, then generally speaking that will never happen. And Ampere will thus continue to scale poorly.

Something is not scalable if given a large set of work, increasing the number of processing units does not see a near linear improvement. So, if Ampere is not given a sufficient workload to fully utilize the increased cores, then you can't say the architecture don't scale. You can say that the architecture do not fit those particular scenarios. Maybe that's same to you, but it technically is not.
The software have to be scalable too. It's not specific to Ampere, developers will have to adjust. Cores can't just keep getting fast, there's a mhz wall. CPU/GPU will just keep getting bigger and even multiple units, the only way to take advantage of that is changing the code. This generation will have to tackle this, if games can't handle higher number of cores, then the Xbox will always perform worse than the ps5. I doubt MS will allow that to continue, games will be more scalable to number of cores.

 

[blckgrffn]

For graphics cards, 50% has always been my minimum
For processors, I'm happy with 20% upgrade ~ I'd still be using my overclocked i7-2600k and still waiting, if I was holding out on something twice as fast for my gaming use

Especially when your fancy new GPU is barely faster than your old one because your CPU/setup has it mired in quicksand. Or worse, the peaks are like what you read in reviews (yay!) but your low FPS is still stupidly low because there are times when the CPU just chokes (no!).

I'll be upgrading to a 5600x when I can just walk into a MC and buy one on any given day and waterfalling my 3600 to my wife. After all those years on a HEDT Intel platform, I am thinking staying on the bandwagon is worth it for now.

Tom's showed OC'ing a 3060ti basically lined it up with the 3070 in a lot of scenarios (non-RTX especially) which further dilutes the value of the 3070. If the 3080 was actually available at MSRP and you really want the bells and whistles, that still seems like the SKU to get imo.

 

[Mopetar]

Tom's showed OC'ing a 3060ti basically lined it up with the 3070 in a lot of scenarios (non-RTX especially) which further dilutes the value of the 3070. If the 3080 was actually available at MSRP and you really want the bells and whistles, that still seems like the SKU to get imo.

Unless you've got a 4K monitor I'd just get a 3060 Ti. The 3060 Ti maintains above 60 FPS in most titles at 1440p and a majority average above 90 FPS or well into the 100's. At 1080p the 3060 Ti performs even better than the 3080 is only ~30% better in terms of average FPS for almost twice the cost and isn't that much better at 1440p.

I don't think it really makes sense unless you're going to run at 4K, which is where the 3060 Ti starts seeing sub-60 FPS in games whereas the 3080 is doesn't see nearly as many games that will drop below 60 FPS and in most games even the 1% lows will still be at or above 60 FPS.

 

[VirtualLarry]

UFD Tech (Hot News) (See around 8:50)

61MH/sec, lower energy usage than RX 5700XT. Miners incoming!

Edit: Talking about the RTX 3060 ti, not the card in the title graphic.

 

[Guru]

Good performance at a solid price, too bad no one can buy this card, its nowhere to be found. Were they even up for like 1 minute before they sold out?

If you are strictly in the Nvidia camp and don't want to see AMD's competitor performance then this is an amazing upgrade from anything like 2060s and lower and older. I mean if you were holding on to a 1080 or lower or 2070 and lower then this upgrade will bring anywhere from 30% uplift compared to the 2070 and 100% uplift when looking at the 1070.

I will definitely be looking to upgrade this generation once availability is good and prices start to drop and reach msrp.

 

[Tup3x]

MSI had surprisingly good availability yesterday here in Finland so many managed to buy one (including me).

 

[Heartbreaker]

MSI had surprisingly good availability yesterday here in Finland so many managed to buy one (including me).

Similar in Canada from what I have read and followed of the retailers. Of the next generation cards so far 3060 Ti seemed to have the most stock. I watched the stock trickle down on launch day and it was also MSI that had the most stock here as well.

 

[Mopetar]

Could be that it's the GDDR6X supply that's holding Nvidia back as much or more than wafers.

 

[Tup3x]

Could be that it's the GDDR6X supply that's holding Nvidia back as much or more than wafers.

That's a non-issue for RTX 3060 Ti and RTX 3070.

 

[maddie]

Could be that it's the GDDR6X supply that's holding Nvidia back as much or more than wafers.

That's a non-issue for RTX 3060 Ti and RTX 3070.

The simplest explanation is that Nvidia is moving to Ti models for the top cards. Few thought that AMD would compete effectively and we know how competitive Nvidia can be. It would appear that the 3060Ti has a lot more availability than the 3070, even though using the same die and memory. Says a lot to me.

 

[Mopetar]

That's a non-issue for RTX 3060 Ti and RTX 3070.

I guess I didn't explicitly state it, but that was kind of my point because the supply for the 3060 Ti has been reported as being better. I realize that GA-104 is also considerably smaller (should produce roughly 140 dies per wafer as opposed to 80) than GA-102 and that Nvidia could just as easily have dedicated more wafers to producing GA-104 chips, but the second factor would be the sensible thing to do if you know there isn't enough GDDR6X for more 3080/90 cards to be produced and you'd just have extra chips setting around in a warehouse doing no one any good.

If they weren't contained, it makes more sense to devote more wafers to GA-102. If we assume that they have a ratio of 4:1 cut dies to full dies that makes the average retail value of a GA-102 die $860 or $1.37 per mm^2. The same ratio with GA-104 produces only $1.07 per mm^2. This does leave out added costs of GDDR6X as opposed to GDDR6 as assumes that there are not any dies that can't be used at all, though I'm not certain to what extent that affects one more than the other. However it seems unlikely to be enough to make the valuations completely even.

 

[CastleBravo]

The simplest explanation is that Nvidia is moving to Ti models for the top cards. Few thought that AMD would compete effectively and we know how competitive Nvidia can be. It would appear that the 3060Ti has a lot more availability than the 3070, even though using the same die and memory. Says a lot to me.

What I find odd is that NV could put all of those dies on 3070s and they would still sell out instantly. Competition from AMD should be irrelevant until supply is able to meet demand. Are they having yield problems?

 

[Heartbreaker]

What I find odd is that NV could put all of those dies on 3070s and they would still sell out instantly. Competition from AMD should be irrelevant until supply is able to meet demand. Are they having yield problems?

I really think the big difference was there was a lot more time spent building 3060 Ti inventory. There were a lot of 3060 Ti Warehouse/retailer inventory pictures before release than there was for 3080/3070.

So physical retailer supply held up better on launch day. But they are just as out of stock since then.

 

[Heartbreaker]

Could be that it's the GDDR6X supply that's holding Nvidia back as much or more than wafers.

Interesting. This says cards are being impacted by GDDR6 shortage, and GDDR6X is less impacted:

Radeon RX 6000, GeForce RTX 3070/3060Ti supply issues also caused by GDDR6 memory shortage - VideoCardz.com

French website Cowcotland, which is usually well oriented in the AIB’s supply channels, claims that the global GPU supply is also affected by a lack of GDDR6 memory. AMD Radeon RX 6000 and NVIDIA GeForce RTX 30 supply not to improve until February? According to the report, NVIDIA and AMD may...

videocardz.com