Nvidia GeForce RTX 50-series Blackwell reviews

[Cappuccino]

Just about everything has liquid coolers now for some reason. Perhaps it's safer to ship a system with a little waterblock attacked than a massive HSF. I'm guessing if the system gets banged around, something bad could happen with a 1KG heatsink hanging off the motherboard.

You don't need a 1kg heatsink, the Peerless Assassin is 135g and cools a 7800X3D just fine. Ryzen 7 chips are incredibly efficient, you don't need watercooling until you get into the 7950X territory, or Intel's higher end offerings.

They do it because liquid cooling AIOs are probably cheap for them to buy in bulk and saying it has liquid cooling means they can sell it for 30% more. These little AIOs aren't even very good, you often get better performance off a good air cooler than a below average 240mm AIO.

 

[homerdog]

You don't need a 1kg heatsink, the Peerless Assassin is 135g and cools a 7800X3D just fine. Ryzen 7 chips are incredibly efficient, you don't need watercooling until you get into the 7950X territory, or Intel's higher end offerings.

They do it because liquid cooling AIOs are probably cheap for them to buy in bulk and saying it has liquid cooling means they can sell it for 30% more. These little AIOs aren't even very good, you often get better performance off a good air cooler than a below average 240mm AIO.

I love talking about this stuff, but we should probably move this to another thread

 

[Sega_Model_4]

NVIDIA restricts GeForce RTX 5060 press drivers, no reviews at launch?

https://videocardz.com/newz/nvidia-restricts-gefoce-rtx-5060-press-drivers-no-reviews-at-launch

 

[Nappe1]

As 5060ti 8GB was beaten by B850 even on 1080p in some cases (hardware unboxed) , I Wonder how bad this really is going be...

NVIDIA restricts GeForce RTX 5060 press drivers, no reviews at launch?

https://videocardz.com/newz/nvidia-restricts-gefoce-rtx-5060-press-drivers-no-reviews-at-launch

 

[troyan]

I think nVidia knows what the conclusion of these reviews will be. So they have no reason to let reviews published at release time. For me this is just evolution from the feedback they have gotten. And becaue these reviewers do not care about all features there is no incentive to send them something.

 

[techuse]

The features the card will often be too VRAM starved to use?

 

[Nappe1]

The features the card will often be too VRAM starved to use?

In some cases yes.
Hardware unboxed's video where they compare B580 to 5060 Ti 8GB shows that not only problems occur in obscure non-usable resolutions but in resolutions and frame rates that can be considered acceptable as well.
Also while B580 was found out to be CPU dependable, in this review, 5060 Ti 8GB uses almost double of CPU time compared to B580. Theory goes that swapping stuff in and out from VRAM causes CPU overhead raising.

And if this is the situation now, the future can't be much brighter.
Of course 5060 is much better balanced card, as it does not have as much as raw power sooo less gets wasted while GPU waits VRAM to be filled with right stuff.

It will be interesting to see 5060 vs B580 reviews when they eventually come up.

 

[troyan]

The features the card will often be too VRAM starved to use?

You can lower texture quality and use DLSS.

In the end these reviews are useless for everyone. Like companies not sending their lower tier products to reviewers (cars).

 

[Nappe1]

You can lower texture quality ...

Yep, you certainly can. But then you get shittier textures. As simple as that. You could do same in B580 as well, but you don't need to as there is enough VRAM.

 

[Dampf]

Nvidia really disappoints me this generation.

Even worse than 5060 8 GBs are 5070 laptops with 8 GB though. Much more expensive and you can't upgrade when inevitably the 8 GB VRAM buffer won't be enough anymore. Planned obsolescence at its best.

 

[techuse]

Doom runs quite poorly on Nvidia GPUs. Blackwell in particular.

 

[Scott_Arm]

Doom runs quite poorly on Nvidia GPUs. Blackwell in particular.

There’s almost no reason I can think of that 50 series wouldn’t have a noticeable advantage over 40 series. If that doesn’t change I’d be worried. I would hope for some improvement either through patches or drivers, otherwise there’d be something seriously wrong with the architecture.

 

[DegustatoR]

There’s almost no reason I can think of that 50 series wouldn’t have a noticeable advantage over 40 series. If that doesn’t change I’d be worried. I would hope for some improvement either through patches or drivers, otherwise there’d be something seriously wrong with the architecture.

There is a very obvious reason - 50 parts below 5080 tend to have less FLOPS than their 40 (Super) counterparts. They compensate for that with memory bandwidth and whatever SM changes Blackwell has.
Why this would affect DTDA to such a degree is a different question however. Generally when something isn't performing as expected it is a sign of the code not being optimized properly.

 

[troyan]

The 5080 is slower in 1080p. The only downgrade on Blackwell is the L2 cache with higher latency. So ID Tech seems to like fast L2 cache.

 

[DegustatoR]

The 5080 is slower in 1080p. The only downgrade on Blackwell is the L2 cache with higher latency. So ID Tech seems to like fast L2 cache.

1080p is almost certainly highly CPU limited on 5080 even on highest settings.
Also worth noting that Eternal has similar performance profile, so whatever that engine is doing just isn't running well on Blackwell.
Considering that you get 60+ fps on 3060 though this isn't much of a problem I'd say. The underperformance here is in 5-10% range, clearing that up wouldn't change the actual experience much.

I've went through all the benchmarks and it's not just Blackwell underperforming which creates this picture - it's also RDNA4 performing better than usually in comparison to RDNA3.
If you remove these cards and look at Ada vs RDNA3 results only (or Ampere vs RDNA2 even) then there's nothing out of the ordinary happening there, RDNA3 is sitting where one would expect it to be - if even a bit lower than one would expect.

This TPU graph actually show this rather well:

Note the comparative deficit is visibly higher on 50 parts and lower on both 9070s.

 

[troyan]

Its not CPU limited in 1080p on a 5090: https://gamegpu.com/action-/-fps-/-tps/doom-the-dark-ages-test-gpu-cpu

A 12600K is enough for the 5090. Its the software running on the GPU limiting the throughput. There is some serious serial limitation on Lovelace and Blackwell.

 

[DegustatoR]

Its not CPU limited in 1080p on a 5090

It is very obviously CPU limited even on a 4090:

Doom: The Dark Ages im Test - Prächtige Grafik und tadellose Performance

CPU-Benchmarks, Latenzen und Fazit

www.pcgameshardware.de

The difference between benchmarks is usually a clear sign of a varying bottleneck between different scenes.
Also the CPU limitation does not work like people seem to think it does when you're looking at AVERAGE fps.
The fact that the performance between GPUs is different doesn't mean that there are no CPU limited scenes during a testing sequence.

 

[troyan]

This makes it not CPU limited. Maybe some scenes but in a GPU limited scenerio the performance do not scale on nVidia.
Using Techpowerups' numbers the difference between a 5060TI and 5090 increases from 2.32x -> 2.61x -> 3.1x with the resolution.

Doom Eternal shows the same scaling with standard RT (less in 4K with 2.8x). Despite needing twice as long to render a frame in Dark Ages the difference between a 5090 and 5060TI is not increasing. So the limitation is a software stack which cant scale the workload over resolutions up to 1440p.

/edit: So, i was interessted in how much faster the 5090 is in comparisions to 2080TI/3070. Techpowerup has the 5090 2.66x faster than the 3070 and Gamegpu put the 5090 2.88x in front of the 2080TI in Doom TDA in 1080p. This is the exact same difference to Doom Eternal. So nVidia has invested 5x more transistors since Turing but only get half of that as a performance increase in a game released seven years after Turing.

Fun fact: The only performance metrics which have not increased by 5x is rasterizing performance and pixel fillrate. 5090 has only 2.75x/2.94x more than the 2080TI.

 

[digitalwanderer]

Is the 5060 out yet, did I miss the launch? I sort of expect the reviews to be a PR disaster for nVidia and don't want to miss all the web drama fun.

SIDENOTE: I freaking love my 2070 super so I'm not really trying to bash nVidia, just expecting this card to be a truly epic fail.

EDITED BITS: May 19th, still a ways away.

 

[trinibwoy]

Is the 5060 out yet, did I miss the launch? I sort of expect the reviews to be a PR disaster for nVidia and don't want to miss all the web drama fun.

The drama has already started. Reviewers are accusing Nvidia of being salty about poor coverage of the 50 series and withholding 5060 review cards/drivers until cards go on sale. Those same reviewers are trying their hardest to convince us that they're not salty about receiving the cold shoulder because they still make money from their youtube videos even if they come out a week late. They want us to believe that they're upset cause they really care about the hordes of people who will buy a 5060 on day one without reading a review first. I'm not sure who buys that nonsense but It's hilarious.