Nvidia GeForce RTX 50-series Blackwell reviews

[Albuquerque]

And to be fair to me, I was the one being quoted in this thread after I made my point about the settings used.

And again to be fair to him, more than one person with more than one perspective is asking him to retort. There was no need for you to declare he was moving goalposts, when you knew he was replying to someone else's question. Even if you weren't aware before, you are now, which means it wasn't he who was moving goalposts.

Yes, and the "it depends" question is answered with benchmark data.

Maybe, are we going to provide benchmarks of games relevant to the SKU being tested, or just those games and settings which we already know will not perform well on that part? As is always in these dicsussions, context actually matters. Handwaving away the context and simply aiming for black or white does none of us any good, because there are black or white answers to any of this.

 

[Flappy Pannus]

Or are we just going to provide games and settings which are obviously out of the realm of reasonability of this SKU

What games and settings specifically are "obviously out of the realm of reasonability" for a $400 GPU that they tested? They go down to 1080p and high/medium with most games.

 

[Subtlesnake]

And again to be fair to him, more than one person with more than one perspective is asking him to retort. There was no need for you to declare he was moving goalposts, when you knew he was replying to someone else's question. Even if you weren't aware before, you are now, which means it wasn't he who was moving goalposts.

He was replying directly to me and suggesting that HUB's data was cherry picked, when I can accept that and still hold that some or all of it is relevant to users. Notice that nowhere in this discussion did I make any claims about HUB's conclusions as a whole. l get that it can be difficult to follow different conversations and will refrain from such language in the future.

Maybe, are we going to provide benchmarks of games relevant to the SKU being tested, or just those games and settings which we already know will not perform well on that part? As is always in these dicsussions, context actually matters. Handwaving away the context and simply aiming for black or white does none of us any good, because there are black or white answers to any of this.

I think it would be most productive to have a substantive discussion about the data itself and what kinds of games should be present.

 

[trinibwoy]

$380 vs $430 is +13% and that's MSRPs, retail prices and model ranges will likely widen that gap a bit.

People need to look at retail prices for their individual purchases but we can’t use that for general comparisons. It’s just a $50 MSRP difference which nowadays isn’t a lot of money.

Steve's arguing that people can't see what they are buying and thus are buying 8GB cards because they didn't know that it's not enough for some games is completely dishonest.

I'd wager that at this point everyone knows that 8GB is the minimum amount of VRAM you should be buying and if you're looking at playing recent AAA SP titles then you should be looking at 12 at minimum.

Yeah this was just posturing . People who are blindly buying 8GB cards cause they saw a review of a $50 more expensive 16GB card are likely not in HUB’s audience so he’s mostly preaching to the choir. He claims to be looking out for the uninformed but it’s not clear who the victims are here.

 

[Flappy Pannus]

DF has a preview of upcoming material, and I like Rich's approach here. Doesn't sound like they'll be bothering with an 8GB review though.

Rich (ie me) almost feels like an RTX 5060 Ti review is perhaps pointless, but I will be taking my existing work and integrating it into a review based more on the anti-benchmark concept we've discussed in DF Direct Weekly. Yes you'll have your benchmarks and value calculations, but how about RTX 5060 Ti vs PS5 Pro? Results there are certainly intriguing. How about some gameplay tests on RTX 3060 Ti vs RTX 5060 Ti covering performance, frame-times and latency? I doubt benchmarks will tell you much different than the reviews already out there and you don't need me to tell you again that the 8GB version shouldn't be considered, so I'm going to see how the anti-benchmark concept works out. I'm pretty pleased with the results so far. AyaNeo 3? Still want to do it. And I want to track down a Strix Halo unit!

Also wanna see some Strix Halo reviews, albeit until their APU's get FSR4 it's hard to get excited about them replacing any low/midrange discrete GPU's (and Strix Halo is not cheap regardless.

 

[Albuquerque]

He was replying directly to me and suggesting that HUB's data was cherry picked,

The part you quoted and were complaining about, in response to his post, was him talking about year-old games and defending that stance. Since you weren't the one talking about year old games, that means he wasn't responding to your take. Thus, no goal posts were being moved, he was simply responding to someone else's claims.

 

[Subtlesnake]

The part you quoted and were complaining about, in response to his post, was him talking about year-old games and defending that stance. Since you weren't the one talking about year old games, that means he wasn't responding to your take. Thus, no goal posts were being moved, he was simply responding to someone else's claims.

I wasn't taking issue with the discussion of year old games. I was taking issue with that discussion leaking over into my exchange, and me having to respond to it. That's what I meant by changing the argument.

 

[T2098]

TPU has some great data for the discussion here:

NVIDIA GeForce RTX 5060 Ti 8 GB Review - So Many Compromises

We review NVIDIA's GeForce RTX 5060 Ti 8 GB, the card that NVIDIA doesn't want tested. Especially in VRAM-heavy titles the performance drops, but in many cases it's slightly faster than the more expensive 16 GB version, thanks to improved power efficiency.

www.techpowerup.com

The Gainward 5060Ti 8GB card they tested is interesting as it isn't overclocked, and comes with reference clocks and the reference 180W TDP limit and can't be changed. I guess not having to have the second set of GDDR7 modules eating power gives some significant extra TDP headroom for the boost algorithm. At 1080p the 8GB card is generally a wee bit faster, but in the VRAM-constrained situations where it loses, it loses big, and those losses start to pile up as the resolution increases, as you'd expect.

It's also super strange that the 4060Ti 8GB card beats the 5060Ti 8GB card in some of those same VRAM-hungry titles. Maybe Blackwell's memory management is significantly different than Ada's? Or certain Blackwell features 'reserve' a chunk of VRAM for internal use, like the hardware flip metering? Very odd. I'm willing to chalk this up to a bug for now, as a lot of the data makes no sense, like the 5060Ti winning at 1080p, losing hard at 1440p, but winning again at 4k, in the exact same title, with the exact same amount of VRAM.

The 5060Ti has more of everything compared to the 4060Ti: more cuda cores, more TMUs, more tensor cores, more RT cores, the same 48 ROPs, slightly higher clocks, PCIE 5.0 x8 vs PCIE 4.0 x8 which should help in VRAM constrained situations, the same 32MB of L2 cache, and most importantly, a whole truckload more memory bandwidth (~56% more!)

A thing to note for all the charts on the page though, is that these are all tested at native resolution and without RT, so the absolute best case scenario for a card that's heavily VRAM constrained.
Lots of common options that people would use with such a card will eat into the VRAM budget further, like:

- Enabling RT (hello, BVH storage costs)
- Enabling DLSS (which you probably realistically would always want to in a card of this class)
- Enabling DLSS FG (which also would probably be leaned on extremely heavily in a card of this class)
- Multitasking / streaming / video encoding / decoding while gaming

Spoiler: Screenshot

 

[homerdog]

TPU has some great data for the discussion here:

NVIDIA GeForce RTX 5060 Ti 8 GB Review - So Many Compromises

We review NVIDIA's GeForce RTX 5060 Ti 8 GB, the card that NVIDIA doesn't want tested. Especially in VRAM-heavy titles the performance drops, but in many cases it's slightly faster than the more expensive 16 GB version, thanks to improved power efficiency.

www.techpowerup.com

The Gainward 5060Ti 8GB card they tested is interesting as it isn't overclocked, and comes with reference clocks and the reference 180W TDP limit and can't be changed. I guess not having to have the second set of GDDR7 modules eating power gives some significant extra TDP headroom for the boost algorithm. At 1080p the 8GB card is generally a wee bit faster, but in the VRAM-constrained situations where it loses, it loses big, and those losses start to pile up as the resolution increases, as you'd expect.

It's also super strange that the 4060Ti 8GB card beats the 5060Ti 8GB card in some of those same VRAM-hungry titles. Maybe Blackwell's memory management is significantly different than Ada's? Or certain Blackwell features 'reserve' a chunk of VRAM for internal use, like the hardware flip metering? Very odd. I'm willing to chalk this up to a bug for now, as a lot of the data makes no sense, like the 5060Ti winning at 1080p, losing hard at 1440p, but winning again at 4k, in the exact same title, with the exact same amount of VRAM.

The 5060Ti has more of everything compared to the 4060Ti: more cuda cores, more TMUs, more tensor cores, more RT cores, the same 48 ROPs, slightly higher clocks, PCIE 5.0 x8 vs PCIE 4.0 x8 which should help in VRAM constrained situations, the same 32MB of L2 cache, and most importantly, a whole truckload more memory bandwidth (~56% more!)

A thing to note for all the charts on the page though, is that these are all tested at native resolution and without RT, so the absolute best case scenario for a card that's heavily VRAM constrained.
Lots of common options that people would use with such a card will eat into the VRAM budget further, like:

- Enabling RT (hello, BVH storage costs)
- Enabling DLSS (which you probably realistically would always want to in a card of this class)
- Enabling DLSS FG (which also would probably be leaned on extremely heavily in a card of this class)
- Multitasking / streaming / video encoding / decoding while gaming

Spoiler: Screenshot

View attachment 13456

I have to be careful with FG on my 4070 even at 1080p in some games. Settings that otherwise work fine cause massive stuttering for lack of VRAM with FG turned on. I can't imagine FG would be very useable with 4GB less memory.

They should market the 8GB cards for esports or something where VRAM doesn't matter. They are fine for that.

 

[DegustatoR]

If Nvidia made the 8GB model an Asian-specific SKU for cyber cafes I think nobody would really care.

Why do you care that it's not "Asian-specific"? Are you afraid of having choices?

This is like whack-a-mole. Every time I reply, the argument changes.

The argument doesn't change from my first post on this matter. You're just not paying attention to what I write and present the same argument in replies again and again - hey, that's a bit like what Steve is doing with his VRAM vendetta!

I entered this thread to point out that HUB were not using unplayable settings to "break" 8GB cards.

But they do which is precisely why they have the selection of games they are using. We are literally going in circles because you don't hear what I'm saying.

Now we're into a dispute about the exact proportion of "recent" titles you need to be representative, when almost all review outlets (especially DF) use a range of both recent and older titles.

The question is not about the proportion, the question is in selection of titles which they use.
Let me help you understand: if Steve would title his video "DOA for AAA games I've chosen to show you" then this would be fine and in line with what's this video is about.

But if you like you can ignore the non-recent games HUB was testing. I still think people are going to be interested in the performance of Star Wars Outlaws, Indiana Jones, Assassin's Creed Shadows, TLOU2, etc.

Why? Recent AAA titles haven't been running well on an 8GB GPU for years now.
What's the point of this testing or the revelation you're getting from it?
The card has a market - it is made specifically because that market exists. Using it in improper ways when there's a 16GB alternative for that is just shoehorning your arguments to your agenda.
And yes, I've said this several times already as well. I'm starting to think that the HUB indoctrination is fatal.

It's also super strange that the 4060Ti 8GB card beats the 5060Ti 8GB card in some of those same VRAM-hungry titles.

Not that strange. When you're VRAM limited to the point where you're essentially rendering through PCIE bus performance can be wildly different because of various interlocks which can appear in such scenario.
It's another one of these things which would actually be interesting to see in reviews and yet no one does them really - does a game hit the VRAM wall "gracefully" or not? What happens when it does? Etc. The outcomes could be anywhere on the spectrum between just lower performance and straight up crashing.

 

[Subtlesnake]

The argument doesn't change from my first post on this matter. You're just not paying attention to what I write and present the same argument in replies again and again - hey, that's a bit like what Steve is doing with his VRAM vendetta!

Because I came into this thread to deny that HUB was using unplayable settings. That's it. Nothing commits me to approve every single choice HUB makes.

But they do which is precisely why they have the selection of games they are using. We are literally going in circles because you don't hear what I'm saying.

I hear what you are saying but doesn't invalidate my point which is that HUB's data is valid for the selection of games they are choosing for testing.

The question is not about the proportion, the question is in selection of titles which they use.
Let me help you understand: if Steve would title his video "DOA for AAA games I've chosen to show you" then this would be fine and in line with what's this video is about.

Then it doesn't sound like we have a disagreement.

Why? Recent AAA titles haven't been running well on an 8GB GPU for years now.
What's the point of this testing or the revelation you're getting from it?

Precisely because users need to know what kind of sacrifices are involved in order to make an informed decision.

 

[DegustatoR]

Precisely because users need to know what kind of sacrifices are involved in order to make an informed decision.

But the problem is that the video doesn't help in making an actual informed decision. That's the whole issue with it.
The card is completely fine in a whole number of games which users could be playing exclusively or mostly.
And it can likely run most of these AAA titles without issues when they'd be properly setup to not exceed the VRAM limits.
The scale of 8 vs 16 GB lies between no difference at all to crashing on VRAM limits, as I've said.
There are several possible outcomes inside that scale which if mapped to games people play may in fact skew towards the "no difference" point more than "crashing".
Showing that would be helping the users to have an actually informed decision.
Selecting games which are well known to have high VRAM usage to showcase that 8GB isn't enough in them tells us only one part of what is needed to have an "informed decision".
Again, as I've said, a person who only play CS2 does not need to pay +$50 for a 16GB card. Showing this as well is what a good review of VRAM limitations would be.

 

[trinibwoy]

The card is completely fine in a whole number of games which users could be playing exclusively or mostly. And it can likely run most of these AAA titles without issues when they'd be properly setup to not exceed the VRAM limits.

Influencers aren’t trying to help people make informed decisions though. They’re pushing whatever narrative generates the most clicks and engagement. Nvidia is at fault too because their aggressive marketing breeds counter marketing to the opposite extreme. I sorely miss the days when GPU coverage skewed more mature and more technical. Those days are long gone.

 

[Subtlesnake]

Selecting games which are well known to have high VRAM usage to showcase that 8GB isn't enough in them tells us only one part of what is needed to have an "informed decision".

I agree but if I am considering the 8 GB 5060 Ti, I think knowing how the card performs at otherwise playable settings in a range of VRAM heavy titles (especially recent ones) is helpful. It's not enough for a complete picture, since I need to know how the titles I want to play perform. But it gives an idea of what I might be missing out on. Like, I pay an extra $50 and in these particular titles, these extra settings or resolutions become viable.

The data of a review can still be helpful, even if you disagree with the editorial conclusion or it doesn't apply to you.

 

[DegustatoR]

I agree but if I am considering the 8 GB 5060 Ti, I think knowing how the card performs at otherwise playable settings in a range of VRAM heavy titles (especially recent ones) is helpful. It's not enough for a complete picture, since I need to know how the titles I want to play perform. But it gives an idea of what I might be missing out on. Like, I pay an extra $50 and in these particular titles, these extra settings or resolutions become viable.

It shows that you may be having issues but it doesn't provide the data necessary for you to understand how widespread and often these issues would be, or how fatal they would be to the experience (i.e. can you solve them by lowering settings?).
Instead it claims that the card is "Instantly Obsolete" using a selection of less played games with a small set out of all possible settings configurations to prove that.
And even then Steve omits the results from even this limited set which doesn't fit his narrative - like where are the 1080p and 1440p results for A Plague Tale or 1080p for AC Shadows?
This isn't informing the users, this is holy crusade.

Nvidia GeForce RTX 5060 Ti 8GB Review

Nvidia tried to bury early reviews of the 8GB RTX 5060 Ti, but now the truth is out–and it's not pretty. With outdated specs and poor value,...

www.techspot.com

Text version for those interested.

 

[Rodéric]

Games are made for 8 GiB VRAM cards because that's what's out there, when 12 or 16 GiB VRAM cards will start to be common, they'll target that instead...

(We do check Steam hardware survey when making games you know...)

 

[Flappy Pannus]

Games are made for 8 GiB VRAM cards because that's what's out there, when 12 or 16 GiB VRAM cards will start to be common, they'll target that instead...

Seems like that ship has sailed, hence the video and resulting discussion about it for the last 5 pages.

 

[DegustatoR]

Seems like that ship has sailed, hence the video and resulting discussion about it for the last 5 pages.

That "ship" has sailed when MS decided on specs of XSS and Valve - on the RAM size for Deck. The latter is admittedly way less important.
There won't be a huge VRAM (or RAM) requirements change until next console h/w gen. Up until then 8GB GPUs will have their niche and will remain in minimal system requirements.

 

[homerdog]

That "ship" has sailed when MS decided on specs of XSS and Valve - on the RAM size for Deck. The latter is admittedly way less important.
There won't be a huge VRAM (or RAM) requirements change until next console h/w gen. Up until then 8GB GPUs will have their niche and will remain in minimal system requirements.

The rub to me is that the 5060Ti is way beyond a min spec GPU, but it has min spec VRAM. The stagnation of GDDR module capacities combined with the narrowing of bus widths (outside of the top end) leaves us in a weird spot. Ideally the 3GB modules come along soon and put all this to bed.

 

[DegustatoR]

The rub to me is that the 5060Ti is way beyond a min spec GPU, but it has min spec VRAM. The stagnation of GDDR module capacities combined with the narrowing of bus widths (outside of the top end) leaves us in a weird spot. Ideally the 3GB modules come along soon and put all this to bed.

You don't need more VRAM for more FPS.