Discussion Nvidia Blackwell in Q1-2025

[jpiniero]

Official pricing announced for 5060/Ti:

https://videocardz.com/newz/nvidia-announces-geforce-rtx-5060-ti-at-429-16gb-and-379-8gb-299-rtx-5060-launches-next-month

A more reasonable $50 to increase from 8GB to 16GB this time on the Ti.

That's assuming there will be product available at $429 tho. That will have to wait until tomorrow.

I am not a fan of the Fake MSRP.

 

[jpiniero]

Sure enough, Newegg has some listings... the 8 GB models are all $419 and 16 GB $472-479.

 

[blckgrffn]

I was thinking $499 street price on something with some ARGB, gamer aesthetic and a thermal pad - with the vast majority of stock being these "premium" ones.

So $480 is a nice little surprise

 

[DAPUNISHER]

I was thinking $499 street price on something with some ARGB, gamer aesthetic and a thermal pad - with the vast majority of stock being these "premium" ones.

So $480 is a nice little surprise

16GB for under $500 will sell through fast in the present market. Paying $400+ for 8GB is it's own punishment.

 

[Heartbreaker]

That's assuming there will be product available at $429 tho. That will have to wait until tomorrow.

I am not a fan of the Fake MSRP.

We live in a bad timeline, that isn't getting better any time soon.

 

[Ranulf]

Tim at Hub breaks down the 5060ti cards. Also, Nvidia is apparently burying reviews of 8gb cards. Launch day reviews will only have 16gb variants. "Board partners can't sample them (8gb cards). Tim was told that 3gb ram modules were still too rare and expensive.

Hah, the 5060 has no set release date in May and no review date embargo.

 

[Tup3x]

Those are definitely a bit cheaper than expected. These are with 25,5 % VAT included.

 

[SolidQ]

Here some results from J2C

Watch videoplayback1080 (online-video-cutter.com) | Streamable

Watch "videoplayback1080 (online-video-cutter.com)" on Streamable.

streamable.com

 

[Mopetar]

16GB for under $500 will sell through fast in the present market. Paying $400+ for 8GB is it's own punishment.

Even $380 for an 8 GB card is a joke. NVidia keeping those models out of reviewers' hands should tell you everything you need to know. Some poor fools are only going to look at the 16 GB reviews and get screwed badly.

NVidia honestly should not sell an 8 GB version of the card. I only hope some reviewers can get their hands on one fast enough to warn off unsuspecting consumers.

 

[jpiniero]

NVidia honestly should not sell an 8 GB version of the card.

And not sell a sub-$400 card? Don't think nVidia is willing to go there, just yet.

 

[SolidQ]

Those NV marketing is disaster for me

 

[jpiniero]

Those NV marketing is disaster for me

Shouldn't it be 16?

 

[Timorous]

I don't get why the 5060 is not just a 96bit 12GB card.

GDDR7 would mean it still has a bandwidth advantage over the 4060 and 12GB @ $300 for what is probably between the 3060Ti and 4060Ti in raw performance seems pretty fine to me.

 

[jpiniero]

I don't get why the 5060 is not just a 96bit 12GB card.

Cheaper since it has 2 less chips.

 

[igor_kavinski]

Was 3070ti 8 GB only? thought that was 3070 and Ti got more? Brain fog here.

The only two options with more than 8GB VRAM from RTX 3000 series were 3060 12GB and 3080 10GB. Everything else was either priced too high or not enough VRAM.

 

[blckgrffn]

I don't get why the 5060 is not just a 96bit 12GB card.

GDDR7 would mean it still has a bandwidth advantage over the 4060 and 12GB @ $300 for what is probably between the 3060Ti and 4060Ti in raw performance seems pretty fine to me.

Nvidia doesn't care that much. They'll sell containers of 5060s anyway.

12/16GB can be next gen's gift to gamers at that price point.

AMD will probably do the same but just closer to price points that are palatable.

OR they will skimp on encoder hardware that so many more people are using these days (live streaming on Discord, clip saving via GG or Steam or whatever) and that will equally cheese off the modern gamer.

It's fun to watch everyone clomp all over themselves rather make a competent $300 option.

"fun" to be clear.

 

[Tup3x]

Shouldn't it be 16?

No. That would be how long it takes to render single frame.

 

[jpiniero]

rather make a competent $300 option.

Cuz they can't at margins they are happy with. See nFlation.

 

[TESKATLIPOKA]

Absolutely TERRIBLE assumptions. Even games don't scale with more cores beyond a certain point. There is quite a bit of evidence that indicates that bandwidth is NOT an issue. Lack of VRAM can be. Poorly optimized games/game engines CAN be. You'd have to show me an actual test that shows bandwidth usage for me to believe that. I've worked extensively with OpenGL and DX3-11 along with some DX12 and Vulkan stuff, and I know what I saw writing code. All of my work was done in native C or C++ btw, no engines involved.

What you quoted is enough proof.

Despite having 42% more Cuda cores and much higher TDP than RTX 4060, It was has only 26% higher performance at 1440p(TPU).

You can't say that this mediocre increase in performance is because of bad scaling by having more SM, that's not true at this SM count.
I can back It up by comparing 4060Ti 16GB vs 4070 12GB -> 32% higher performance at 1440p while having 35% more SM. TPU
And Vram was not an issue, because at average 4060Ti 16GB vs 4060Ti 8GB saw only 1% difference in performance at that resolution.
So the only thing left is BW bottleneck and If you check the specs, It's only a bit better 4060: 24MB L2 + 128-bit 17Gbit vs 4060Ti: 32MB L2 + 128-bit 18Gbit.

NVIDIAs *90 series is faster, sure, but look at the power budget and increased VRAM.

RTX 5060Ti 16GB has the same amount of Vram and power budged did increase, but only by 15W vs 4060Ti 16GB.
And 9060XT 16GB likely will have very high TDP, If the boost of 3230MHz turns out to be true. Wouldn't be surprised If It was higher than 5060Ti's 180W.
But with that boost, It could be really close to 5060Ti at least in raster.

I'm willing to bet (admittedly with no evidence other than profiling code) that games would see far more improvements with faster clocks over wider cores. AMD seems to be proving that so far. We'll see what happens long term.

Honestly, I don't care. I wasn't even talking about which method provides better scaling. I was pointing out that 4060TI is BW starved, so 5060Ti could have additional gains by using GDDR7.

 

[TESKATLIPOKA]

RTX 5060 8GB for $299?
Performance should be ~ RTX 4060TI 8GB.
If they can make a refresh with 12GB GDDR7 and keep the price, that would be pretty good, especially for Laptops.

 

[blckgrffn]

There should have been one 5060 Ti SKU - 192 bit, GDDR6. 12GB. $399 ($449 for the fancy ones). "Change my mind" picture here.

What they are doing is silly and known (and cheap) GDDR6 would have solved their problem and not have had anything to do with new GDDR7, its cost or availability. Basically take what the 3060 has been for years and put a new core on it, if they had been purposeful AIB's could have reused a lot of their existing designs. Looks like board power limits between the 5060TI and 3060 are even really similar.

I know there are probably reasons they didn't do this - including money - but this all seems pretty dumb as it stands now.

 

[TESKATLIPOKA]

There should have been one 5060 Ti SKU - 192 bit, GDDR6. 12GB. $399 ($449 for the fancy ones). "Change my mind" picture here.

GB206 is limited to 128-bit wide memory controller.
It would have to be a heavily cutdown GB205 to have that bus width.

 

[blckgrffn]

GB206 is limited to 128-bit wide memory controller.
It would have to be a heavily cutdown GB205 to have that bus width.

How its designed now, yeah. And can it even work with GDDR6? I have no idea.

"I am just saying" that the 5060 Ti should have always been the top of the small chip stack and had a 192 bit bus and 12GB of ram. IF they were serious about the gamer GPU segment, maybe the would have gifted that to us.

But here we are, and nvidia is apparently so ashamed of the 8GB Ti that they are withholding it.

 

[TESKATLIPOKA]

How its designed now, yeah. And can it even work with GDDR6? I have no idea.

"I am just saying" that the 5060 Ti should have always been the top of the small chip stack and had a 192 bit bus and 12GB of ram. IF they were serious about the gamer GPU segment, maybe the would have gifted that to us.

But here we are, and nvidia is apparently so ashamed of the 8GB Ti that they are withholding it.

GDDR7 has 24Gbit modules, so with that you can have 12GB Vram. Maybe Nvidia thought those modules will be more widespread and cheaper, who knows.
They could have made such a chip, 200mm2 GTX 1060 had 192-bit GDDR5, but laptop makers wouldn't like a chip, which needs 6 memory chips for the low end.

 

[blckgrffn]

GDDR7 has 24Gbit modules, so with that you can have 12GB Vram. Maybe Nvidia thought those modules will be more widespread and cheaper, who knows.
They could have made such a chip, 200mm2 GTX 1060 had 192-bit GDDR5, but laptop makers wouldn't like a chip, which needs 6 memory chips for the low end.

Yeah, they could have still had their 5060 8GB though. nvidia is obviously no rush to add memory to laptop skus.

It's a hindsight is 20/20 sort of thing but it seems to me the lower risk approach would have been to go with in hand designs based on GDDR6. The 4060 series was actually the first in a bit that didn't have a 12GB option, the 3060 and 2060 12 GB (late addition) were both on the ground, and that's back in the good old days when the xx60 series was up to 256 bit

It would have made sense to migrate the 4070 memory subsystem - maybe with even less last level cache - down a tier to the 5060Ti. Again, IMO. Slow, trickle down progress.