RTX 4060 Ti could be Nvidia's first compact 40-series graphics card

[AaronK]

Why it matters: A new leak is suggesting Nvidia's RTX 4060 Ti will feature an exceptionally short PCB and a relatively low power consumption of 220W. This should allow the upcoming GeForce RTX 4060 Ti to use significantly smaller coolers compared to the RTX 4080 and 4090, and have the cards fit inside smaller computer chassis where clearance issues are a problem with Nvidia's flagship GPUs.

Hardware leaker @kopite7kimi, shared a tweet yesterday revealing specifications of Nvidia's GeForce RTX 4060 Ti. One of the highlights listed is the use of a short PCB on the 4060 Ti reference design, suggesting the GPU could use compact cooling solutions. If true, the 4060 Ti has a good chance of becoming Nvidia's first normal sized graphics card being able to fit in cases where the RTX 4080 and RTX 4090 cards cannot.

RTX 4060 Ti has a very short reference board. The PG190 still uses CEM5 connector.
AD106-350-A1
4352FP32
8G 18Gbps GDDR6
32M L2
220W

— kopite7kimi (@kopite7kimi) December 13, 2022

Kopite says the RTX 4060 Ti will use Nvidia's AD106-350-A1 GPU featuring 4352 cores paired to 32MB of L2 cache and 8GB of GDDR6 running at 18Gbps. Power consumption is targeting 220W, but the 4060 Ti will still utilize the 600W 16-pin power connector found in the RTX 4080 and RTX 4090.

Based on these specifications, the RTX 4060 Ti will probably target the same performance class as Nvidia's RTX 3070 or 3070 Ti if we had to guess. The 4060 Ti features less than half the amount of CUDA cores as Nvidia's RTX 4080 which will make it substantially slower. The 4060 Ti would feature the same 8GB capacity as the 3070 series cards to justify this claim, but without any firm knowledge about clock speeds or memory bus width, it's difficult to make accurate assumptions. So take this with a grain of salt.

The most exciting aspect of this leak is the 4060 Ti's short PCB length. Kopite's wording suggests the 4060 Ti's PCB will be exceptionally short - possibly even shorter than the RTX 4080s Founders Edition PCB. This will give Nvidia and its partners an opportunity to build not just regular sized graphics card models, but also ultra-compact Mini-ITX solutions as well.

There's also a chance Nvidia and its AIB partners will take advantage of the RTX 4060 Ti's short PCB to build more powerful cooling solutions instead, though we won't know for sure until the RTX 4060 Ti launches. As for a possible release date, the RTX 4060 could arrive as late as mid-2023, which would allow Nvidia to move out current RTX 30 series inventory in that price segment.

Permalink to story.

https://www.techspot.com/news/96968-rtx-4060-ti-could-nvidia-first-compact-40.html

 

[VitalyT]

B.t.w. DELL produced the smallest RTX 3090 on the market (not bigger than this 4060 Ti). But those were specifically for their Alien Ware desktops. And in all fairness, that card was the only good thing in those computers. But it just shows that nVidia and others oversize their lazy card design, which could have been much-much smaller.

 

[Skjorn]

No mention of bus width. I'll guess 64bit bus width because they're starting to get sad.

 

[neeyik]

No mention of bus width. I'll guess 64bit bus width because they're starting to get sad.

8GB of GDDR6 is either four lots of 2GB modules or eight lots of 1GB modules. GPU vendors pretty much only use Samsung or Micron for GDDR6, these days, and only Samsung offers 18 Gbps modules. They're 2GB in size, so that means the bus width would be 4 x 32 = 128 bits.

The RTX 3060 Ti has a global memory bus width of 256 bits, and while 128 bits sounds like a massive step backward, 32MB of L2 cache would go a long way to offset the narrower bus width (the 3060 Ti has 4MB). This is why AMD, Intel, and Nvidia have all gone down the 'massive L2 cache' route.

 

[takaozo]

So now compact cards are 2.5 slots with a pcb not longer than the socket?
If the 4070 has 192 bit bus this one has 128 bit? 4060 on 96 bit and 4050 on 64 bit?
This still leaves room for a 4030 32 bit gddr3

 

[pcnthuziast]

Nvidia sickens me. No ethics, no shame.

 

[Theinsanegamer]

8GB of GDDR6 is either four lots of 2GB modules or eight lots of 1GB modules. GPU vendors pretty much only use Samsung or Micron for GDDR6, these days, and only Samsung offers 18 Gbps modules. They're 2GB in size, so that means the bus width would be 4 x 32 = 128 bits.

The RTX 3060 Ti has a global memory bus width of 256 bits, and while 128 bits sounds like a massive step backward, 32MB of L2 cache would go a long way to offset the narrower bus width (the 3060 Ti has 4MB). This is why AMD, Intel, and Nvidia have all gone down the 'massive L2 cache' route.

But as infinity cache demonstrated, having high capacity cache only goes so far. You still need to be able to feed the GPU with those giant textures that fill up that 8GB framebuffer.

 

[human7]

Is the leaker credible, though? Have their past leaks proven accurate?

 

[Neatfeatguy]

If provided info about the DLSS 3.0 graphs from Nvidia is anything to go by....

4070Ti will land right around 3090 performance.....(just behind the 7900 XT) and probably still be priced at $900
4070 will probably land right around the 3080 12GB.....probably price around $750
4060Ti would be around the 3070/3070Ti....and probably price around $600.

Quick! Everyone! Go snatch up a 4060Ti for $600 for that 3070 performance level!

I don't have much faith in Nvidia's (heck, even AMD's) cards this generation. Too pricy for actual decent performance gains or too pricy for minor performance gains.

 

[neeyik]

But as infinity cache demonstrated, having high capacity cache only goes so far. You still need to be able to feed the GPU with those giant textures that fill up that 8GB framebuffer.

The 4080 has less VRAM, a narrower bus, and less bandwidth than the 3090, but has no problem beating that card. It also has fewer SMs but counters all of this with way higher clock speeds and a shed load of fast, low latency cache. Assuming the rumoured specs are correct, it’ll be the same for 4060 Ti.

 

[yRaz]

>makes "normal" sized graphics card
>calls it MTX

the 7000 series has a, well, series of issues, but that is far from the point. Stop with the pricing, naming and now, sizing

The 4060 will never be a midrange card and I'll cite the 3060, 2060 and 1650 as evidence of that

 

[Pensiveclown]

Lots of great info! Just missing one tiny little thing...

The price tag

 

[sorten]

220W draw makes me think this isn't the card to replace my 1660Ti. I might have to wait for next gen and more efficient designs.

 

[Julnor]

Problem is the price, if it's going to be $600+ it won't be enticing at all. The TDP is a little high but I suppose just at the edge of viability. From experience I'd say you'd usually want to be 200 and below, the laws of physics dictate that heat has to go somewhere, and most people prefer it doesn't become a space heater. Not sure why this isn't mentioned for newer high TDP cards. I had a GTX 470, that put me off high TDP cards.

 

[Hodor]

But as infinity cache demonstrated, having high capacity cache only goes so far. You still need to be able to feed the GPU with those giant textures that fill up that 8GB framebuffer.

Call me a conspiracy theorist, but recently I've developed suspicions whether Infinity Cache is really infinite.