RTX 4060 Ti Geekbench entry confirms 4352 CUDA cores with 8GB of RAM, possibly for $399

[Daniel Sims]

Something to look forward to: As the rumored date for the GeForce RTX 4060 Ti's unveiling approaches, its appearance on Geekbench marks a typical sign of its imminent launch. The post confirms prior rumors and offers a vague sense of where the GPU sits among its predecessors. A trusted leaker also recently stated the likely final prices for the 8GB and 16GB variants.

A new Geekbench listing confirms that at least one GeForce RTX 4060 Ti variant, likely the first to see release includes only 8GB of VRAM. Moreover, known leaker MEGAsizeGPU agrees with prior rumors that the 8GB card will cost about $400.

The VRAM capacity is concerningly low in light of the performance of recent prominent PC games. Video memory bottlenecks have constrained 8GB GPUs like the RTX 3070 and the 10GB 3080. Furthermore, the amount presents a downgrade from the 12GB RTX 3060.

The Geekbench entry verifies specs suggested by a die shot and retail listing from last month. The multiprocessor count of 34 confirms the 4060 Ti will have 4,352 CUDA cores. Prior information states it has 32 MB of L2 cache with a 160W TGP rating.

Whoever submitted the GPU paired it with a 3GHz Intel Core i9-13900K and 32GB of DDR5 RAM running Windows 11. The overall CUDA score of 46170 sits just above typical scores for the 3060 Ti, but final drivers could widen the gap. Retail units should outperform the 3070 Ti.

MSRPï¼Â
4060Ti 8G 399
4060Ti 16G 499

– MEGAsizeGPU (@Zed__Wang) May 17, 2023

Recent claims suggest the 4060 Ti could launch next week on May 24, with AMD's direct competitor in the mid-range space, the RX 7600 XT, following a day later. That GPU is also said to have 8GB of VRAM but might retail for $350. Team Red could unveil the 7800 XT, 7700 XT, and 7600 XT at Computex late this month or early next month. The 7800 XT also recently made its Geekbench debut.

Another Tweet from MEGAsizeGPU last week said that Nvidia plans to assuage VRAM concerns by releasing a 16GB variant of the 4060 Ti sometime in July. The leaker's latest information indicates it will retail for $499. Meanwhile, the non-Ti version of the RTX 4060 could arrive in July with 8GB, based on the same GPU as its laptop counterpart.

Permalink to story.

https://www.techspot.com/news/98728-rtx-4060-ti-geekbench-entry-confirms-4352-cuda.html

 

[ScottSoapbox]

Disappointing.

 

[pcnthuziast]

I'd buy the 16gb version for 500 or thereabouts.

 

[tylerb1489]

I'd buy the 16gb version for 500 or thereabouts.

Getting a $100 discount on a 3070 ti with 16gb…. I guess it’s not bad.

 

[emmzo]

8GB versions will be DOA.

 

[Thatsdisgusting]

190mm2 chip for almost half-a-grand.

You're being screwed hard, people.

 

[hahahanoobs]

I hope 8GB and 16GB versions are coming. I wanna see the performance difference within the 40 series.

 

[nnguy2]

Don't worry 85 score from Techspot.

 

[Shadad]

The should have been 4050 = 4060
let it be n shelf.

 

[Nobina]

12GB should be the minimum for this class of GPU imo.

 

[takaozo]

12GB should be the minimum for this class of GPU imo.

You can't fit 12GB on a 128 bit bus. 2,4,8,16 only.

 

[KIAg1978]

I'd buy the 16gb version for 500 or thereabouts.

Are you crazy? Wait 4 months and get a 4070 at the price or even Ti version !

 

[mcnabney]

Yeah, I'm going to be buying my first console since the Atari 2600. PC gaming is going to remain a boutique market. Enjoy the zero effort console ports as the market contracts

 

[Thatsdisgusting]

You can't fit 12GB on a 128 bit bus. 2,4,8,16 only.

How they've fit 2gb on 660/660Ti?

 

[neeyik]

How they've fit 2gb on 660/660Ti?

All GDDR memory modules, regardless of the version, have 32-bit data buses so a card with a 128-bit memory bus has four 32-bit memory controllers, with each one hardwired to one or two GDDR modules.

A 128-bit card with 2 GB, across four modules, is using 512 MB modules; if it's a 4GB card, then it's using 1 GB modules. In the case of the 16 GB 4060 Ti, it's using 8 x 2GB modules - two per controller. GDDR6 and 6X are currently limited to 2 GB modules.

 

[Thatsdisgusting]

All GDDR memory modules, regardless of the version, have 32-bit data buses so a card with a 128-bit memory bus has four 32-bit memory controllers, with each one hardwired to one or two GDDR modules.

A 128-bit card with 2 GB, across four modules, is using 512 MB modules; if it's a 4GB card, then it's using 1 GB modules. In the case of the 16 GB 4060 Ti, it's using 8 x 2GB modules - two per controller. GDDR6 and 6X are currently limited to 2 GB modules.

AFAIR, MCs were 64bit up to Pascal, when they decided to improve pipeline "flexibility" by rising dispatch unit complexity, along with moving to 32bit controllers. Anyway, GK106 utilized 192bit bus, and it still had 8(4?) mem banks - as I get it, 64bit interface for four parts and 2x32bit for another four. Maybe it's a bad idea to combine 32bit and 16bit for top-notch products, but smth like AD108 under AD106 nametag wouldn't "mind" having uneven mem configuration simply for VRAM size

 

[neeyik]

AFAIR, MCs were 64bit up to Pascal, when they decided to improve pipeline "flexibility" by rising dispatch unit complexity, along with moving to 32bit controllers.

The memory controllers are classed as being 64-bit by Nvidia, and the management units themselves may well have processed 64-bit read/write requests, but the PHYs are always 32-bits. For example, the Maxwell-GM204 GeForce GTX 980 has 'four 64-bit MCs' according to Nvidia's whitepaper, but die images clearly show 8 repeated patterns for the PHYs around the perimeter of the die.

Anyway, GK106 utilized 192bit bus, and it still had 8(4?) mem banks - as I get it, 64bit interface for four parts and 2x32bit for another four.

This GTX 660, which uses the GK106, has 8 memory modules. The chip has 3 MCs in the block diagram, but again, the die shot shows six repeated patterns for the PHYs. Four of these are hard-wired to one module each and the remaining two get a pair of modules apiece.

 

[Freddie159]

8GB versions will be DOA.

Only for gamers, alot of other people, namely those who run Boinc of Folding will be happy with the 8gb version but happier with the 16gb version so they can run multiple tasks at one time.

 

[Julnor]

Only for gamers, alot of other people, namely those who run Boinc of Folding will be happy with the 8gb version but happier with the 16gb version so they can run multiple tasks at one time.

I was under the assumption those types of tasks are even more bandwidth intensive and resource heavy, hence why professionals would use the Titan level enterprise cards. I've no interest in them but I do know both AMD and Nvidia have a professional line of GPUs that is oriented for non-gaming tasks.