Nvidia recently scored a big win with its RTX 4070 Ti graphics card, which has me pretty optimistic about the future of Nvidia’s midrange offerings, but a new spec leak from a pretty credible Twitter leaker has me seriously questioning what the company is thinking.
For starters, it’s important to qualify spec leaks like this one, as Nvidia hasn’t announced anything yet, and any “leaks” online need to be considered with caution. But kopite7kimi (opens in new tab) it’s been pretty accurate in the past, so those specs can’t be entirely ruled out either.
RTX 4060 still uses PG190. AD107-400-A13072FP328G GDDR6 18Gbps115W24M L2I will try to remain neutral about any leaks in the future. 😁😁😁February 13, 2023
For starters, the RTX 4060 looks like it will use the AD107 GPU, which is a step down from the AD106 we would expect to see in the RTX 4060, as the RTX 3060 used the GA106 GPU. Even the RTX 3050 8GB used a scaled-down GA106 GPU, so the AD107 GPU appears to be a regression here.
What’s more, the alleged specs of the RTX 4060 (as well as some of our calculations on the back of the napkin) are nearly identical to the specs of the RTX 4060 Mobile for which we have official numbers for comparison.
|Component||RTX 4060 Mobile (Official)||RTX 4060 Desktop (rumored)||RTX 3060 Desktop|
|Ray Tracing Cores||24||24||28|
|base clock frequency||1545MHz||TBD||1,320MHz|
|Increase clock frequency||1,890MHz||TBD||1777MHz|
|VRAM||8GB GDDR6||8GB GDDR6||12GB GDDR6|
|memory clock||2000 MHz||2250MHz||1875MHz|
|Effective memory speed||16 Gbps||18 Gbps||15 Gbps|
|memory bus width||128 bit||128 bit||192 bits|
|memory bandwidth||256GB/s||288GB/s||360 GB/s|
As you can see, the biggest difference in the leaked specs of the confirmed RTX 4060 Mobile is the big downgrade in the number of streaming multiprocessors coming off the CUDA core count, from 3,584 on the RTX 3060 and 3,072 on the RTX 4060 – a 14.28% decrease . Without the dip, this is light years short of the best graphics card in the Nvidia lineup, the RTX 4090, and it’s only on the northern half of the RTX 4070 Ti, so the dip in core count here is going to limit the RTX potential of the 4060.
There’s also the matter of the slightly higher effective memory clock over the RTX 4060 Mobile, which gives the desktop RTX 4060 a 12.5% boost in memory bandwidth, but is otherwise more or less the same. even on the memory front than its mobile variant.
One’s guess right now is what the final base and boost clock speeds will be for the RTX 4060, but given that the Nvidia Lovelace architecture is a 4nm process versus Nvidia Ampere’s 8nm process, we expect base clock speeds to north of 2000MHz, with boost clock possibly above 2500MHz if base clock roughly 56% faster and boost clock gen-on-gen standard 48% faster for RTX 4090 and RTX 4080 hold on to RTX 4060.
Two reasons why these specs worry me
First, let me start with this: the decrease in streaming multiprocessors and CUDA cores is no big deal. At 128 CUDA cores per SM, the leaked specs mean a decrease from 28 SMs on the RTX 3060 to 24 SMs on the RTX 4060. A 14.28% drop in tensor cores and ray tracing cores isn’t huge, but consider that too that they are fourth generation tensor cores and third generation ray tracing. They are simply much, much better than the Ampere’s 3rd and 2nd generation cores respectively, so they will actually perform better despite there being fewer of them.
What concerns me, however, is the memory. 8GB VRAM at this stage is pretty insignificant, and meanwhile technically should be a top-end 1080p graphics card, considering how well the RTX 4070 Ti handles 4K and how well the RTX 3060 Ti handles gaming at 1440p, we expect the RTX 4060 to be a solid contender for the best 1440p graphics card . Given its specs, though, I’m dubious.
The problem here is that 8GB of VRAM is fine for 1080p, as the size of texture files that can quickly fill the VRAM is much smaller than it is at 1440p or 4K. And while the RTX 3060 Ti also had 8GB GDDR6 VRAM, it also had a much wider memory bus (256-bit), providing an effective memory bandwidth of 448.0 GB/s.
This is more than enough to get 1440p textures loaded and processed efficiently despite the smaller VRAM pool, whereas the RTX 3060 had 50% more VRAM (12GB) and a larger memory bus (192bit). /s, also giving it decent 1440p performance. The RTX 4060, meanwhile, appears to have much lower memory bandwidth, so larger textures at 1440p will bottleneck much faster, limiting 1440p performance and pretty much restricting this card to 1080p gaming.
There’s also the issue of using the same base GPU as the RTX 4060 Mobile. We haven’t been able to test the RTX 4060 Mobile for ourselves yet, but typically mobile chips are about a level to a level and a half lower than their desktop counterparts in terms of GPU used (so the RTX 4090 the desktop’s AD102 is a notch above the RTX 4090 Mobile’s AD103), as well as being scaled-down versions of those chips.
The RTX 4060 Mobile is definitely a scaled-down variant of the AD107, but if the RTX 4060 desktop card is also an AD107, then you definitely can’t expect much more performance from the desktop version than you would get with the mobile GPU. That’s not great for desktop gaming, even at 1080p.
Still, there’s a big reason why these specs could be good news for gamers.
While this could all be bad news for gamers in terms of performance, there’s a sliver of hope here, and that’s the price. If Nvidia tried to make an aggressively priced RTX 4060 (I’m thinking under $300 / £300), then the trade-off here would not only be justified, it may very well be what gamers are asking for.
Most gamers still play at 1080p, according to Steam Hardware Survey (opens in new tab) with an increasing number upgrading to 1440p max. While memory restrictions don’t bode well for 1440p gaming, getting an affordable graphics card into gamers’ hands where they currently are would definitely be something to celebrate.
Many gamers are still running GTX-era GPUs, so the performance upgrade with an RTX 4060 will still be substantial enough that many gamers won’t care how much better it is. it could be with an AD106 GPU, and in the end, that’s really all that matters – especially if Nvidia can get this card’s pricing right.