Launching Today: NVIDIA’s GeForce RTX 3060 - Aiming For Mainstream At $329

[harlan4096]

NVIDIA this morning is launching their previously announced GeForce RTX 3060. First unveiled back at CES 2021, the latest member of the GeForce RTX 30 series is continuing NVIDIA’s ongoing top-to-bottom launch of Ampere-based video cards, with today’s card in some respects being the most popular one.

Aimed at the mainstream market, the RTX 3060 is designed to be a more balanced option for the larger market of gamers who probably aren’t trying to drive high-end 4K displays, but still want the latest graphical features on a 1080p or 1440p display. RTX 3060 cards will go on sale a bit later this morning – at 9am Pacific – with prices starting at $329.

Underpinning the new cards is the first new GPU out of NVIDIA since last fall, GA106. This part has already been shipping in laptops for just under a month as part of the GeForce RTX 3060 Laptop GPU, and now it’s finally coming to desktops. We’ll dive into GA106 more in a bit, but at a high-level it’s a fairly traditional scaled-down NVIDIA GPU design, cutting down on the number of SMs and memory channels to produce a cheaper GPU. Like it’s other consumer siblings, this is being produced on Samsung’s 8nm process – a potential blessing at a time when TSMC is slammed with orders, but not completely meeting NVIDIA’s needs amidst the insatiable demand for chips across the industry.

As previously mentioned, NVIDIA’s 60-tier parts are traditionally their big volume parts for North America and Europe, and the RTX 3060 will be no exception.

Though the $329 price tag is arguably on the north side of “mainstream” – and there’s the whole matter of unmet demand – there’s little question that as the cheapest Ampere card yet, it’s going to be the most popular one with gamers. NVIDIA sold nothing short of a massive amount of cards to this mainstream market 3-4 years ago with their Pascal-based GTX 1060 series, and now they’d like to sell that market on an upgrade.

That said, the RTX 3060 comes at a very odd time for the video card industry, enough so that, volume aside, this may be the most important RTX 30 series video card launch yet. Specifically, this is the first GeForce video card to launch since the video card market went cuckoo bananas (to use the technical term) thanks to a spike in the price of the Ethereum cryptocurrency. The compounding demand has made it nearly impossible to get a video card of any sort over the last few months, and has led to NVIDIA taking the unprecedented step of adding an Ethereum throttling mechanism to the RTX 3060 cards. Only time will tell how effective this measure actually is, but addressing the miner problem is the first step towards ensuring a more regular and predictable supply of video cards for NVIDIA’s core gaming market.

But even with those mechanisms in place, pent-up demand for video cards in general means that the supply of cards for today’s launch will go quickly. For those who are fortunate enough to get a card, they will have a significant number of models to pick from, as the absence of a retail NVIDIA reference card means that AIBs’ custom designs making up the entirety of the RTX 3060 lineup. None of the usual suspects are releasing anything too exotic here – at least not right away – so expect to see a mix of dual and triple fan cards at a variety of price points.

Moving on, let’s dive into the specifications for NVIDIA’s latest Ampere card. As previously mentioned, this is based on NVIDIA’s new GA106 GPU. At 12 billion transistors, the chip is notably lighter than the 17.4B chip that is GA104, and at a high level it’s roughly equivalent to two-thirds of a GA104 both in the number of functional units and in the transistor count. Out of the fab, GA106 comes with 30 SMs, 48 ROPs, and 6 memory controllers. However for yield reasons, NVIDIA is shipping their leading GA106 card with only 28 of those SMs enabled. So as far as GA106 is concerned, there’s a little more gas in the tank than what we’ll see from the RTX 3060.With its 28 SMs enabled, RTX 3060 offers a total of 3584 FP32 CUDA cores. Meanwhile everything else in GA106’s SMs is in proportion per the Ampere architecture, so this means there is also 112 TMUs, 112 tensor cores, and 28 RT cores. The ROP count, on the other hand, stands out a bit more here since NVIDIA has significantly trimmed down this aspect of the chip; whereas GA104 had 96 ROPs, GA106 gets just 48. This means that, relative to GA104 and the RTX 3070/3060Ti cards, RTX 3060 is going to be a bit more hamstrung in terms of total pixel pushing power and thus higher resolutions, underscoring how this is a card aimed first and foremost at 1080p gaming.

Given that NVIDIA had decoupled the ROPs from the memory controllers in this generation, I had been hoping we’d see 64 ROPs for the RTX 3060. But as it turns out, that won’t be the case. NVIDIA has not increased the number of ROPs for a 60-tier card since the launch of the GTX 1060 in 2016, so it will be interesting to see what this means for the generational performance of the card, especially in 1080p games that still want to rasterize a relatively large number of pixels.

All of these blocks, in turn, will be clocked a little more aggressively for an Ampere part, with NVIDIA setting the baseline specifications for the RTX 3060 at 1.32GHz for the base clockspeed, and 1.777GHz for the rated boost clock. On paper, this means the RTX 3060 will offer around 80% of the RTX 3060 Ti’s shader and texture performance, with the card’s relatively high boost clock helping to make up for the overall lack of SMs. Meanwhile we’re looking at just 64% of RTX 3060 Ti’s ROP throughput.

On the memory side of matters, GA106 introduces a smaller 192-bit memory bus, which is comprised of six 32-bit memory controllers. This is standard fare for 60-tier cards, and does come with some trade-offs, mainly in terms of memory capacity. At three-quarters the size of the bus on the likes of the RTX 3070, NVIDIA normally opts to include three-quarters the memory, which would have been 6GB. However, as a 6GB card is looking kind of long in the tooth when game consoles have 16GB of total memory, likely impacting the card’s long-term performance prospects, NVIDIA has decided to finally bite the bullet and make the jump to 12GB of VRAM on their mid-tier video cards. The net result is that there’s currently an odd progression/regression in the GeForce RTX 30 lineup – going from 12GB down to 8GB and then back up to 10GB on the RTX 3080 – but at the end of the day it ensures the RTX 3060 ships with enough VRAM to be useful for years to come.

NVIDIA’s choice of VRAM, in turn, ends up being quite interesting. The company’s official specifications call for RTX 3060 cards to use (at least) 15Gbps GDDR6 memory; the GDDR6 is to be expected, the speed grade is not. To the best of our knowledge, none of the GDDR6 vendors offer a 15Gbps speed grade – after 14Gbps, the next speed grade is 16Gbps. This means that in order to meet the 15Gbps requirement, AIBs will need to use 16Gbps chips to begin with. So why the odd choice in memory speed? The most rational explanation is so that it leaves a bit of room for vendors to sell factory overclocked cards, as this way they can sell cards with 16Gbps memory clocks as a premium product without actually exceeding the memory’s specifications.  In any case, using 15Gbps GDDR6 gives the RTX 3060 a total memory bandwidth of 360GB/second, 24GB/second more than what the RTX 3060 offered.

With lower performance and memory amounts, TDP is also dropping accordingly relative to the higher-end RTX 30 series cards. NVIDIA is rating the RTX 3060 at 170 Watts, which is 30W below the RTX 3060 Ti, and making this the first RTX 30 series card with a sub-200W TDP. However, much like the other RTX 30 series cards, TDPs are rising on a generational basis. Whereas the GTX 1060 was a 120W card and the RTX 2060 a 160W card, RTX 3060’s 170W TDP is a further 10W higher. If nothing else, this is consistent with what NVIIDA has done with this entire generation of cards, but it means that the RTX 3060 series is going to be more powerful in more than just rendering performance.

An Extra Feature: Throttling Ethereum Mining Performance

As briefly mentioned towards the introduction, the RTX 3060 also introduces an interesting new anti-mining feature for the first time in any NVIDIA GeForce card. Specifically, the RTX 3060 is designed to throttle back on Ethereum mining, essentially capping the mining performance for the cryptocurrency at 50% of the card’s actual (native) rate. As announced by NVIDIA last week, this is part of an effort to segment the gaming and mining markets – and to keep gaming cards from being pressed into a life of chasing digital riches.

When NVIDIA originally announced the throttling mechanism, they merely called it a “driver” feature – raising immediate questions about how this might be bypassed. However since then, NVIDIA has clarified that the feature is actually a mix of driver, hardware, and firmware functionality, relying on a “secure handshake” between the three of them. The involvement of the firmware makes things far more interesting (and far harder to break) since NVIDIA requires signed firmware on all GeForce video cards, meaning that an RTX 3060 card’s firmware can’t be modified nearly as easily as drivers.
...

Continue Reading

[harlan4096]

Additional Info: https://www.tomshardware.com/news/rtx-30...-synthetic