NVIDIA this morning is quietly adding to its menagerie of high-end video cards with a third version of the GeForce RTX 3080, the simply-named GeForce RTX 3080 12GB. Just as the name says on the tin, this latest GeForce card is more or less a version of the existing RTX 3080 with 12GB of memory, and the additional capacity and memory bandwidth benefits that come from that. This latest video card launch is relatively subdued launch for the company, and NVIDIA is not making much fanfare for the new card – nor are they announcing a price for it.
The third member of the RTX 3080 family comes as the cryptocurrency-driven GPU shortage has officially entered its second year. NVIDIA and its partners are selling every card they can make, and as a result traditional product stack logic has gone flying out the window at 360 fps. Instead, NVIDIA (and AMD) are left optimizing their product stacks to best match that insatiable demand, along with getting every useable chip in a card and on the market. And even with NVIDIA having switched its non-3090 cards to Ethereum hashrate nerfed LHR versions over half a year ago, there’s no immediate sign that the heavy demand for video cards will wind down any time soon.
Consequently, NVIDIA isn’t saying much about the new RTX 3080 SKU, primarily because they don’t need to. Which to be sure isn’t a criticism of NVIDIA, but it is a sign of the times. Officially the card exists for high-end gaming, but NVIDIA isn’t bothering to put together any kind of promotional campaign outlining the benefits of the card, or why they thought it necessary to introduce a 12GB SKU now, etc. Even the announcement of the card itself was buried in an announcement about DLSS support for a port of a Playstation 4 game (God of War). Simply put, the GeForce RTX 3080 12GB now exists, and for right now that’s enough for NVIDIA.
| NVIDIA GeForce Specification Comparison | ||||||
| RTX 3080 Ti | RTX 3080 12GB | RTX 3080 10GB | RTX 3070 Ti | |||
| CUDA Cores | 10240 | 8960 | 8704 | 5888 | ||
| ROPs | 112 | 96 | 96 | 96 | ||
| Base Clock | 1.37GHz | 1.26GHz | 1.44GHz | 1.58GHz | ||
| Boost Clock | 1.67GHz | 1.71GHz | 1.71GHz | 1.77GHz | ||
| Memory Clock | 19Gbps GDDR6X | 19Gbps GDDR6X | 19Gbps GDDR6X | 19Gbps GDDR6X | ||
| Memory Bus Width | 384-bit | 384-bit | 320-bit | 256-bit | ||
| VRAM | 12GB | 12GB | 10GB | 8GB | ||
| Single Precision Perf. | 34.1 TFLOPS | 30.6 TFLOPS | 29.8 TFLOPS | 21.7 TFLOPS | ||
| Tensor Perf. (FP16) | 136 TFLOPS | 122 TFLOPS | 119 TFLOPS | 87 TFLOPS | ||
| Tensor Perf. (FP16-Sparse) | 273 TFLOPS | 244 TFLOPS | 238 TFLOPS | 174 TFLOPS | ||
| TDP | 350W | 350W | 320W | 290W | ||
| GPU | GA102 | GA102 | GA102 | GA104 | ||
| Transistor Count | 28B | 28B | 28B | 17.4B | ||
| Architecture | Ampere | Ampere | Ampere | Ampere | ||
| Manufacturing Process | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm | ||
| Launch Date | 06/03/2021 | 01/11/2022 | 09/17/2020 | 06/10/2021 | ||
| Launch Price | MSRP: $1199 | MSRP: [undefined] | MSRP: $699 | MSRP: $599 | ||
So what does the latest RTX 3080 SKU bring to the table compared to the 10GB RTX 3080 classic? The high point is of course the memory capacity, but there are actually a few different things going on here.
On the GPU front, NVIDIA is actually using a slightly better version of their venerable GA102 GPU, which now is used across 5 different desktop video cards. The version of the GA102 used here has a slight increase in the number of SMs enabled versus the OG RTX 3080, with 70 SMs as opposed to 68 on the original card. Clockspeeds have also changed a bit; while the official boost clock rating is still 1.71GHz, the base clockspeed for the new SKU is 1.26GBz, 180MHz below the more basic 3080. Ultimately this seems to be a function of TDP, as the additional memory and additional transistors being lit up on the GPU will increase the power needs of the card, especially in a maximum-load scenario.
As for the memory, the increase to 12GB of GDDR6X comes with a matching increase in the width of the memory bus. The RTX 3080 12GB sees GA102’s full 384-bit memory bus enabled, reflecting the addition of 2 more GDDR6X memory chips (64-bits) to the memory bus, bringing the total to 12 chips/384-bits. According to NVIDIA’s specifications, they’re using the same 19Gbps GDDR6X chips here as on the classic RTX 3080, so memory clockspeeds have neither been dialed up or dialed down. So the expansion of the memory bus brings with it both an additional 2GB of VRAM – which will come in handy at 4K – as well as a 20% increase in memory bandwidth. Compared to the 10GB RTX 3080 and its 760GB/second of memory bandwidth, the 12GB RTX 3080 offers 912GB/second of bandwidth.
But to pay the bill for all of this, so-to-speak, the TDP of the newer 12GB SKU is also higher than the 10GB cards. Here NVIDIA’s official/minimum TDP has gone from 320W to 350W, a 9% increase. And as we noted before, even with this TDP increase, the minimum/base clockspeed still needed to be turned down a bit. This gives the RTX 3080 12GB the same official TDP ratings as both the RTX 3080 Ti and RTX 3090, and if these values are accurate, then it implies that the new card will have the lowest energy efficiency out of all of them.
Speaking of the RTX 3080 Ti, the configuration of the new card immediately raises the question of what to expect in terms of performance versus NVIDIA’s best 3080 card – itself essentially a slightly cheaper RTX 3090. The RTX 3080 Ti was about 10% faster than the original RTX 3080, and while benchmarks will be needed to draw exact figures, I expect the RTX 3080 12GB to essentially split the difference. That would put it around 5% faster than the OG card, and the RTX 3080 Ti about 5% faster than that. But along with underscoring the fact that this is an estimate, it should be noted that the difference will vary from game to game, and that games that are especially bandwidth sensitive have the most to gain, particularly at 4K.
Unfortunately, pricing won’t offer much of a guide here. Seeing as how NVIDIA isn’t even selling an Founder’s Edition of the card, they’re not providing an official MSRP – and it’s not as if the irrational market would follow it anyhow. The best guidance we have right now is looking at what NVIDIA’s board partners are charging/trying to charge for their cards. And in that case, the cheapest RTX 3080 12GB being listed this morning is an EVGA model at $1249. That’s $50 over the RTX 3080 Ti MSRP and $40 over their own cheapest RTX 3080 Ti, but also generally a couple of hundred below the rest of their RTX 3080 Ti lineup.
At best, it’s fair to say that the RTX 3080 12GB is unlikely to be priced much differently than the RTX 3080 Ti. Which shouldn’t be too surprising since so much of the current crypto ecosystem is based around memory bandwidth, and the two cards are identical in that respect. Though even in a gaming context, the RTX 3080 12GB is very likely to be within a few percent of the RTX 3080 Ti. Put another way, don’t expect to pay less than $1200 for the RTX 3080 12GB, even if you can get it at manufacturer (as opposed to market) prices. Otherwise the more positive news, at least, is that even following the launch of the new card, according to NVIDIA the RTX 3080 classic isn't going away; so it will still be produced for the consumer market.
Wrapping things up, expect to see cards in the coming days and weeks from the usual suspects, including EVGA, Zotac, ASUS, Gigabyte, and MSI. NVIDIA is not announcing a hard availability date, so given the ongoing GPU shortage, we expect to see cards slowly filter into the market, and then leave it almost immediately. Happy hunting!
ncG1vNJzZmivp6x7orrAp5utnZOde6S7zGiqoaenZH54fo9tZqeumZm2onnQrqCerJyueq2t1KeaoZ2jYrSmss6rmp5loqnFbn%2BPcWdmaWKcr265zqucZq6ilrpuuc6rnGaon6yys3nMqKmeZZ2ku6bF