The autumn of 2020 will undoubtedly bring a huge amount of exciting technological news from a wide flora of companies. Microsoft and Sony are launching new game consoles with graphics technology from AMD, which uses the architecture RDNA 2 in this autumn’s dedicated graphics card with “Big Navi” at the forefront. Intel launches “Tiger Lake” for laptops with Xe graphics. However, Nvidia is ahead of all companies with the unveiling of the Geforce RTX 3000 series, with the Ampere architecture under the hood.
During an event entirely dedicated to the new product line, the graphics cards Geforce RTX 3090, RTX 3080 and RTX 3070 will be unveiled. the step up in computing resources is real, but shockingly enough, the step is taken to as many as 10,496 CUDA cores, which is paired with 24 GB of fast GDDR6X memory in the series’ top card.
Specifications: Geforce RTX 3090 and RTX 3080
RTX 3090 | RTX 2080 Ti | RTX 3080 | RTX 2080 | |
---|---|---|---|---|
Technical | 8 nm Samsung | 12 nm TSMC | 8 nm Samsung | 12 nm TSMC |
Circuit | GA102 * | TU102 | GA102 * | TU104 |
Circuit surface | 627 mm2* | 754 mm2 | 627 mm2* | 545 mm2 |
Transistors | 28 billion | 12 billion | 13.6 billion * | 7.2 billion |
Architecture | Ampere | Turing | Ampere | Turing |
CUDA cores | 10 496 st. | 4 352 st. | 8 704 st. | 2 944 st. |
Clock frequency | 1 410 MHz* | 1 350 MHz | 1 440 MHz* | 1 515 MHz |
Boost frequencies | 1 700 MHz | 1 545 MHz | 1 710 MHz | 1 710 MHz |
Computational power | 35 580 GFLOPS* | 13 448 GFLOPS | 29 770 GFLOPS* | 10 068 GFLOPS |
Memory bus | 384-bit | 352-bit | 320-bit | 256-bit |
Memory amount | 24 GB GDDR6X | 11 GB GDDR6 | 10 GB GDDR6X | 8 GB GDDR6 |
Memory frequency | 19 500 MHz | 14 000 MHz | 19 000 MHz | 14 000 MHz |
Memory bandwidth | 936 GB/s | 616 GB/s | 760 GB/s | 448 GB/s |
Power supply | 12-pin | 8+8-pin | 12-pin | 8+6-pin |
Outputs | 2 st. HDMI 2.1 | 1 st. HDMI 2.0b | 2 st. HDMI 2.1 | 1 st. HDMI 2.0b |
Interface | PCI Express 4.0 | PCI Express 3.0 | PCI Express 4.0 | PCI Express 3.0 |
TDP | 350 W | 250 W | 320 W | 215 W |
Rec. Award | 1 499 USD | 999 USD | 699 USD | 699 USD |
* The information is not officially confirmed by Nvidia, but will be updated as soon as it becomes available.
Geforce RTX 3090 – Ampere on steroids
As early reports indicated, the RTX 3090 is a real monster with 24 GB of GDDR6X memory, which according to Nvidia makes it possible to play games at 60 FPS in 8K resolution. The large amount of memory operates at 21,000 MHz over a 384-bit memory bus, giving a total bandwidth of 936 GB / s. This is a step up of about 34 percent compared to the Geforce RTX 2080 Ti.
What was not expected before today’s event was how many CUDA cores the new beast is equipped with. Preliminary reports stated 5,248 cores, something Nvidia is literally blowing out of its seat as the actual number is an insane 10,496 cores. The hefty number was also blowing editor-in-chief Anton Nilsson out of the chair.
The abundant capacity also comes with a significantly increased TDP budget, despite the fact that the RTX 3000 cards are manufactured on more energy-efficient 8-nanometer technology from Samsung. Where RTX 3090 has a TDP level of 350 W, little brother RTX 3080 has lower but still high 320 W.
Geforce RTX 3080 – the new “normal” flagship
When Nvidia unveils the Geforce RTX 3080, it is called the series’ new top card, but the company then refers to the fact that it is a top card for the typical gaming fan who is not really willing to cough up the money an RTX 3090 costs. The card seems to justify the name flagship as the RTX 3080 according to Nvidia is twice as powerful as the Turing equivalent Geforce RTX 2080.
This claim is supported by the fact that the RTX 3080 is equipped with as many as 8,704 CUDA cores, which is more than a doubling of the 2,944 cores that the RTX 2080 possesses. The circuit is here capable of performing standard shader calculations with a capacity of 30 TFLOPS, 58 TFLOPS in ray tracing performance and 238 TFLOPS in Tensor calculations.
Geforce RTX 3070 – the entry board for RTX generation two
The entry card in Nvidia’s performance segment is in accordance with the company’s naming Geforce RTX 3070. Nvidia goes out of its way to announce that RTX 3070 is faster than the previous generation’s top card Geforce RTX 2080 Ti. In terms of performance figures, the card powers 20 TFLOPS shader performance, 40 TFLOPS ray tracingperformance and 163 TFLOPS Tensor performance.
Specifications: Geforce RTX 3070 vs RTX 2070
RTX 3080 | RTX 2070 | RTX 3070 | |
---|---|---|---|
Technical | 8 nm Samsung | 12 nm TSMC | 8 nm Samsung |
Circuit | GA102 * | TU106 | GA104* |
Circuit surface | 627 mm2* | 445 mm2 | 450 mm2* |
Architecture | Ampere | Turing | Ampere |
CUDA cores | 8 704 st. | 2 304 st. | 5 888 st. |
Clock frequency | 1 440 MHz* | 1 410 MHz | 1 500 MHz* |
GPU Boost | 1 710 MHz | 1 620 MHz | 1 730 MHz |
Computational power | 29 770 GFLOPS* | 7 465 GFLOPS | 20 310 GFLOPS* |
Memory bus | 320-bit | 256-bit | 256-bit |
Memory amount | 10 GB GDDR6X | 8 GB GDDR6 | 8 GB GDDR6 |
Memory frequency | 19 000 MHz* | 14 000 MHz | 16 000 MHz* |
Memory bandwidth | 760 GB/s | 448 GB/s | 512 GB/s* |
Power supply | 12-pin | 8-pin | 12-pin |
Outputs | 1 st. HDMI 2.1 | 1 st. HDMI 2.0b | 1 st. HDMI 2.1 |
Interface | PCI Express 4.0 | PCI Express 3.0 | PCI Express 4.0 |
TDP | 320 W | 175 W | 220 W |
Rec. Award | 699 USD | 499 USD | 499 USD |
* The information is not officially confirmed by Nvidia, but will be updated as soon as it becomes available.
To keep the high heat development that such complex circuits inevitably produce in check, Nvidia applies a new cooling solution, which the company unveiled before today’s launch. The new solution introduces, among other things, leaf springs of the low-profile type, and the fans have been optimized to better take in air and distribute it over the circuits on the graphics card. The electrical efficiency of the RTX 3000 series should also be improved compared to its predecessors.
A new current power supply is also that Nvidia introduces a more compact 12-pin power connector, which replaces the previous arrangement with 8 + 8-pin connectors for Nvidia’s flagship products. However, Nvidia announces that adapters for 12-pin to 8 + 8-pin connectors are included with the purchase of the company’s Founder’s Edition card. However, a number of partner manufacturers have presented their own products that apply 8 + 8-pin connectors.
RTX – Ampere fulfills the promise from Turing
When Nvidia launched the Turing architecture and the RTX 2000 series with pomp and circumstance, it was with the promise that the time had finally come for ray tracing in real time for games. As is so common among first-generation products, the ambition was perhaps a bit higher than what the practice delivered, and those who want to play 4K-resolution games near 60 FPS have usually had to really thumbs up on graphic finesse to get close to this.
With the Geforce RTX 3000, Nvidia addresses this concern in two ways. Partly through the new architecture Ampere which works more efficiently than its predecessor, and partly by simply throwing more resources at the problem. All cards in the RTX 3000 series are equipped with more shader processors than their predecessors, which together with a more efficient architecture according to Nvidia provides doubled RTX performance compared to previous generation graphics cards.
The long-awaited game Cyberpunk 2077 is used to highlight ris tracing i Geforce RTX 3000-serien.
Nvidia calls the Ampere “second-generation RTX core”, which comes with 2.7 times higher shader performance compared to Turing. In terms of ray tracing performance, the step will be 1.7 times what is offered in Turing at most, and even calculations of AI and machine learning take a real step up with 2.7 times higher Tensor performance than in its predecessor. The architecture is not only faster, but also more energy efficient where Nvidia claims twice as energy efficient operation compared to Turing.
It’s not just graphics performance and Tensor capabilities that are taking the step forward in Ampere. Nvidia is also introducing a technology called RTX IO, which, as the name suggests, increases the possible bandwidth between system storage and graphics card memory by compressing data in real time. According to Nvidia’s data, RTX IO provides as much as 100 times better bandwidth compared to a system that uses a conventional storage structure.
It will also reduce the load by 20 times compared to conventional SSD compression over PCI Express 4.0, which undeniably sounds like a real step forward for game developers if it matches real scenarios. RTX IO also supports Microsoft’s Directstorage technology used in Xbox Series X, which will be added to Windows 10 in the future.
The Ampere generation will not be as terribly expensive as many expected. The price for the entry-level model Geforce RTX 3070 lands at SEK 5,625. With the Geforce RTX 3080, the step is taken up to the easy price of SEK 7,825. Finally, little unexpectedly, the Geforce RTX 3090 requires a slightly maffier wallet with a price “from SEK 16,800”, which suggests that more expensive configurations will be added.
The Geforce RTX 3070 will go on sale in October, while the RTX 3080 will arrive a little earlier on 17 September. The monster card Geforce RTX 3090 speculators have to wait until September 24.