Nvidia RTX 3070, 3080 & 3090 Preview – Everything about the new generation Nvidia GeForce

By admin On Sep 1, 2020

Spread the love

At six o’clock tonight, Nvidia aired a pre-recorded event in which CEO Jensen Huang introduced the new Ampere video cards for gamers. A lot had already been leaked about the GeForce RTX 3070, RTX 3080 and RTX 3090, but Nvidia nevertheless managed to surprise on points – the new GPUs appear to have twice as many Cuda cores as previously suggested. In this preview you can read everything we know so far about Nvidia’s brand new gaming cards.

Nvidia GeForce RTX 3000: Ampere + 8nm

The main innovations in the GeForce RTX 3000 series are the Ampere architecture, which succeeds both the Volta architecture for servers and the Turing architecture for gaming, and Samsung’s 8nm process ‘co-developed with Nvidia’. This process is a further development of Samsung’s 10nm process, which means that no EUV is used in production yet.

The line-up: RTX 3070, RTX 3080 & RTX 3090

The line-up currently consists of three models, with the flagship GeForce RTX 3090 (nicknamed BFGPU – feel free to think for yourself what that stands for) based on the GA102 GPU. In fact, this is the successor to the RTX 2080 Ti (or even the Titan RTX, as Nvidia itself says), which contained the TU102 chip, but to simplify the naming, they simply chose a higher model number this time. The RTX 3080 is based on the same chip, but with 17% fewer computing units enabled. Only the RTX 3070 uses the GA104 chip – for comparison, the TU104 was already used for the RTX 2080 in the previous series.

	RTX 3090	RTX 2080 Ti	RTX 3080	RTX 2080	RTX 3070	RTX 2070
Architecture	8nm, GA102	12nm, TU102	8nm, GA102	12nm, TU104	8nm, GA104	12nm, TU106
Cuda cores	‘10496’	4352	‘8704’	2944	‘5888’	2304
Boost speed	1700MHz	1635MHz	1710MHz	1800MHz	1730MHz	1710MHz
Vram	24GB gddr6x	11GB gddr6	10GB gddr6x	8GB gddr6	8GB gddr6	8GB gddr6
Memory Speed	19.5GBit / s	14Gbit / s	19Gbit / s	14Gbit / s	16Gbit / s	14Gbit / s
Memory bus	384bit	352bit	320bit	256bit	256bit	256bit
Bandwidth	936GB / s	616GB / s	760GB / s	448GB / s	512GB / s	448GB / s
Tgp	350W	260W	320W	225W	220W	175W

Later in this article we will discuss the most important new features of the Ampere architecture, the GA102 and GA104 GPUs already mentioned and the design of Nvidia’s own Founders Edition. We must rely entirely on what Nvidia announced today; no samples have arrived in our lab yet and the extensive press briefings, which usually go deeper into the architecture, have not yet taken place. Fortunately, the first GeForce RTX 3000 video card will be in the shops on September 17, so it will not be too long to wait for all the details and of course extensive, independent benchmarks – but that does not mean that we are holding you back from our preliminary analysis.

Amps for gamers: 10,000+ cores?

The beating heart of RTX 3000 series graphics cards is of course the Ampere architecture. Not entirely new for the entire GPU market, because we saw Ampere previously in Tesla cards for servers, but Ampere is now coming out in GeForce products for the first time.

Improved sms, rt cores and tensor cores

The promises with the RTX 3090, 3080 and 3070 are great: up to twice the performance and 1.9 times the efficiency of Turing GPUs. The second generation RTX should achieve this through the enhancements that a new generation of sms, rt cores and tensor cores have received. With technical details, Nvidia was still sparse on Tuesday.

Two new GPUs: GA104 and GA102

We already mentioned that the three video cards that Nvidia announced today are based on two GPUs: GA104 and GA102, where the G and A respectively stand for GeForce and Ampere. There is an even bigger chip, the A100, which Nvidia does not yet use in consumer products, but which was already in the Tesla A100 earlier this year.

If we compare the specifications, we first notice that the GA102 GPU is a lot smaller than GA100: it contains almost half the number of transistors. This is partly because it contains fewer streaming multiprocessors (SMS), namely 82 instead of 108, but also because Nvidia has replaced the hbm2e memory controller with one that works with cheaper gddr6x memory, about which more later.

	GA100	GA102	GA104	TU102
Applied in	Tesla A100	RTX 3080, RTX 3090	RTX 3070	RTX 2080 Ti
Manufacturing process	7nm TSMC	8nm Samsung	8nm Samsung	12nm TSMC
Transistors	54 billion	28 billion	nnb	18.6 billion
That size	826 mm²	627 mm²	nnb	754 mm²
Cuda cores	6912	5248 *	2944 *	4352
TGP	400W	350W	220W	260W
Memory	HBM2e	gddr6x	gddr6	gddr6

Against all rumors and even specifications of video card manufacturers, Nvidia claims that the GA102 and GA104 GPUs contain twice as many cuda cores, namely 10496 pieces in GA102 and 5888 pieces in GA104. That is impossible in the conventional way: both in terms of tdp and that size as transistors, so many cuda cores do not fit in the mentioned GPUs.

The most important clue for what’s going on can be found on a specifications page that Nvidia put online after the stream. It states that the RTX 3000 series SMSs each contain two fp32 compute units, compared to one in all previous generations.

In the documentation of the A100-GPU we find a block diagram showing the internal structure of an Ampere-sm. For a total of 64 FP32 units, often simply referred to as shader units, there are four clusters each containing sixteen of those FP32 units in one SM. However, eight fp64 units have also been placed per cluster. Most likely, Nvidia has equipped those FP64 units with the option of also serving as a double FP32 unit, which effectively means that there are no 16 but 32 FP32 units per cluster, or 128 per SM.

If this hypothesis is correct, the Ampere GPUs for gamers have an enormous ditch of FP32 computing power. A doubling of the performance seems too optimistic, as other parts of the chip (dispatcher, scheduler, caches, memory bandwidth) will likely create a significant bottleneck, but we could potentially see some very impressive gaming performance.

Memory, PCI-E 4.0 and RTX IO

Nvidia has worked in collaboration with Micron on gddr6x, a further development of gddr6. The latter version was already used on Turing-based video cards and is also making a return on the RTX 3000 series. However, the most high-end models, the RTX 3080 and 3090, use gddr6x. We officially encounter this new type of video memory for the first time, and it delivers twice the throughput speed of normal gddr6. This is made possible by processing four values per clock cycle on gddr6x, versus two on gddr6. You could almost compare it with slc versus mlc with flash memory: we no longer only work with 0 and 1, but with 00, 01, 10 and 11.

On the RTX 3080, this, combined with 19Gbps memory coupled to a 320-bit memory bus, means a memory bandwidth of no less than 760 GB per second, 70 percent more than its predecessor. With the X version, the maximum is actually squeezed out of gddr6, which in turn does not benefit the consumption and heat production – this is probably one of the reasons that the stated consumption of Ampere cards is significantly higher than its predecessors.

PCI Express 4.0

PCI-express 4.0 is one of the other specifications that Nvidia briefly shed light on during the presentation. Not like the first in the world, that was competitor AMD with the introduction of the Navi architecture. But if we are to believe Nvidia’s performance claims, the high-end RTX 3000 cards will probably benefit more from the new standard to make the most of the graphics memory buffers.

From pci-e 4.0 ssd to vram with RTX IO

In addition to the memory on the video card itself, according to Nvidia, a higher throughput speed is needed between the vram and the system memory to be able to play future games properly. Modern SSDs on pci-e 4.0 are ideally suited for this, but the amount of processor power throws a spanner in the works here, according to the chip designer. Many processor cores are needed to have data from a storage medium processed by the CPU and placed in the system memory, after which it is sent via the PCI-e bus to the GPU and eventually the vram. Particularly with compression – the rule rather than the exception in today’s games – the processing power required for this is large.

Nvidia wants to solve this problem with RTX IO, by directing data from an SSD to the Vram and lossless decompression in the GPU itself. This should free a large number of CPU cores from the above work and drastically reduce loading times. Details about how RTX IO will work exactly are not yet available – at first sight it does not seem possible to have SSDs communicate with the video card without the intervention of the CPU, but we will undoubtedly return to this in our review.

It seems that the manufacturer with RTX IO will use a just announced api from Microsoft: DirectStorage . Not surprisingly, that api will also be used in the upcoming Xbox Series X, and we will also see a similar technique in the PlayStation 5. The name suggests that Pascal-generation and earlier video cards cannot support RTX IO, but it is as yet unclear whether RT or tensor cores will actually be used when moving the data.

8K, AV1 and HDMI 2.1

The first 8K gaming GPU. For example, Nvidia puts its new top model, the RTX 3090, in the announcement. This video card would be so fast that 8K should be within reach in this generation of video cards. With four times the number of pixels of 4K, a lot of computing power is required to make this possible, but the manufacturer demonstrated that this should be possible at a frame rate of 30 frames per second, especially when using dlss.

AV1 decoding, but no encoding yet

Also for the first time we see support for the AV1 video compression standard on a video card. The GeForce Ampere cards can play this hardware as standard without claiming computing power outside the media block. It is important that the manufacturer is specifically talking about decode, which gives the impression that encoding with the same part will not be possible – the unchanged nvenc generation seems to confirm this.

HDMI 2.1 for 4K / 144Hz and 8K / 60Hz

Finally, HDMI 2.1 was mentioned as a specification. Much has been done to achieve this standard, and we discussed this in depth in our background article earlier this year. For the RTX 3000 series, the advantages of HDMI 2.1 are relevant, as the higher bandwidth 144Hz at a 4K resolution or 60Hz at 8K becomes possible. Support for variable refresh rates is also obvious, which Nvidia has already opened up under the name G-Sync Compatible.

Founders Edition: design and cooling

One of the most controversial leaked details about the new generation of Nvidia video cards was the futuristic cooler of the Founders Edition, as the GPU designer has called his reference model for several generations. Nvidia promises double the cooling compared to the open dual-fan cooler it used for the RTX 2000 series.

Founders Edition: futuristic cooler and ultra short pcb

In order to be able to cool the GA102 GPU in the RTX 3080 and RTX 3090 – with well over 300 watts, these are not economical cards – Nvidia has developed a unique cooler, where the two fans are placed on either side of the heat sink. The outer fan is placed at the back, but also placed the other way around, so that they both blow in the same direction. Under each fan is a conventional heatsink with cooling fins, which are connected to the GPU via five heat pipes.

Because the printed circuit board (also equipped with 18 power phases for the GPU) of the Founders Edition is very short, the outer fan placed on the back sucks completely ‘through the video card’ – the air flows in at the front, through the cooling block and then on. the back out again. In a standard housing, the warm air will be blown out of the cabinet either by the outtake fans placed at the top or by the CPU cooler. The inner fan in turn blows through the short PCB practically straight on the GPU and VRM, and can partially discharge its warm air directly from the housing.

RTX 3070 and 12-pin power connector

The GeForce RTX 3070 has a cooler more like what we’re used to, with two fans on the front. As evidenced by the air outlet in the backplate, this card also has a short pcb and the second fan blows (partly) through the card.

All RTX 3000 Founders Editions have a striking power connection; the compact 12-pin connector that previously surfaced in rumors. You don’t have to worry about compatibility though, as an adapter to two regular eight-pin peg plugs is included in the box. Manufacturers of custom video cards seem to use the trusted connections anyway.

Dimensions

The RTX 3090 Founders Edition is the only triple-slot card that, with a width of 13.8 cm, also extends well above the standard width of an expansion slot. This video card is 31.3 cm long, on the edge of what fits in most cabinets. The RTX 3080 and RTX 3070 are exactly the same width as a standard pcie slot at 11.2 cm, only two locks high and at 28.5 and 24.2 cm respectively also a lot shorter.

Availability and prices

The Nvidia GeForce RTX 3080 is the first of the new cards to hit stores, on September 17th. The top model RTX 3090 will follow a week later on September 24, for the RTX 3070 no more concrete date than ‘in October’ has been released.

Nvidia has now announced the euro prices of its own Founders Edition cards. The RTX 3070 will cost 519 euros, the RTX 3080 will be available for 719 euros and for the RTX 3090, Nvidia will charge a whopping 1549 euros, almost 300 euros more than the RTX 2080 Ti cost.
Naturally, Nvidia also bundles a game with the RTX 3000 video cards, namely Watch Dogs: Legion .

Released on October 29, this Ubisoft game takes place in a future London, where the city has turned into a police state where you as a citizen are completely controlled. As a member of a secret group of hackers, you take on the state. If you purchase an RTX 3000 video card or a PC equipped with it from a participating retailer, you’ll get game code for Watch Dogs: Legion and a year-long GeForce Now, Nvidia’s cloud streaming service for games.

Performance claims and initial benchmarks

Nvidia showed some performance claims during its presentation, usually in somewhat vague terms and moreover often based on tests with real-time ray tracing and dlss enabled. In such scenarios, the improved rt and tensor scores probably play a major role, making them not representative of the vast majority of gameplay, even without ray tracing and at the native resolution of your screen.

Nvidia’s own claims

Still, we’re taking a look at Nvidia’s claims – it’s one of the few benchmarks we have so far, and everyone wants to know how the new cards will perform. In a whole host of rtx applications, Ampere is between 60 and 100 percent faster than Turing, according to Nvidia – which new card Nvidia is comparing to which old card here is not clear.

Furthermore, according to Nvidia, Ampere achieves the performance level of a Turing card with a tdp of 250 watts (read: the RTX 2080 Ti) with a consumption of only 120 watts, which results in an improvement in performance per watt by 90%. However, to achieve the 80% higher performance shown in this graph, a much higher power of 320 watts is also required. The latter corresponds to the tgp of the RTX 3080.

The first in-game benchmarks

One UK website has already been able to benchmark a GeForce RTX 3080 under the supervision of Nvidia. Digital Foundry was not allowed to choose the games, resolution and settings itself, so take the results with a grain of salt for the time being. On average over six games, the RTX 3080 performed 80% better than the RTX 2080, and it is noticeable that the games that were tested without RTX also show good profits.

4K @ ultra settings	RTX 3080 vs. RTX 2080
Borderlands 3	+ 81.6%
Doom Eternal	+ 84.2%
Shadow of the Tomb Raider	+ 69.8%
Control	+ 77.6%
Ray Tracing: Quake 2 RTX	+ 92.3%
Ray Tracing: Control / High RT / DLSS Perf	+ 77.6%

Source: Eurogamer / Digital Foundry

Finally

After the presentation of Nvidia we can say that the expectations are high and that improvements and increases in performance do not seem to be small. Regardless of what competitor AMD comes up with, Nvidia doesn’t seem to have slowed down with Ampere and the next generation of Radeons must also see significant improvements to keep up with the green camp.