NVIDIA RTX 3090, RTX 3080, RTX 3070: Everything you need to know about the latest series of graphics cards

The next RTX generation

Ampere is Nvidia’s second RTX generation. The first appeared exactly two years ago as Geforce RTX 20 (codename Turing) with the promise of having “reinvented graphics”. The focus of architecture is no longer just an increase in performance – as in the case of its predecessor Pascal – but a quantum leap on the road to photorealism: Hybrid Raytracing instead of rasterization. RTX 3000 series is the new and shining series.

To this end, Nvidia, as the first GPU manufacturer, implemented specialized hardware units that can handle highly complex calculations at real-time frame rates.

California, along with many other companies, including AMD and Intel, provided the basis for DirectX:For two years now, the Microsoft Interface Collection has included the DXR – DirectX Raytracing – and thus creates the necessary software environment to speak standardized with the hardware.

How the latter deal with the orders, however, is not prescribed. Graphics chips can roll out raytracing rendering requirements on their generalized computing, equip them with appropriate special features, or lift stand-alone fixed-function cores out of the box.

Last resort of Nvidia at Turing – all information can be found atequip these with appropriate special functions or lift independent fixed-function core from the baptism. Last resort of Nvidia at Turing – all information can be found atequip these with appropriate special functions or lift independent fixed-function core from the baptism. Last resort of Nvidia at Turing – all information can be found at detailed technical dossier .

Nvidia recently agreed that Turing could not meet the high expectations of PC gamers. These assumed that RTX 20 could not only display handsome graphics, but also be faster than GTX 10. Surprisingly, the Geforce makers claimed in the same breath that this wish was fulfilled with Ampere alias RTX 30.

Thanks to a “gigantic step” in effective performance, the new Geforce RTX 3080 should be significantly faster with active raytracing than the 2017 top model Geforce GTX 1080 Ti. Of course, we’ll see if that works out – in addition to many other things. Beforehand, however, it is important to outline the improvements and specifications. Who wants to know exactly, scrolls directly to the second page of this article,on which we look under the hood of the Gamer Ampere GA102.

As mentioned at the outset, Nvidia is announcing the amp era with the Geforce RTX 3080. This is according to the Nvidia boss’ revelation presentation officially around the flagship – however, only if one follows the arbitrary definition.

In fact, the Geforce RTX 3080 is not the fastest model in the just-launched RTX-30 portfolio, but as the classic “80s” only the number 2. Already on September 24, the Geforce RTX 3090 goes to the start, whose appearance with full intention the Category “Battleship” fulfills, with a price point of 1,499 euros, however, the budget exceeds many interested parties.

At the same time, the Geforce RTX 3080 with its price tag of 699 euros looks like a bargain – and the RTX 3070 for 499 euros anyway. You will find out what the RTX 3090 and RTX 3070 do in a nutshell at PCGH. First of all, it’s all about the Geforce RTX 3080, whose Founders Edition (reference design) will be available from now on.

Performance And Memory

Like the RTX 3090, the Geforce RTX 3080 relies on the brand new Ampere GA102, Nvidia’s most advanced gaming processor to date. The new chip houses 10,752 shader ALUs in full construction – a value that you only have to lower once.

Possibly this is due to a refined Turing production, which in contrast to the previous top chips does not originate from the house TSMC. Instead, Nvidia agreed with Samsung, whose 8LPP procedure does not come with new EUV exposure, but this shows a good yield and above all a good price-performance ratio.

Nvidia uses a custom process called 8N (-> 8 nm Nvidia). For comparison: The professional accelerator GA100, a close relative of the GA102 for players, runs at TSMC in state-of-the-art 7-nm fabrication of the band, however, due to its size and complexity, makes completely different demands on the Foundry (and in the bank account).

Neither the RTX 3090 nor the RTX 3080 are working on the full expansion of the GA102, aptly named “full-fat” in the English-speaking world. Instead, the Geforce RTX 3080 has to make do with 68 of the physically available 84 shader multiprocessors, bringing 8,704 ALUs to the fore. The incision has mainly production-technical reasons, because in this way partially defective chips can be used meaningfully.

As with the RTX 2080 Ti, GTX 1080 Ti and many other Nvidia graphics cards, this also includes the omission of parts of the memory interface: 320 of 384 data paths are active on the RTX 3080. Although this interface width is not entirely new – one remembers the Geforce 8800 GTS 14 years ago – but leads to a brand new storage capacity: 10,240 MiByte or 10 GiByte are installed on the RTX 3080.

This corresponds to +2 GiByte compared to the previous 80s models GTX 1080 and RTX 2080, but -1 GiByte compared to the officially dependent RTX 2080 Ti – and also -1 GiByte compared to the three year old GTX 1080 Ti. Whether Nvidia has constructed a target breaker here, which should be targeted at the RTX 3090 and its 24 GiByte, you will find out below.

The most important change within the core concerns the number of sophisticated FP32 ALUs. These arithmetic-logical units are responsible for the numerous semicolon operations that are widely used in modern games, and thus to a large extent for performance. The GA102 is now able to run 128 FP32 calculations per shader multiprocessor in parallel – at Turing there are only 64, so half.

Load scenarios, which consist of pure slide calculations, therefore add up to factor 2 – the number is, as one might first assume, no marketing invention, but real. However, there is a catch. Turing is able to solve 64 INT32 tasks in parallel in addition to 64 FP32 ops.

These integer calculations run with commas instead of commas, are consequently less time consuming,and moreover rarely in games. In such a case, Ampere must select a mode for the second ALU phalanx, so that the distance to Turing shrinks. The latter is different in each application, as we will see in the benchmarks.

The Core

When the width of the memory interface stagnates or even backs up, only one means leads to increased throughput: Height – and indeed at the data rate of the installed memory. For the result, it hardly matters whether you put the tasks in double so many hands or let half the workforce double work so quickly. Of course, the best is both, but of course this is not for nothing. To meet the growing bandwidth requirements, Nvidia and storage specialist Micron have teamed up.

Released is GDDR6 X -DRAM, a further development of GDDR6, which should achieve up to 21 gigatransfers per second (GT / s). For comparison: GDDR6 officially ends at 16 GT / s. Possibly this is due to a fundamentally old, but refined intervention in signal transmission, which now knows four instead of only two states (“PAM4”).

Thus, per beat, the double amount of information is transmitted, which lowers the internal beat and saves energy, without the effective performance decreasing. Nvidia Geforce RTX 3080 in test: GDDR6X memory for maximum transfer rate despite narrow 384 and 320 bit connection respectively.

Nvidia Geforce RTX 3080 in test: GDDR6X memory for maximum transfer rate despite narrow 384 and 320 bit connection respectively.
In the case of the Geforce RTX 3080 Founders Edition, the storage rate is “only” 19 GT / s, translatable at 9,500 MHz compared to the previous 14 GT / s and 7,000 MHz respectively.

What counts in the end is the product of interface width and tact. Here, the RTX 3080 closes at 770 GByte per second from A to B – almost 70 percent more than the RTX 2080/2070 (S) / 2060S, 53 percent more than the RTX 2080 Super and at least 23 percent more than the RTX 2080 Ti. The upcoming RTX 3090 even pushes around 936 GByte / s and thus fails at the magical limit of 1 TByte / s – that works in the end customer area only AMD’s Radeon VII .

These values ​​compare unusually well between Turing and Ampere, as Nvidia confirms on demand, the “GA102 does not have any further improved memory compression.” GA102 over no further improved memory compression, but draws in this regard with Turing. Previous generations have always had new tricks in place to pack data losslessly and thus reduce data traffic on the memory bus.

This may be one of the reasons why the raw transfer rate was raised quite clearly – when switching from Pascal to the Turing top model (Titan to Titan) it was “only” around 23 percent. Below the line, Amperes storage transfer rates can be seen, but the capacity is conceivable in relation to computing performance.

In the following table, which has been tested and calculated by us several times, we summarize all relevant performance data. Included is not only the Geforce RTX 3080, but also its two siblings with the final specifications, as well as various older models. Particularly exciting for starters is the look at the GTX 1080 Ti, which next to its modern counterparts looks a bit dusty. On the next article page we will see the GA102 under the Heatspreader.

Price and Bottom line

The RTX 3090 will be available from September 24th, the RTX 3080 from September 17th. The RTX 3070 is expected to follow in October.

However, the prices are a neat surprise. Due to the large number of leaks in the run-up, which promised significant increases in performance, everyone assumed that the prices would be quite substantial. But only the RTX 3090 is in the four-digit range at 1,499 US dollars. GeForce RTX 3080, on the other hand, costs just as much as its predecessor card, namely $ 699. The RTX 3070 price-performance card costs $ 499.

All cards rely on PCI-Express 4.0 and come with the HDMI 2.1 and Displayport 1.4a video connections. Nvidia dispenses with the VR connection Virtual Link via USB-C – the connection planned as a new VR standard can thus be finally buried.

Three new GeForce RTX 3000 graphics cards from Nvidia are casting their shadows ahead – the GeForce GTX 3090, which costs 1,500 euros, is only aimed at a presumably extremely small group of enthusiasts. So I leave out this top model here.

For the vast majority of gamers, the two other models in the trio that Nvidia announced a few days ago are more interesting: On September 17th, the starting shot for the new Ampere GPUs with the Nvidia GeForce RTX 3080 will be given, followed by the GeForce in October RTX 3070. Both graphics cards are said to be stronger than the current top model GeForce RTX 2080 Ti, for which more than 1000 euros is still being called.

This alone is quite remarkable, but the message only becomes something special when you consider the prices. Because the two new graphics cards will cost 499 and 699 euros, respectively – you will get the previous top performance for less than half the price, provided the announcement does not exaggerate the performance.

Although we have always described the current top model of the 2000 series in our purchase recommendations as overpriced, as was the case with our last graphics card purchase advice in June – the potential performance of the RTX 3000 models seems to be so high that the announced prices appear to be comparatively good.

You can find more of such articles at MYESPORTSGLOBE

Follow our Social media handles for more exclusive eSports News Instagram Twitter Facebook