AMD Navi GPUs

**Winjer** · 30-11-20, 11:19

AMD Radeon RX 6700 XT as a real clock monster? Exclusive leaks of first board partner BIOSes including clock rates and TGP | Investigative

On what we can expect from Navi Flounder (Navi22), two very different board partner BIOSes now provide information. One of the two BIOSes is rather moderately designed and could well be the reference, the other one uses a much higher TGP and should therefore also allow higher clock rates. Also the specifications for the fans are interesting here, because one sees here for BIOS 1 rather a cheaper graphic card, while the presumed OC model should be equipped clearly better.
But let’s first come to the PowerPlay specifications, which are identical for both graphics cards. The slider in the Wattman now goes up to very theoretical 2950 MHz, the maximum memory clock is set to 1075 MHz as a simple value (slider 2150 MHz effective, i.e. x2). The maximum power limit can be set to +15%, but this refers to the TGP stored in the BIOS. If this increases, then the final value is also correspondingly higher. Also the GFX maximum increases compared to the RX 6800 XT reference to now 2854 MHz. You probably won’t reach that, but it will certainly go much higher than with Navi21.

Let us now come to the first differentiated. The BIOS 1 (left) looks much more modest in power consumption. The 186 watt TGP (for GPU and memory together) should also be above but still closer to the reference and end at a TBP of about 210 to 220 watts, while the card with the BIOS 2 (right) can grab a whopping 25 watts more and then already scratches at the 250 watt mark at the TBP. Don’t forget: in both cases, up to 15% can be added here (Wattman). Now let’s do the math. Based on the shaders, I had estimated about 67% of the power of a Radeon RX 6800 at the corresponding clock rates.
The Radeon RX 6800 has a TGP of 203 watts as reference card. The 186 watts of the BIOS 1 would therefore be 92% of the TGP, which also suggests an OC BIOS in view of the smaller memory (12 instead of 16 GB GDDR6). The BIOS 2 is even clearly above what AMD already gives the RX 6800 reference card! That’s a little bit of violent action, but we’ll see what kind of performance comes out in the end. By the way, both cards have 1.2 Volt as GPU voltage, also more than the RX 6800 and RX 6800 XT.

The simple card with the BIOS 1 looks like a 1-fan solution, whose small whirl has to whirl around properly in accordance with the GPU clock. The 2300 rpm for the Acoustic Limit is really a lot, especially since the 80 °C tjunction is still set quite low as a fan target. The other card looks better and quieter, despite higher TGP. The 1700 rpm already sounds a lot more bearable on paper and the tjunction of 95°C should also equalize the whole fan curve a bit and not let it rise so steeply.

Interim conclusion

So either these two small cards are overclocked power dwarves and wild barrel organs with immensely high clock rates, far above the sweet spot and the reference target, or Navi22 would have to work much more inefficiently than expected, which is rather unlikely. So you can rather assume that the board partners, whose name I hide, have put on a little bit of the electric crowbar to attack at least the upcoming GeForce RTX 3060 Ti or rather the RTX 3060. One can hardly reach the RTX 3070 in such a way, despite brutally high clock rates. Especially since the chip itself is also reduced in size compared to Navi21 and even the video encoder is supposed to be affected.
If everything goes wrong, the GeForce RTX 3070 and maybe even the RTX 3060 Ti would hardly have any real competition and at AMD there would be a gap of almost 25% between the RX 6800 and RX 6700 XT. Not a good idea, as it seems, especially since one expects a RRP of about 400 euro for a GeForce RTX 3060 Ti. If there is no pressure from AMD, it should be difficult to keep this price as low as it is. Unfortunately. But of course all this also has the usual speculoos taste. Let us be surprised for once! Perhaps everything comes nevertheless differently, than Jensen thinks, who placed the coming RTX 3060 Ti surely very consciously in such a way and still holds back the RTX 3060.

**Winjer** · 08-12-20, 11:41

Largest Swiss Retailer Digitec to Receive a Grand Total of 35 AMD RX 6900 XT Graphics Cards for Launch

AMD's launch of their top of the line RX 6900 XT graphics card seems that it will have even less availability than the company's high-end RX 6800 and RX 6800 XT graphics card. This isn't surprising; the RX 6900 XT is a 590 mm² beats of a GPU with all of its execution units enabled - that's a lot of die space to harvest without a single silicon fault, no matter how good TSMC's 7 nm manufacturing process really is. Stock will be scarce, and likely will be scarce throughout the lifetime of the product, especially with the clogged, unmet, existing demand for high performance GPUs from a world population that has turned to gaming as a solace in times of quarantine.

Digitec, the largest Swiss retailer (serving a population of 8.5 million people), is only receiving 35 RX 6900 XT graphics cards for launch. We don't know, of course, what exactly is the Swiss demand for high-performance graphics cards, but it being one of the world's wealthiest countries (when it comes to its population's average income) it's expected to be higher than other countries with comparable population but lower income. As a result, the retailer isn't even putting the cards up for sale as they normally would; instead, there's a sweepstakes of sorts where 35 random users that opt-in for the event will receive a code that allows them to purchase the graphics card for its retail price of $999. An interesting solution, albeit of course, it just signals the dimension of the cards' availability issues.

Está complicado...

**LPC** · 08-12-20, 13:49

Post Original de Winjer

Largest Swiss Retailer Digitec to Receive a Grand Total of 35 AMD RX 6900 XT Graphics Cards for Launch

Está complicado...

Boas!
Já estava por tudo... já até estou a ver se compro uma dessas na Europa, mas não me devo safar...

Cumprimentos,

LPC

**Winjer** · 08-12-20, 14:02

Chegaram as reviews das RX 3090

**MTPS** · 08-12-20, 15:28

Belo monstro! Haja dinheiro...e stock...

**Winjer** · 09-12-20, 19:14

O pessoal da AMD mudou de ideias e vão continuar a fazer as 6800 de referência, durante mais tempo.

We've just officially extended the reference design builds indefinitely due to popular demand. Thanks for the feedback. — Scott Herkelman (@sherkelman) December 9, 2020

**MTPS** · 09-12-20, 19:37

Boa noiticia...quero a reference...

**Mystic_red** · 09-12-20, 23:11

Quero a 6800. Mas porra, os preços das gráficas atualmente são para gente doida. 600 paus uma 6800, espera por ela. Quando bater nos 400 conversamos. Antes de cá entrar a 6800 ainda dou o salto de 1TB para 2TB no ssd, assim como um upgrade na ram para 32GB... acho que os dois upgrades juntos não pagam uma 6800 ao preço atual!

**LPC** · 10-12-20, 01:42

Post Original de MTPS

Boa noiticia...quero a reference...

Boas!

X2... ou a Reference ou então a Red Dragon...

Cumprimentos,

LPC

**Winjer** · 16-12-20, 19:06

Foi lançado a versão 1.4 do Quake 2 RTX, com suporte para as extensões ray-tracing do Vulkan.

Por isso, este jogo também corre nas RX 6800. Este é um dos poucos que é full path tracing.

Aqui ficam uns benchs.

https://www.pcgameshardware.de/Raytr...acing-1363692/

**Winjer** · 06-01-21, 11:55

**Winjer** · 23-01-21, 20:24

AMD is Allegedly Preparing Navi 31 GPU with Dual 80 CU Chiplet Design

AMD is about to enter the world of chiplets with its upcoming GPUs, just like it has been doing so with the Zen generation of processors. Having launched a Radeon RX 6000 series lineup based on Navi 21 and Navi 22, the company is seemingly not stopping there. To remain competitive, it needs to be in the constant process of innovation and development, which is reportedly true once again. According to the current rumors, AMD is working on an RDNA 3 GPU design based on chiplets. The chiplet design is supposed to feature two 80 Compute Unit (CU) dies, just like the ones found inside the Radeon RX 6900 XT graphics card.

Having two 80 CU dies would bring the total core number to exactly 10240 cores (two times 5120 cores on Navi 21 die). Combined with the RDNA 3 architecture, which brings better perf-per-watt compared to the last generation uArch, Navi 31 GPU is going to be a compute monster. It isn't exactly clear whatever we are supposed to get this graphics card, however, it may be coming at the end of this year or the beginning of the following year 2022.

**LPC** · 23-01-21, 23:55

Post Original de Winjer

AMD is Allegedly Preparing Navi 31 GPU with Dual 80 CU Chiplet Design

Boas!
Sim parece-me o passo lógico para eles...

Alias o Cache Infinito que eles estão a usar nestas novas, irá ser muito importante para fazer a ligação interna entre os chiplets...

Mas para é preciso.... reduzir mais um pouco o processo de fabrico, meter 2 dies num só monobloco irá concerteza ter desafios de engenharia (apesar de em teoria ser simples).

Depois a questão do funcionamento é que terá que ser bem vista... terá que funcionar a nível de hardware e ser tipo AI para saber dividir as cargas pelos dies...

Cumprimentos,

LPC

**Winjer** · 08-02-21, 13:10

Por falar no diabo...
Supostamente será lançado na primavera.
AMD finally ready to launch its DLSS competitor?

A new report puts a possible timeframe on the FidelityFX Super Resolution technology launch.

It has been more than two years since NVIDIA introduced its Deep Learning Super Sampling technology. At first, it was mocked for how blurry the games looked liked with the DLSS enabled, but the second revision of the technology has significantly improved the quality. At this point, most users would rather use DLSS and sacrifice almost indistinguishable image quality in favor of a much higher framerate. Some games such as Cyberpunk 2077 are almost unplayable at 4K with ray tracing enabled, unless DLSS technology is enabled.At Radeon RX 6000 series launch AMD made a promise it will launch its own version of the AI-powered super-resolution technology. The company has not provided any details whatsoever on how would it work or when exactly should we expect it to be available. AMD has only hinted it could be based on Microsoft DirectML technology, but at the same time AMD revealed that its DLSS competitor would launch as a part of the FidelityFX technology package. Thus, it is unclear if AMD has plans to stick to DirectML and push the gaming market into a single standard, or rather create their own implementation that will require collaboration from game studios. The slide from AMD does however suggest it might be the latter option:

AMD FidelityFX Super Resolution, Source: AMD
According to Prohardvare, AMD now plans to launch its big driver update somewhere in this spring. This means that at the earliest we might hear more about the FFSS in March. The report claims that the big driver update would provide an update on two of the resolution-based technologies: FidelityFX SR and Radeon Boost.
AMD Radeon Boost does not utilize machine learning algorithms to boost framerate. Instead, the technology dynamically changes the resolution during faster-paced gameplay. What this means is that when there is a lot of movement the resolution will be adjusted and the performance will improve. This technology is relatively an easy way to increase framerate, but at the same time, it may lead to inconsistent performance during gameplay.
A new implementation of Radeon Boost that would launch with the new driver is said to offer a higher quality image by dynamically adjusting the resolution only on the fast-moving parts on the screen. By comparison, the current implementation affects the whole frame. This would, for instance, solve the problem of lower quality weaponry in first-person shooters during fast-paced shooting.
An alternative was developed by Sapphire, which is called TriXX Boost. Unlike Radeon Boost, the resolution change does not happen dynamically. Instead, the application creates a custom resolution for the system and uses Radeon Image Sharpening technology to increase the quality of the image. Therefore it produces a more consistent performance increase but at the cost of lower image quality which is then artificially improved with sharpening filters, not AI algorithm.