AMD RX 6900 XT Reference: Review, Mining and Test in Games

In this review we will look at the 16 GB version of the AMD 6900 XT Reference video card.

The card is equipped with 2×8 Pin connectors and consumes from 170 Watt to 300 Watt depending on overclocking, miner and algorithm.
Specifications

Specifications	AMD 6900 XT Reference
Video memory	GDDR6 16GB
CPU frequency	Frequency: 2250 MHz
Energy consumption	From 170 Watt to 300 Watt
Power connectors	2×8 Pin
Noise	~44 dB
Dimensions (mm)	267 x 115
Official site	AMD

Internal structure (characteristics)

With the RX 6800 XT, AMD has a very competitive product that can really challenge the RTX 3080 with its $650 MSRP.

AMD has set new goals for itself with the release of the RX 6900 XT; the company claims that this new flagship will offer similar performance to Nvidia's RTX 3090, but the card will be several hundred dollars cheaper.

Of course, we need to find out how true these claims are, so today we'll take a look at the performance of the RX 6900 XT.

Although the RX 6900 XT is a new model, it shares many similarities with the RX 6800 XT. The card is still manufactured on TSMC's 7nm process and uses the same 519mm² Navi 21 GPU.

The key difference is that the RX 6900 XT's silicon is fully populated, with 80 Compute Units (CUs) compared to the 6800 XT's 72. This means the new flagship has an additional 512 stream processors for a total of 5,120 shaders.

The nominal clock speed of these two RDNA 2 GPUs is also identical. This means that the RX 6900 XT has a game clock of 2015 MHz and a boost clock of up to 2250 MHz. For comparison, frequency: RX 5700 XT game clock 1755 MHz.

AMD has also increased the memory capacity, putting 16GB of GDDR6 memory in every RX 6000 model (for now).

The memory is clocked at 16 Gbps and operates through a 256-bit memory interface with a total bandwidth of 512 GB/s.

However, AMD has also implemented a 128MB Infinity Cache on the Navi 21 GPU to significantly increase effective memory bandwidth without excessive power consumption.

Speaking of power, the RX 6900 XT's TBP is 300W, which is the same as the RX 6800 XT despite having 512 more cores.

Sapphire Toxic Radeon RX 6900 XT Limited Edition (16 GB) video card review

Research object : commercially available 3D graphics accelerator (video card) Sapphire Toxic Radeon RX 6900 XT Limited Edition 16 GB 256-bit GDDR6.

Briefly about the main thing

Friends, I am a new author Stanislav Andreevich (many readers of Andrei Vorobyov’s video card reviews will be able to guess who I am). I have been present in video card reviews for 2 or even 3 months now, but in the form of a certain hero in the videos accompanying the articles in this section. The method of creating video card reviews, which has been used here in recent years, does not provide for extensive authorial text, well, except in the conclusions, so perhaps many readers will not notice the change in authorship until they see the end of the article. However, I will still try to add my thoughts into the reviews about what is happening now with video cards and with the personal computer market as a whole. Of course, without going far from the topic of the current device. Well, actually, this is where I’ll finish my introduction, and then the usual tests and conclusions.

Yes, as usual, we bring to the attention of our readers summary charts of the performance of all modern families of video cards. It is clear that somewhere among them is the accelerator being considered today. We evaluate all this on a scale of five gradations (subjectively, of course). And one more thing: we decided to leave only the latest generation of GPU accelerators from both AMD and Nvidia for analyzing the competition between video cards, so as not to constantly clutter the diagrams with the results of 28 video cards. Nowadays there are many times fewer of them. And therefore, for the last time in the summary diagrams below there are video cards of the GeForce RTX 20, GeForce GTX 16 and Radeon RX 5000 families.

Games without ray tracing (classic rasterization)

Yes, guys, we have learned well that the Radeon RX 6900 XT is a direct competitor to the GeForce RTX 3090, if we are talking about classic games without RayTracing (we are not talking about mining at all now, only a little about it at the end of the material), which means , that you can play very decently (and even comfortably) in 4K resolution, subject to maximum graphics settings in the game settings. It is clear that in lower resolutions there will be the same excellent gameplay, or even faster. Some games will work in 8K resolution (well, how they will: I mean that you can play passably even on such a huge thing as a TV with a diagonal of 65 inches or more with a resolution of 7680x4320), but for them you definitely need to enable AMD FSR ( AMD Fidelity Super Resolution). If the game does not support this technology, then everything will not be very good with 8K.

The Sapphire card is slightly faster than the AMD reference card: on average by 5%, but at 4K the difference reaches 6 or even 7 percent.

Games using ray tracing

And here, men, axes are in sheaths, and sabers are in stumps. Alas, our Radeons do not digest traces well. The drop in FPS can be one and a half, or even two, and sometimes even two and a half times relative to performance without RT enabled. Why? Volumes of essays have already been published, where it is described in detail that the GeForce RTX has separate blocks for RT calculations, but the Radeon RX 6000 does not, there are general-purpose cores doing the tracing, and it is impossible to simultaneously calculate the scene and run the rays. That's why things are like this. Therefore, if we are talking about the work of the Radeon RX 6900 XT with ray tracing, then there is no talk of any GeForce RTX 3090 as a competitor. Even the GeForce RTX 3070 is sometimes faster, and, alas, the Radeon RX 6900 XT falls to the embarrassing level of the GeForce RTX 3060 Ti. But if GeForce RTX cards have their own dope in the form of DLSS, then Radeon RX already has a similar one in the form of FSR, there’s just one problem: games with FSR make the cat cry. But that's it for now. Let's see further.

Card characteristics

The Sapphire Technology company (Sapphire trademark) was founded in 2001 in Hong Kong as a subsidiary of the largest concern for the production of PC components - PC Partner. Focused on producing products based on ATI cores (graphics processors) (later part of AMD). Headquarters are in Hong Kong, production is in China. The largest manufacturer of Radeon series accelerators. It also produces mini-PCs and other products.

Sapphire Toxic Radeon RX 6900 XT Limited Edition 16GB 256-bit GDDR6
Parameter	Meaning	Nominal value (reference)
GPU	Radeon RX 6900 XT (Navi 21)
Interface	PCI Express x16 4.0
GPU operating frequency (ROPs), MHz	BIOS 1: 2365(Boost)—2587(Max) BIOS 2: 2365(Boost)—2573(Max)	2250(Boost)—2470(Max)
Memory operating frequency (physical (effective)), MHz	4000 (16000)	4000 (16000)
Memory bus width, bits	256
Number of computational units in the GPU	80
Number of operations (ALU/CUDA) in block	64
Total number of ALU/CUDA blocks	5120
Number of texturing units (BLF/TLF/ANIS)	320
Number of rasterization units (ROP)	128
Number of Ray Tracing blocks	80
Number of tensor blocks	—
Dimensions, mm (card itself / remote radiator SVO)	270×130×45 / 394x120x53	270×110×55
Number of slots in the system unit occupied by a video card	3	3
PCB color	black	black
Peak power consumption in 3D, W (BIOS1/BIOS2)	293	265
Power consumption in 2D mode, W	25	25
Energy consumption in sleep mode, W	4	4
Noise level in 3D (maximum load), dBA (BIOS1/BIOS2)	27,6/23,2	25,5
Noise level in 2D (video viewing), dBA	18,0	18,0
Noise level in 2D (idle), dBA	18,0	18,0
Video outputs	1×HDMI 2.1, 3×DisplayPort 1.4a	1×HDMI 2.1, 2×DisplayPort 1.4a, 1×USB Type-C (USB 3.2 Gen2)
Multiprocessing support	no data
Maximum number of receivers/monitors for simultaneous image output	4	4 (including output via USB Type-C
Power: 8-pin connectors	2	2
Power: 6-pin connectors	1	0
Maximum resolution/frequency, DisplayPort	3840× [email protected] Hz, 7680× [email protected] Hz
Maximum resolution/frequency, HDMI	3840× [email protected] Hz, 7680× [email protected] Hz
Sapphire Card Retail Offers	find out the price

Memory

The card has 16 GB of GDDR6 SDRAM memory, located in 8 16 Gbit chips on the front side of the PCB. Samsung memory chips (GDDR6, K4Z80325BC-HC16) are designed for a nominal nominal operating frequency of 4000 (16000) MHz.

Features of the card and comparison with the reference AMD Radeon RX 6900 XT 16 GB

Sapphire Toxic Radeon RX 6900 XT Limited Edition (16 GB)	AMD Radeon RX 6900 XT (16 GB)
front view

back view

The total number of power phases for the reference Radeon RX 6900 XT card is 16. The Sapphire card has 18.

The core power supply circuit is marked in green, and the memory in red. We see a 16-phase PWM controller XDPE132G5D (Infineon), which in this case controls 15 phases of GPU power supply (the controller itself is located on the reverse side of the PCB).

Also on the reverse side of the board there is an IR35217 PWM controller (International Rectifier, which is now also part of Infineon). It controls three phases of power supply to memory chips.

The power converter, traditionally for all modern video cards, uses DrMOS transistor assemblies (the same IR/Infineon) - in this case TDA21472 for memory chips (maximum current 70 A) and TDA21490 for the graphics core (maximum current 90 A).

The PCB itself is very complex, with 14 layers, including 4 copper layers to reduce interference and improve cooling. Based on the video outputs, we see the usual set of 1 HDMI version 2.1 + 3 DP version 1.4a.

The backlight control is given to the Nuvoton NUC029 controller (it also performs monitoring).

The board (at the tail end) has its own ARGB connector for connecting a cable to synchronize the backlight with the motherboard (however, the cable itself is not included in the package).

A typical scheme with two BIOS versions is used, so a switch is installed at the top end:

The Performance/Silent versions differ mainly in fan operating curves, as well as consumption limits (there is no particular difference in core frequency ranges). By default, the switch is set to position “3,” allowing Trixx’s proprietary utility to control the selection of a BIOS copy to boot.

The standard memory frequencies in both BIOS versions are equal to the reference values. The core frequency boost value in both BIOS versions is significantly higher than the reference analogue (about +7%), which on average gives a performance increase of 5%.

Manual overclocking was tested using the standard AMD utility included in the driver control panel. At the same time, I placed the remote radiator on a tripod near the open door to the balcony (the air temperature outside was about zero).

At the maximum possible frequencies, provided that stable operation is maintained, of +125 MHz for the core and +650 MHz for the memory, maximum frequencies of 2915/17120 MHz were obtained. At the same time, the performance increase relative to the reference card was slightly more than 6%.

When the load on the GPU is low, the fans (all of them, both on the card itself and on the external radiator) stop, regardless of the selected mode.

Power is supplied through three connectors (two 8-pin and one 6-pin), near which there are fuses (a very rare and valuable feature!).

The card's operation is controlled using the proprietary Sapphire Trixx utility.

Main information tab, but no frequency control

Switching between BIOS chips with different operating modes can be done not only with a switch on the end of the card, but also in Sapphire Trixx.

The most interesting tab is Trixx Boost. With its help, you can enable rendering of 3D graphics at a lower resolution than what you set in the game, so that the card will produce a higher speed, and at the end it is stretched to... oh, that is, the resulting image is stretched to the current resolution ( this is cleverly called upscaling). In previous reviews of Sapphire cards, we already wrote about this know-how. In essence, this is a rough analogue of AMD FSR, but according to the personal method of Sapphire specialists.

In games running via the DX9, DX12 or Vulkan API, you can enable proprietary Radeon Image Sharpening technology to minimize quality loss during upscaling. Of course, this will require some costs, like any post-processing, and the speed will drop by 3%-5%, but “the game is worth the candle”: upscaling can give a performance increase of about 15%-30% without visual changes in quality.

Each user can decide for himself whether he needs it or not, whether he agrees to some degradation in picture quality or not. A universal calculation of performance gains depending on Resolution Scale is impossible; every game is different. Well, by the way, this is a good time for games where there is no AMD FSR yet; you can increase FPS through Trixx Boost.

The Trixx utility has several more features.

Small overclocking panel: enabling Toxic Boost will increase the boost frequency by 3%, but the maximum will remain the same; overall gives about 2.5% performance gain

Fan system testing panel; after testing, information about their status will be displayed

What utility for setting up the operation of a card can do without monitoring?

Heating and cooling

Well, that’s the main feature of this accelerator; it’s not for nothing that it arrived in such a huge box. We have a card with a combined cooling system: some are cooled by a fan on the card itself, and some are cooled through a “water tank”, that is, a cooling system.

And yes, we see that such a powerful thing as a liquid cooling system is used in this case to cool the graphics processor.

The copper base of the dropsy pump is clearly visible, which is pressed against the GPU. And for memory chips and power converters there is a special black heatsink, which from the side of the removed video card is visible as a large frame with thermal pads glued to it.

The back plate is made of aluminum with electrical protection coating and serves as a protection element for the PCB.

To cool that same black radiator, the card has a large ∅95 mm fan with special hybrid blades designed to reduce noise as much as possible with good airflow efficiency.

The remote radiator of the LSS has three ∅120 mm fans with transparent blades equipped with ARGB lighting.

Fans stop (both on the card itself and on the radiator) under low load when the GPU temperature drops below 53-55 degrees. Of course, the CO becomes silent in this case. When the PC starts, the fans work, then turn off before the OS loads. Below is a video on this topic. Fans turn off when idle with any BIOS option.

Temperature monitoring using MSI Afterburner utility:

BIOS 1 (Performance):

After a 2-hour run under load, the maximum core temperature did not exceed 52 °C, which is simply a phenomenal result for video cards of this level! The maximum power was recorded at 293W and the maximum heating was observed in the center of the PCB and the main source of heating is the GPU. Memory chips heated up to 62 degrees.

Below is a 9-minute warm-up of the card, accelerated by 50 times.

With the manual overclocking described above, the card's operating parameters remained almost unchanged, the maximum consumption increased to 299 W.

BIOS 2 (Silent):

In this mode, the core heat was slightly less (51 °C), and the fans rotated more slowly. The heating of the memory chips did not exceed 68 degrees. We can conclude that the fan setting in BIOS 1 is clearly excessive, just low rotation speeds are enough, and the heating will still be quite low, but there will be no noise at all!

By the way, here I will simply repeat a paragraph from the previous review. A very important point should be especially noted about “certain” sensors that report heating of about 100 or even higher degrees. These sensors can be called either Hot Spot or Junction.

The temperature of these sensors can differ dramatically from the generally accepted indicator of GPU heating temperature. The fact is that AMD installs several sensors on its processors, and Hot Spot (Junction) means the maximum heating point of the graphics core. It can reach 110°C , and it's safe ! The fact is that the driver will continue to increase the operating frequencies of the GPU units until the Hot Spot sensor gets close to this value, having received the maximum that a given GPU crystal can do. Of course, specific instances of chips, as well as types/types of CO, plus GPU loading in a particular game (driver operation) will influence the ability to achieve maximum heating, and therefore maximum operating frequencies. Therefore, these values may differ for different cards, but, I repeat once again: there is no need to be afraid of such temperature values, GPUs tolerate such modes quite normally (the heating limit of silicon is even higher).

Noise

Here I will not be clever, but will simply present a purely noise measurement technique as it is.

The noise measurement technique assumes that the room is soundproofed and muffled, and reverberations are reduced. The system unit in which the noise of video cards is studied does not have fans and is not a source of mechanical noise. The background level of 18 dBA is the noise level in the room and the noise level of the sound level meter itself. Measurements are taken from a distance of 50 cm from the video card at the level of the cooling system.

Measurement modes:

Idle mode in 2D: Internet browser loaded with iXBT.com, Microsoft Word window, a number of Internet communicators
2D mode with movie viewing: SmoothVideo Project (SVP) is used - hardware decoding with insertion of intermediate frames
3D mode with maximum load on the accelerator: FurMark test is used

The noise level rating is as follows:

less than 20 dBA: relatively silent
20 to 25 dBA: very quiet
25 to 30 dBA: quiet
30 to 35 dBA: clearly audible
35 to 40 dBA: loud but tolerable
above 40 dBA: very loud

When idle, BIOS 1 (Performance) and BIOS2 (Silent) were not at all different: in 2D the temperature was no higher than 36 °C, the fans did not work, but the constantly running pump slightly raised the noise level to 22 dBA.

When watching a movie with hardware decoding, nothing changed.

In maximum load mode in 3D, the temperature in BIOS 1 (Performance) reached 52 °C. At the same time, the fans spun up to 1300 rpm, the noise increased to 27.6 dBA: this is quiet. In the video below, the noise was recorded for a couple of seconds every 30 seconds.

BIOS 2 (Silent) mode the temperature reached 51 °C. At the same time, the fans spun up to 940 rpm, the noise increased only to 23.2 dBA: this is very quiet.

Backlight

The backlight on the card is phenomenally beautiful, located on all 4 fans, and the company and series logos on the card itself also light up.

It is possible to synchronize the backlight of the video card with the motherboard; this requires an ARGB cable (5V) with two “female” connectors at the ends. The video above shows everything perfectly.

To control the backlight, the same Sapphire Trixx utility is used. The lighting system is called Nitro Glow.

You can organize very good lighting effects in conjunction with the motherboard and/or case fans, as well as the central processor.

Delivery and packaging

The delivery set, in addition to the traditional user manual, includes a bonus screwdriver with many attachments (in fact, a whole set!)

Test results, configuration

Test bench configuration

Computer based on AMD Ryzen 9 5950X processor (Socket AM4) : Platform: AMD Ryzen 9 5950X processor (overclocked to 4.6 GHz on all cores);
ZhSO Cougar Helor 240;
Asus ROG Crosshair Dark Hero motherboard based on AMD X570 chipset;
RAM TeamGroup T-Force Xtreem ARGB (TF10D48G4000HC18JBK) 32 GB (4x8) DDR4 (4000 MHz);
SSD Intel 760p NVMe 1 TB PCI-E;
hard drive Seagate Barracuda 7200.14 3 TB SATA3;
Seasonic Prime 1300 W Platinum power supply (1300 W);
Thermaltake Level20 XT case;

operating system Windows 10 Pro 64-bit; DirectX 12 (v.21H1);

TV LG 55Nano956 (55″ 8K HDR, HDMI 2.1);

AMD drivers version 21.11.1;

Nvidia drivers version 496.49;

VSync is disabled.

List of testing tools

All gaming tests used the maximum graphics quality in the settings.

Hitman III (IO Interactive/IO Interactive)
Cyberpunk 2077 (Softclub/CD Projekt RED), patch 1.2
Death Stranding (505 Games/Kojima Productions)
Assassin's Creed Valhalla (Ubisoft/Ubisoft)
Watch Dogs: Legion (Ubisoft/Ubisoft)
The Medium (Bloober/Bloober)
Godfall (Gearbox Publishing/Counterplay Games)
Resident Evil Village (Capcom/Capcom)
Far Cry 6 (Ubisoft/Ubisoft)
Metro Exodus (4A Games/Deep Silver/Epic Games)

To calculate the hashrate when mining “ether” (Ethereum/ETH/ETC) and “raven” (Ravencoin/RVN), the T-Rex miner (0.21.06) was used, the average per hour was recorded in two modes:

default (consumption limit reduced to 70%, GPU frequency reduced by 200 MHz, default memory frequency, fans set to manual mode at 70%)
optimization (consumption limit reduced to 70%, GPU frequency reduced by 200 MHz, memory frequency increased by 500-1000 MHz (depending on the card), fans set to manual mode at 80%)

To test the GeForce RTX 3060 (without LHR), we used the same “leaked” driver version 470.05, in which mining protection is disabled; on other versions it is 24/26 MH/s.

Test results in 3D games

Standard test results without hardware ray tracing at 1920x1200, 2560x1440, and 3840x2160 resolutions

Hitman III

Cyberpunk 2077

Death Stranding

Assassin's Creed Valhalla

Watch Dogs: Legion

The Medium Godfall

Resident Evil Village Far Cry 6

Metro Exodus

Test results with hardware ray tracing and/or DLSS/FSR enabled at 1920x1200, 2560x1440 and 3840x2160 resolutions

Cyberpunk 2077, RT

Watch Dogs: Legion, RT

The Medium, RT

The Medium, RT+DLSS / RT+FSR

Resident Evil Village, RT

Far Cry 6, RT

Far Cry 6, RT+FSR

Metro Exodus, RT

iXBT.com rating

The iXBT.com accelerator rating shows us the functionality of video cards relative to each other and is presented in two versions:

iXBT.com rating option without RT enabled

The rating is based on all tests without the use of ray tracing technologies. This rating is normalized by the weakest accelerator from the group of cards - GeForce RTX 3060 (that is, the combination of speed and functions of the GeForce RTX 3060 is taken as 100%). The ratings are based on 14 accelerators that we periodically research as part of the Best Video Card of the Month project. In this case, a group of cards for analysis was selected from the general list, which includes the Radeon RX 6900 XT and its competitors.

The rating is given in total for all three resolutions.

№	Accelerator model	iXBT.com rating	Usefulness rating	price, rub.
01	Sapphire Toxic RX 6900 XT, overclocked to 2915 /17120	210	12	171 000
02	Sapphire Toxic RX 6900 XT 16 GB, 2365—2587/16000	210	12	171 000
03	RX 6900 XT 16 GB, 2015—2470/16000	200	13	158 000
04	RTX 3090 24 GB, 1695—1965/19500	200	7	300 000
05	RX 6800 XT 16 GB, 2015—2401/16000	190	13	148 000
06	RTX 3080 Ti 12 GB, 1665—1965/19000	190	10	186 000
07	RTX 3080 10 GB, 1710—1965/19000	180	12	150 000

I already said earlier that in classic games the RX 6900 XT is just a “cannon”! We see complete and unconditional leadership of the Sapphire card (as well as the Radeon RX 6900 XT in general) in modern games with settings at maximum. Let me emphasize again - without ray tracing.

iXBT.com rating option with RT enabled

The rating is based on 8 tests using ray tracing technology (with Nvidia DLSS and with AMD FSR in the games The Medium and Far Cry 6). This rating is normalized by the weakest accelerator in this group - GeForce RTX 3060 (that is, the combination of speed and functions of the GeForce RTX 3060 is taken as 100%).

The rating is given in total for all three resolutions.

№	Accelerator model	iXBT.com rating	Usefulness rating	price, rub.
01	RTX 3090 24 GB, 1695—1965/19500	220	7	300 000
02	RTX 3080 Ti 12 GB, 1665—1965/19000	210	11	186 000
03	RTX 3080 10 GB, 1710—1965/19000	190	13	150 000
04	Sapphire Toxic RX 6900 XT, overclocked to 2915 /17120	180	11	171 000
05	Sapphire Toxic RX 6900 XT 16 GB, 2365—2587/16000	170	10	171 000
06	RX 6900 XT 16 GB, 2015—2470/16000	170	11	158 000
08	RX 6800 XT 16 GB, 2015—2401/16000	150	10	148 000

Yes, unfortunately, if we use Ray Tracing, then our “cannon” becomes a “Tsar Cannon” and only poses beautifully. Alas, in this case, even a decent overclocking of the Sapphire card did not help it at least catch up with the RTX 3080, so we can no longer say that in such games 4K resolution is completely suitable for the RX 6900 XT. Of course, if we have games with AMD FSR, then yes, FSR will somehow be able to compensate for the drop from ray tracing, and you can play in 4K comfortably. But in others it is a controversial issue.

Usefulness rating

The utility rating of the same cards is obtained if the previous rating is divided by the prices of the corresponding accelerators. To calculate the utility rating, retail prices were used conditionally for November 2022 . Given the Radeon RX 6900 XT's focus on 4K resolution, we've provided ratings calculated only at 3840x2160 resolution (which is why the iXBT.com rating is different from the one above).

Option for utility rating without including RT

№	Accelerator model	Usefulness rating	iXBT.com rating	price, rub.
05	RX 6900 XT 16 GB, 2015—2470/16000	14	218	158 000
06	RX 6800 XT 16 GB, 2015—2401/16000	14	203	148 000
07	Sapphire Toxic RX 6900 XT, overclocked to 2915 /17120	14	234	171 000
09	RTX 3080 10 GB, 1710—1965/19000	13	198	150 000
11	Sapphire Toxic RX 6900 XT 16 GB, 2365—2587/16000	13	224	171 000
13	RTX 3080 Ti 12 GB, 1665—1965/19000	12	217	186 000
14	RTX 3090 24 GB, 1695—1965/19500	8	229	300 000

Today, the realities of life are such that top-end video cards cost much more than 100,000 rubles. Therefore, there is no need to discuss this anymore. At the same time, we observe that the previously made conclusions that the RX 6900 XT is the king in classic games are confirmed in this rating. The Sapphire card took third place only because of the increased cost, and this is logical, because the product is really sophisticated, and the usefulness ratings do not take into account either the features of the card or the delivery package.

Utility rating option with RT included

№	Accelerator model	Usefulness rating	iXBT.com rating	price, rub.
05	RTX 3080 10 GB, 1710—1965/19000	14	205	150 000
06	RTX 3080 Ti 12 GB, 1665—1965/19000	13	235	186 000
07	RX 6900 XT 16 GB, 2015—2470/16000	12	184	158 000
08	Sapphire Toxic RX 6900 XT, overclocked to 2915 /17120	11	196	171 000
09	RX 6800 XT 16 GB, 2015—2401/16000	11	164	148 000
10	Sapphire Toxic RX 6900 XT 16 GB, 2365—2587/16000	11	187	171 000
14	RTX 3090 24 GB, 1695—1965/19500	8	250	300 000

In this case, we again see a logical picture: the RTX 3090 is an outsider due to its simply incredible cost, but the RX 6900 XT has almost caught up with its more expensive brother, and also its competitor from Nvidia, the RTX 3080 Ti. Of course, again, due to the higher price, the Sapphire card is somewhat behind, but I repeat that everyone chooses for themselves: if you have the funds to buy such flagships, then it probably makes sense to carefully choose one or another product based on the sophistication of the SO or kit, and not grab what's cheaper.

Mining testing results (mining, hashrate)

Hashrate, MH/s

In general, tests have long shown that the Radeon RX 6900 XT (like the 6800 XT) has a hashrate slightly higher than the GeForce RTX 3070 (without LHR), while such 3070 cards are almost impossible to find on sale. The RTX 3080 Ti has approximately the same hashrate due to the built-in protection against Ethereum mining (however, given the latest attempts to crack the protection, the RTX 3080 Ti can give a hashrate of up to 75).

Note that Radeon cards do not have any protection against being used for mining; they work as physically as they can.

If we start mining ETH with the card's default settings, we get approximately 60 MH/s with a consumption of 252 W. Heating is within normal limits.

RVN, as you know, is more demanding, so the card already consumes 285 W. Heating is generally at the same level. In this case we have 32.4 MH/s. (Different cryptocurrencies have different hashrates; they cannot be compared with each other; cards can only be compared within the framework of working using the same algorithm.)

Optimizing the settings for mining in our case does not involve strong overclocking of video memory; external airflow for video cards is also mandatory. With a strong reduction in core frequency and voltage, a slight increase in memory frequency and increased cooling, we got 63.4 MH/s in ETH, while GPU consumption dropped to 152 W. The heating has decreased.

For the kawpow (RVN) algorithm, our optimization gives almost nothing, because a strong decrease in PL leads to a drop in the hashrate (in the case of ethash, it was the decrease in PL that greatly saved the day, because the hashrate remained the same with much less consumption). As a result, there is an increase in hashrate, but it is insignificant.

conclusions

Sapphire Toxic Radeon RX 6900 XT Limited Edition (16 GB) is probably the most sophisticated flagship video accelerator I've ever seen. A key feature of the card is the use of liquid cooling, which seriously complicates the installation of a video accelerator in the case, but it ensures very low GPU temperatures under maximum load and very quiet operation of the cooling system. This is a serious competitor to the famous GeForce RTX 3090 (the cost of which has long been above the clouds). All Radeon RX 6000 video cards have a vulnerability in the form of a decrease in FPS when using ray tracing, but games have already begun to come out with support for AMD FSR technology, which helps to partially compensate for this dramatic drop in performance.

Therefore, overall we can still say that the Radeon RX 6900 XT (and the Sapphire card in particular) is very, very suitable for gaming at 4K resolution with graphics settings at maximum. Maybe in a couple of games when RayTracing is enabled, he will have to switch to 2.5K resolution or lower the graphics settings a little, but this is acceptable.

I would also like to remind you of the new AMD technologies accompanying the Radeon RX 6000 family, including support for the HDMI 2.1 standard, which allows you to output 4K images at 120 FPS or 8K using a single cable, support for hardware decoding of video data in AV1 format, Smart Access Memory technology, which can provide a slight increase in performance when the new accelerators work together with Ryzen 5000 processors, as well as with the latest generations of Intel Core processors. It is also worth noting the Radeon Anti-Lag latency reduction technology, useful for esports players.

Reference materials:

Gaming Graphics Card Buyer's Guide
AMD Radeon HD 7xxx/Rx Reference
Guide to Nvidia GeForce GTX 6xx/7xx/9xx/1xxx

In the “Original Design” category, the Sapphire Toxic Radeon RX 6900 XT Limited Edition (16 GB) received the award:

In the “Excellent Delivery” category, the Sapphire Toxic Radeon RX 6900 XT Limited Edition (16 GB) received the award:

We thank Sapphire Russia Sergei Kozhevnikov personally for providing us with the video card for testing.

We thank TeamGroup and Ethnie Lin for providing RAM for the test bench

For the test bench: AMD Ryzen 9 5950X processor provided by AMD , ROG Crosshair Dark Hero motherboard provided by Asus

Equipment

The AMD RX 6900 XT comes in identical packaging to the RX 6800 XT, with a large graphic of the graphics card on the front.

The back of the box is mostly blank except for some technical text in the top left corner.

The 6900 XT also comes with an AMD mouse pad and a Radeon keyboard cap.

Mining on AMD Radeon RX 6900XT: how many hashes does the card produce using other algorithms

On such productive GPUs you can mine not only ether. Speaking of the RX 6900, how many hashes does the card produce for other cryptocurrencies? We have compiled a card hashrate table (RX 6900 XT hashrate) for popular algorithms based on data from whattomine.com. The numbers may be lower than those stated in the reviews, since GPU performance directly depends on the manufacturer and the mining card settings.

Autolykos	110.00 Mh/s	MTP	6.30 Mh/s
CuckooCycle	8.90 h/s	ZelHash	67.00 h/s
Ethash	64.00 Mh/s	KawPow	32.00 Mh/s
Ethash4G	64.00 Mh/s	Zhash	89.00 h/s

Cooling system

The RX 6900 XT uses the same design as the RX 6800 XT, so both look the same.

This means we have a premium aluminum casing that is mostly black with some silver accents.

The card also uses the same cooling system. AMD RX 6900 XT features three 80mm

axial fans. Each cooler features the Radeon "R" logo on the center hub.

The card size is 267 x 115 mm , the thickness of the RX 6900 XT is 2.5 slots . This is exactly the same card as the RX 6800 XT, but the RX 6800 looks the same, but is still a little thinner - it takes up two slots.

The card itself is quite massive, its weight is 1515 grams!

On the front of the card is the Radeon logo, and there's a little surprise here - this logo is actually an RGB zone!

As for the back panel (backplate), it is almost entirely silver, which increases the length of the card. There is a cutout behind the chip itself, but otherwise the backplate is completely blank.

It is also worth noting that the back panel is made of metal and protects the card from damage during installation and operation.

The AMD Radeon RX 6000 series does not support multi-GPU, at least not in the classic sense. DirectX 12 has a multi-GPU feature that works universally regardless of GPU or architecture, but is not used in any serious games.

Note that unlike Nvidia (3080 and 3090), AMD is not using the new power outputs. The card features two standard 8-pin PCIe power connectors on the top edge of the card.

Internal cooling system

The copper plate contacts the GPU and memory, and the aluminum support plate conducts heat away from the VRM.

Both plates are in contact with an aluminum radiator, which is blown by three 90 mm coolers. The card supports the function of stopping coolers during idle time or when power consumption is low.

Having disconnected the radiator grille, we see a clean printed circuit board.

AMD uses a 13-phase GPU VRM design, which is powered by an expensive 16-phase "Infineon XDPE132G5D" VRM controller.

AMD uses International Rectifier TDA21472 DrMOS components on the Radeon RX 6900 XT.

The memory voltage uses a three-phase circuit controlled by an International Rectifier IR35217 controller.

AMD's "Navi 21" uses an external USB 3.2 controller for the USB-C port, supplied by Cypress Semiconductor under the part number " CYPD5137-40LQXI ". This controller provides USB 3.2 Gen 2 data transfer at 10 Gbps.

The GDDR6 chips are manufactured by Samsung, under the number " K4ZAF325BM-HC16 " and are designed to operate at 16 Gbps.

Built on the same TSMC N7 (7 nm) as Navi 10, the Navi 21 GPU has a die area of 519.8 mm² and contains 26.8 billion transistors. The pinkish-red tint we saw on the Navi 10 is gone.

NVIDIA Overclocking Settings

Please note that the power limit expressed as a percentage (%) may differ for different video card models. Not all graphics cards have the same base TDP, so setting the power limit to 80% will not necessarily match the 285W power limit.

How to calculate the power limit for a video card as a percentage?

First, determine the TDP (power consumption) model on your video card manufacturer's page. Example. Follow this formula to calculate the percentage change in power limit:

Reducing the power limit = TDP of the video card - Power limit (from the table below)
Reducing the power limit as a percentage = Reducing the power limit / TDP of the video card * 100

Model	Core frequency delta	Memory frequency delta	Power Limit (W)	Expected Performance (DaggerHashimoto)
RTX 3090	-300	+1000	285	125MH/s
RTX 3080	-150	+900	220	100MH/s
RTX 3070	-500	+1100	130	62 MH/s
RTX 3060 Ti	-500	+1200	130	62 MH/s
RTX 2080 Ti	-200	+1100	150	57 MH/s
RTX 2080 Super	-50	+1000	175	44 MH/s
RTX 2080	-50	+800	155	44 MH/s
RTX 2070 Super	-50	+800	150	44 MH/s
RTX 2070	-50	+800	125	44 MH/s
RTX 2060 Super	-50	+850	125	43 MH/s
RTX 2060	-50	+700	115	29 MH/s
GTX 1080 Ti	+750	185	49 MH/s
GTX 1080	+700	135	35MH/s
GTX 1070 Ti	+500	135	29 MH/s
GTX 1070	+450	115	27 MH/s
GTX 1060 6GB	+900	80	23MH/s

Thermal images

All thermal images were taken under maximum load for a general overview and understanding of the cooling system with a slight overclocking of the core and memory.

By installing a thermal chamber on top, we get the following indicators:

The GPU temperature stays at around 70 degrees - this is not a very good indicator!

Memory at ~54 degrees.

Unfortunately, unlike the RX 6800 XT (and especially the RX 6800), the RX 6900 XT performs much worse in terms of cooling system.

By installing the thermal chamber on the side (on the cooler side), we get the following indicators:

The GPU temperature remains at ~68 degrees.

Overall the map looks very cold!

Tests in MSI Afterburner show terrible results, the GPU temperature is at around 80 (81) degrees:

Unfortunately, AMD did not make the ideal cooling system for such a powerful card. 80 degrees according to MSI Afterburner is already a very bad indicator.

Watch the video from our YouTube channel - Is the video card temperature of 80 degrees normal? What is the normal temperature of the card?

Overclocking 6900 XT and 6800 XT for Ethereum mining

We overclock through Radeon Software following the example of reviewers and miners. Before overclocking, disable MSI Afterburner and do not turn it on again. After overclocking, we record the indicators for the farm, and in order not to lose them, press CTRL + D and add this page to your bookmarks.

Through Radeon Software we specify the following parameters. [How to do it].

6800XT : 1400 core frequency / 881 mv core voltage / 2150 memory frequency
6900XT : 1100 core frequency / 825 mv core voltage / 2088 memory frequency

At the output we get an average of 64 MH/s per card. The difference in the number of cores and texture units does not help the cards, but the older model gets its 64 MH/s at a core frequency 300 MHz lower than that of the RX 6800 XT. Therefore, it will be more energy efficient than the younger model.

How much can you earn

Knowing that this video card works well on the Ethash algorithm (Ether Coin) and knowing its hashrate, you can calculate the payback of this video card, how much it will bring per day, month, year.

To calculate, enter the hashrate of this video card in the “Hashing Power” column, in our case it is 64

“ Pool Fees ” is the pool commission, you don’t have to enter it if you don’t know. The average commission for all pools is 1%.

Next, in the “ Power Usage ” column, you need to enter the power consumption of the video card (farm). In our case it is 200 . We enter this number in this column.

“ Power Cost ” is the price for your electricity (“per socket”). Enter the value in cents. For example, our electricity costs an average of 3 rubles – that’s $0.05. Enter this or another number of yours.

That's all! Click on the “ Calculate ” button and find out how much you will receive from 1 video card per day, week, month, year!

AMD video cards hashrate table.

GPU	ETHash	Zhash	BEAM3	Grin32	MTP	ProgPow	KawPow
RX560	7H/s	7H/s 70W	4.3Mh/s 53W
RX470/570	22-31Mh/s 120W	24H/s 100W	13H/s 110W	0.1H/s 110W	6.7Mh/s 130W	13Mh/s 120W
RX480/580	24-33Mh/s 130W	28H/s 110W	14.7H/s 120W	0.16H/s 110W	1.6Mh/s 120W	7.9Mh/s 140W	14Mh/s 150W
RX590	25-33Mh/s 150W	14Mh/s 150W
RX5500XT	26Mh/s 65W	10.5Mh/s 125W
RX5600XT	42.5 Mh/s 94W	21.3Mh/s 150W
RX5700	48-60Mh/s 140W	23.5H/s 130W	0.35H/s 170W	26MH/s 150W
RX5700XT	48-60MH/s 140W	24.5H/s 130W	0.4H/s 190W	2.8Mh/s 155W	28Mh/s 160W
Vega 56	45-55Mh/s 180W	34H/s 230W	23H/s 160W	0.3H/s 230W	2.5Mh/s 230W	15.5Mh/s 250W	25Mh/s 190W
Vega 64	45-55Mh/s 180W	44h/s 250W	28H/s 180W	0.34H/s 250W	2.8Mh/s 260W	17Mh/s 270W	26Mh/s 190W
Vega 7	100-116Mh/s 230W	49H/s 180W	43H/s 190W	0.5H/s 200W	4.2Mh/s 250W	24.6Mh/s 270W	40Mh/s 220W
RX6600	29Mh/s 55W	13.5Mh/s 70W
RX6600XT	32Mh/s 60W	16.5Mh/s 140W
RX6700XT	45-50Mh/s 130W	67H/s 180W	24Mh/s 200W
RX6800	64Mh/s 170W	90H/s 200W	36Mh/s 200W	32Mh/s 200W
RX6800XT	64MH/s 170W
RX6900XT	64MH/s 170W

The maximum performance of AMD video cards using the Ethash mining algorithm is achieved by changing the video memory timings through firmware and subsequent correct overclocking, as well as downvolting. Read more here.

All video cards with 4GB memory (RX470, RX480, RX570, RX580) in 2021 cannot mine Ethereum due to an increase in the size of the DAG file.

List of mining software for AMD video cards:

Ethash: Phoenix miner, TeamRedMiner
Zhash, Beam: Lolminer, Bminer,
Grin: Bminer, Gringoldminer, acegrinminer, gminer, lolminer
ProgPOW: progpowminer amd, ethminer
KawPow: NBminer, kawpowminer, teamredminer
MTP: teamredminer

I recommend installing the HiveOS operating system on your farms, where it is very convenient to manage your mining equipment, and most importantly, the hashrate on Linux is higher than on Windows. At the top of the text there are links to instructions with detailed mining setup.

Memory and power consumption

The video card is equipped with 16 gigabytes of GDDR6. Supplied from Samsung. The memory marking is “K4ZAF325BM-HC16”.

Card consumption is from 170 to 300 Watt, depending on the miner and overclocking. The card is equipped with 2×8 Pin power connectors.

For a farm of 6 AMD 6900 XT video cards and a connected power system (motherboard, SSD, processor), you will need 2 power supplies of 1000 Watt each!

AMD RX 6700 XT specifications and characteristics.

Name	Index
Core	Navi 22
Technical process, nm	7
Transistors, million	17 200
Core frequency, MHz	2321–2581
Shader unit operating frequency, MHz	2321–2581
Shader blocks	2560
TMU	160
ROP	64
Memory operating frequency (DDR), MHz	2000 (16000)
Memory bus	192-bit
Memory type	GDDR6
Memory	16384
PSP, GB/s	384
Shaders Model	6.5
Fill Rate, Mpix/s	165200
Fill Rate, Mtex/s	413000
DirectX	12 (12_2)
Interface	PCI-E 4.0
Energy consumption, W	230
Date of issue	17.03.2021

Conclusion

AMD 6900 XT Reference is not a bad card, with an excellent build and an average (closer to low) cooling system.

AMD has really come a long way in the quality of card design and power.

Despite the fact that the RX 6800 copes well with loads, the same cannot be said about its older brother 6900 XT.

The Bytwork team does not recommend the AMD 6900 XT reference for purchase. If +- you find cards from vendors at the same price, then we advise you to consider them first, since they will most likely be better suited to the cooling system for the XT version card.

Mining profitability on AMD Radeon RX 6900XT

We figured out what hashrate the Radeon 6900 xt has, but what about the profitability? In the table below you can see how much profit one card will bring per day. Data is presented as of August 2022. Current information can be calculated at whattomine.com.

Cryptocurrency	Income in dollars per 24 hours	Cryptocurrency	Income in dollars per 24 hours
Ethereum(ETH) Ethash	$4.17	Nicehash-CuckooCycle CuckooCycle	$2.14
Nicehash-Ethash Ethash	$3.76	Tent(TENT) Zhash	$2.09
Ravencoin(RVN) KawPow	$3.63	Nicehash-Autolykos Autolykos	$2.14
Flux(FLUX) ZelHash	$3.12	Callisto(CLO) Ethash	$1.92
Nicehash-KawPow KawPow	$3.31	QuarkChain (QKC) Ethash	$1.94
Aeternity(AE) CuckooCycle	$2.12	BitcoinZ(BTCZ) Zhash	$2.25
EthereumClassic(ETC) Etchash	$2.95	EtherGem(EGEM) Ethash	$2.05
Ergo(ERG) Autolykos	$2.32	Etho(ETHO) Ethash	$1.88
Firo(FIRO) MTP	$3.94	Nicehash-Zhash Zhash	$1.73
BitcoinGold(BTG) Zhash	$2.03	BitTubeCash (TUBE) CuckooCycle	$2.36
Metaverse(ETP) Ethash	$2.36	Ellaism(ELLA) Ethash	$1.62

Installation of a full coverage water block

Now let’s get down to the main part of the experiments - installing a water block, getting rid of problems with temperatures and seeing the real potential.

We managed to obtain an EK water block from the Quantum line with a nickel-plated copper base. The marks from the cutter are clearly visible, but all the important areas that directly remove heat are treated better - it will not affect performance.

The black acetal top and silver backplate are responsible for the visual uniqueness. But appearance is the tenth thing. It is much more interesting how the water block will manifest itself in the main thing - cooling. Well, let's get started.

There are no special difficulties in disassembling reference radeons. We sequentially unscrew the screws from the backplate side, and also unscrew the metal crosspiece.

It is curious that the AMD design does not provide any thermal pads on the back side of the printed circuit board, but a metal backplate could be useful in dissipating heat.

Then you can unscrew the plate from the video outputs, or you can leave it - it won’t interfere with the final result. We continue to get rid of screws, unscrewing all available ones.

Now we carefully disconnect the printed circuit board and the radiator; all kinds of thermal interfaces will resist this, but the main thing is not to forget to disconnect the fan wires. The reference cooling system looks quite solid: an evaporation chamber, a massive radiator, a little confusing is the solution with a separate plate for the power subsystem, which transfers heat to the main radiator through small thermal pads.

The easy part is over. Now comes the difficult part: carefully remove traces of the graphite thermal interface from the video chip.

Then everything will be easier. The kit includes thermal pads of different thicknesses - we sort and cut them. In our case, EKWB provided Quantum Vector with detailed instructions with pictures so you can’t go wrong.

There is no need to really sweat over the perfection of right angles - the main thing is that the elements are completely covered.

We cover the chip with our favorite thermal interface (we’ll make do with regular thermal paste) and place it on top of the water block - it’s more convenient. We collect the required number of screws and place them in the seats, then we check the backplate fastening and correct the errors. It is worth tightening evenly, starting from the area near the video chip.

The backplate from EK is also metal, but the Slovenes did not spare thermal pads - we cut them according to the instructions, combine them, and secure them.

All is ready. Beauty and RGB, but the main thing is the temperature mode of operation. A completely different level, this cannot be achieved in the air. Let's start testing and comparisons.

Increased power limits

Initially, there was not even a choice to remain within the factory limits or not - of course, we will increase them. To do this, we will use the More Power tool application - an extremely useful utility, but within the framework of the latest generation Radeons it can only help with increasing the available power consumption and current. The remaining values important for overclocking cannot be changed without physical intervention. Let's set 360 W as the limit on the chip, and increase the permissible current not so much - to 360 A.

Based on the results obtained above, it was expected that we had to make a correction to the values of the maximum and minimum core frequencies: first by 50 MHz to minus, and during further testing they had to be lowered by another 50 MHz.

In the case of a high load on an unheated card, the boost algorithm most likely climbed along the frequency voltage curve into the instability zone, and in the case of operating temperatures, these values were no longer included in the sample of possible ones and the tests passed without errors. As an additional option, the comparison will be the same overclocking, but with the fan speeds fixed at maximum.

In the Time Spy Extreme stress test, for the first three minutes both right options performed similarly in terms of frequencies and FPS, but after that the mark of 110 degrees was reached on the Hot Spot of the video chip and the effective frequency began to decrease.

Without maximum fan speeds, 400 W on the chip was expectedly too much for a reference cooling system. Naturally, 3000 rpm is no longer very suitable for constant use, but it was interesting to look at the behavior in such conditions.

In the resource-intensive Resident Evil Village, the temperature effect is slightly less pronounced, but still noticeable in the final result. Using such a high limit on an ongoing basis, and especially in a closed case, does not seem safe.

On the other hand, what really sets this game apart is its ability to utilize all available power.

In Watch Dogs: Legion, too, the situation looks much more acceptable, but the difference between the maximum fan and auto speeds is visible only in temperatures; it no longer affects the result in FPS.

In most games the picture will be similar. The tests are located in the final block.