AMD Radeon RX 6800 XT Roundup: ASRock, Asus, and Sapphire Reviewed

AMD Radeon RX 6800 XT Roundup
(Image credit: Tom's Hardware)

AMD launched the Radeon RX 6800 XT and RX 6800 on November 23, 2020. The first of the new RDNA2 architecture graphics cards had plenty to offer, ranking near the top of our GPU benchmarks hierarchy and earning a place on our list of the best graphics cards. AMD does particularly well with games that don't support ray tracing. In such cases, there are quite a few games where the 6800 XT leads the (theoretically) more expensive RTX 3080, though enabling ray tracing or DLSS quickly turns the tables. The biggest problem, as we've seen with all of the recent GPU launches, is actually finding one in stock. Now we're looking at three third-party custom cards, from ASRock, Asus, and Sapphire, to see what they bring to the table.

The core features and RDNA2 architecture are all unchanged, so the main differences between the cards will be in clock speeds, cooler designs, and aesthetics. There are also a few third-party add-ons, in the way of software, that might sway your purchasing decision. But let's be real: Finding any of these cards in stock can be an exercise in futility, and with the recent surge in cryptocurrency mining, it could be months before supply is anywhere close to matching demand. In other words, if you want an RX 6800 XT as soon as possible, the brand and model of card will be far less of a consideration than whatever you can actually lay your grubby little mitts on.

The good news is that performance across all of the tested RX 6800 XT cards is very close. At factory stock settings, the speediest of the cards we've tested is only 2-3 percent faster than the reference RX 6800 XT. Between the three custom cards, the performance deltas are even smaller, to the point of being effectively non-existent. But that doesn't mean the cards are all equal, as the cooling designs and other elements come into play. Here's a quick overview of the specs before we get into the individual card analysis and benchmark results. 

Swipe to scroll horizontally
AMD Radeon RX 6800 XT Specifications
Header Cell - Column 0 ASRock Taichi RX 6800 XTAsus ROG Strix LC RX 6800 XTSapphire Nitro+ RX 6800 XTReference RX 6800 XT
ArchitectureNavi 21Navi 21Navi 21Navi 21
Process TechnologyTSMC N7TSMC N7TSMC N7TSMC N7
Transistors (Billion)10.310.310.310.3
Die size (mm^2)251251251251
SMs / CUs72727272
GPU Cores4608460846084608
Ray Accelerators72727272
Boost Clock (MHz)2360236023602250
VRAM Speed (Gbps)16161616
VRAM (GB)16161616
VRAM Bus Width256256256256
ROPs128128128128
TMUs288288288288
TFLOPS FP32 (Boost)21.721.721.720.7
Bandwidth (GBps)512512512512
TDP (watts)350?350?350300
Pricing$829 $899 ($1,080)$769 ($999)$649

Let's first address the elephant in the room: The pricing is either fantasy land or, in the case of actual 'street' pricing, egregious. Theoretically, the Sapphire Nitro+ initially launched at $769, but of course, it was sold out — just like every other GPU. Newegg currently lists it at $999, and it’s still out of stock. The Asus Strix LC is a similar story, with a launch price of $899 but a current Asus store price of $1,080, and it's also sold out. ASRock gave a launch price of $829, but retail prices are much higher than that, and naturally, the reference AMD RX 6800 XT can't be had for anything close to $649.

Beyond price, the only difference in specs is the TDP. Sapphire lists 350W, while Asus and ASRock don't give any value. We put in 350W with a question mark based on our testing. All three AIB cards have the same 2360 MHz Boost Clock, which they can exceed in some cases. That's where the cooling solutions come into play.

That's it for the introduction. Let's get to the individual cards, and we'll dig into the finer points of each one, including any extra features that can help it stand out. We're going to dispense with actual scores on these cards, mostly because they all feel like ghost launches. Yes, they technically went on sale, but both pricing and availability are so limited that we don't know where they'll really land. They're all more or less equal, depending on your wants and needs.

MORE: Best Graphics Cards

MORE: Desktop GPU Performance Hierarchy Table

MORE: All Graphics Content

Jarred Walton

Jarred Walton is a senior editor at Tom's Hardware focusing on everything GPU. He has been working as a tech journalist since 2004, writing for AnandTech, Maximum PC, and PC Gamer. From the first S3 Virge '3D decelerators' to today's GPUs, Jarred keeps up with all the latest graphics trends and is the one to ask about game performance.

  • Makaveli
    Canadian pricing for these cards. Asrock from newegg.ca and the last two from canada computers.

    ASRock Radeon RX 6800 XT Taichi Gaming $1499

    SAPPHIRE NITRO+ Radeon RX 6800 XT $1149

    ASUS ROG STRIX LC Radeon RX 6800 XT $1299

    And for the money Asus is asking for this card, they could have alteast sleeved those fan cables its mess. For $1000+ asking price kinda of a slap in the face.
    Reply
  • Jobeker
    Your testing methodology is flawed.
    The memory should not be overclocked in these cards.
    These cards show much better results once you push the GPU to the max & leave the memory at stock.
    It seems that you & all the other testers I have come to respect in almost 2 decades of reading tests & benchmarks , have become fixated on maxing the mem to the point of "no crush" without even comparing the results.
    I hope you still have at least one of the cards at hand to make 1 more test, you will see that stock mem@2000mhz + GPU@2600mhz gets much better results than mem@2140mhz + GPU@2600mhz .

    I am active on a different language forum & a local system builder/fine tuner corroborated these results with several different 6800XT cards.
    ( he is the one that Identified this issue , I don't own such a card at the moment)
    He is now testing a 6900XT .

    As a bonus , once you leave the memory at stock you get a few extra watts for higher gpu oc.

    I don't care that much about the actual value of the specific cards ( definitely not at current pricing ) it is however very important for me to make sure you testing methodology isn't flawed.

    I seriously hope you still have one of the cards for one more test .
    Reply
  • chalabam
    Nice cards. Unfortunetely, they are useless for Tensorflow.

    What a wasted potential.
    Reply
  • JarredWaltonGPU
    Jobeker said:
    Your testing methodology is flawed.
    The memory should not be overclocked in these cards.
    These cards show much better results once you push the GPU to the max & leave the memory at stock.
    It seems that you & all the other testers I have come to respect in almost 2 decades of reading tests & benchmarks , have become fixated on maxing the mem to the point of "no crush" without even comparing the results.
    I hope you still have at least one of the cards at hand to make 1 more test, you will see that stock mem@2000mhz + GPU@2600mhz gets much better results than mem@2140mhz + GPU@2600mhz .

    I am active on a different language forum & a local system builder/fine tuner corroborated these results with several different 6800XT cards.
    ( he is the one that Identified this issue , I don't own such a card at the moment)
    He is now testing a 6900XT .

    As a bonus , once you leave the memory at stock you get a few extra watts for higher gpu oc.

    I don't care that much about the actual value of the specific cards ( definitely not at current pricing ) it is however very important for me to make sure you testing methodology isn't flawed.

    I seriously hope you still have one of the cards for one more test .
    Define "much better results" for me, please. The overclocking is not supposed to be the major focus, because silicon lottery and other elements come into play. Frankly, I wouldn't bother overclocking most GPUs -- it's just not enough of a gain to warrant the added power and potential stress on the hardware. Anyway, a 7.5% memory OC isn't much, and neither is the 3-7% increase in performance I measured with the 'max' OC I achieved.

    Dropping the memory OC and trying for a slightly higher core OC is totally within the parameters of what can be done, and may improve performance more than what I've shown. More effort on tuning voltages, fan speeds, etc. could also improve performance. Without physically modding the cards, though, I strongly doubt you'll see more than a 5% improvement over what I achieved, which is a 10% potential total improvement. In practice, I'm sure it would be far less than that -- probably only a 1-2% difference from my max core + max RAM OC results. And there's a very good chance that, despite what you're positing, overclocking the memory actually does improve performance.

    Let me give you just one example, because based on this I see no reason to bother retesting anything more.


    SettingAsus Strix LC RX 6800 XT OCAsus Strix LC RX 6800 XT OC StockRAMAsus Strix LC RX 6800 XTMetro Exodus1080p Ultra
    125.09
    123.2
    119.46

    1440p Ultra
    105.43
    103.69
    99.77

    4k Ultra
    69.44
    68.52
    64.81

    So, in at least one game, using the same card in all three cases (the Asus Strix LC), dropping the RAM OC but leaving the GPU at the same 2600 MHz setting reduced performance by 1-2%. So much for my "flawed" methodology.
    Reply
  • shady_021
    The only thing I see here is all 6800 XT are equal or slightly ahead of a 6900 XT... so except for the price difference what's the point of having a 6900XT?
    Reply
  • JarredWaltonGPU
    shady_021 said:
    The only thing I see here is all 6800 XT are equal or slightly ahead of a 6900 XT... so except for the price difference what's the point of having a 6900XT?
    The overclocked RX 6800 XT cards match the stock RX 6900 XT. If I overclocked the 6900 XT as well, it comes out a bit ahead again.

    This is something I pointed out in the RX 6900 XT review: it has identical clocks, TDP, and VRAM, so the only difference is 80 CUs instead of 72 CUs. That means at most an 11% advantage for the 6900 XT, but because of VRAM being the same it is usually more like 4-5%. So it's not really worth the extra $350, except none of the 6800 XT cards cost $649, and none of the 6900 XT cards cost $999 -- and in fact, neither of those really exists in any meaningful quantity AFAICT. Tens of thousands of cards worldwide perhaps, but that's a drop in the proverbial bucket.
    Reply
  • Jobeker
    JarredWaltonGPU , Thank you for taking interest in my comment & performing the relevant tests.
    I've ( we actually ) been fed with a lot of information in the past 2 weeks that led me to believe the claim I made .
    I apologize that it came out rude.
    I trust your findings better .
    Thank you.
    Reply
  • HC1Gunner
    Admin said:
    We've rounded up multiple Radeon RX 6800 XT cards to see how the various models stack up. Higher factory overclocks, liquid cooling hybrids, massive coolers, and increased pricing are the general trend while GPUs continue to be in short supply.

    AMD Radeon RX 6800 XT Roundup: ASRock, Asus, and Sapphire Reviewed : Read more
    Question, how is this a round up, when other manufactures like Gigabyte weren't included?
    Kind of stupid to do these reviews, when you can't find any of these cards for retail sale.
    Reply
  • pbergonzi
    Admin said:
    We've rounded up multiple Radeon RX 6800 XT cards to see how the various models stack up. Higher factory overclocks, liquid cooling hybrids, massive coolers, and increased pricing are the general trend while GPUs continue to be in short supply.

    AMD Radeon RX 6800 XT Roundup: ASRock, Asus, and Sapphire Reviewed : Read more
    Thank you for your thoughtful testing and article.
    Reply
  • JarredWaltonGPU
    HC1Gunner said:
    Question, how is this a round up, when other manufactures like Gigabyte weren't included?
    Kind of stupid to do these reviews, when you can't find any of these cards for retail sale.
    A roundup isn't every card or manufacturer possible; it's a roundup of the cards we were sent. So rather than three individual 6800 XT reviews, you get a roundup of three cards plus the reference model.
    Reply