Analysis: XFX RX590

RX590


Introduction

Ah, greetings dear reader and welcome to another peoples…review? I’ll admit I have no idea what to call this as it has mutated in to an amalgam of things this time which you will find out as we do a deep dive. I know I’m late with a look at the RX590 what with writing other reviews and articles but 590 reviews are still occasionally trickling out and as most of you know I always like to bring something different and new to the table compared to anywhere else and today will be no different, I always say if you’re going to be late to something you better make damn sure you’re the best. On to the XFX RX590 then, some of you may have heard about the reputation the XFX RX590 has, I am not here to comment or judge on such things, yet. We’ll see what we find first with an objective look at things. So let us find out how the XFX RX590 really does with a fresh perspective as we take it for a spin down the road.


Gallery

This is the FIRST XFX RX590 I started with pictured, we will get to this topic as we push forward, but as things like the cooler and overall look are identical we will use these original images.





First impressions then, there isn’t much to say about the XFX RX590 (I refuse to call it “the Fatboy”) the packaging is quite generic and doesn’t particularly feel premium in any sense of the word which is a little off considering this model is meant to represent the finest XFX has. Inside the card is also poorly protected sitting in just an anti-static bag with a few thin pieces of cardboard around it. Nestled away in all that minimally protective cardboard you’ll find a molex to 6 pin PCI-E adapter and an 8 pin to twin 6 pin PCI-E adapter, both of these seem rather pointless in today’s age.

You won’t find a driver DVD in the box with any accompanying software either which was a surprise instead you will get a little note from XFX telling you to go download the drivers from the XFX site or the AMD site. This you could say is greener for the planet so I don’t mind that however here’s the thing, the driver DVD for Joe Average is essential XFX, if you are trying to be greener any information that might be needed make a PDF document and put it on the driver DVD and stop wasting paper and cardboard the box is about twice the size it needs to be.

As for the look of the XFX RX590 itself, well, let’s take a look at it. One of the first things that should jump out is the picture on the box clearly depicts the top shroud neatly coming down in steps to meet with the rear plate but in reality as you can see this isn’t even close to happening, like there were issues getting the mould to do this or something. Visually this card isn’t pretty at all being clunky, lumpy and very, very cheap looking. I’m not a fan, and I bet you aren’t either.
Lastly you probably can’t see it very well where it is tiny but this card does have a dual vBIOS, the switch is right next to where the PCI-E cables plug in.



The rear IO is decent, 1x HDMI, 3x Displayports, and 1x DVI port. XFX really should have ditched the DVI for a few more HDMI ports or taken away one of the DPs for an extra HDMI as a single HDMI port really isn’t enough, ventilation on the IO is about as good as you can expect. The rear IO isn’t the worst I’ve seen (I’m looking at you Powercolor Red Dragon RX590) but it’s not particularly good either. Average at best.


Laid To Bare

We know the drill here, total strip down and see what lay beneath, so let’s have a look.

Let’s start with the FIRST XFX RX590;



I’ll get straight to the point, this cooler is dreadful on closer inspection, the plastic shroud is flimsy and if you grip the card even a little tight you’ll hear the heatsink fins stress slightly they literally sound like tin foil, the fans are also non-replaceable without totally stripping down the cooling solution, not something I would normally worry about but this RX590 is meant to be the XFX version of a premium variant but it looks and feels anything but that.



The rear plate visually isn’t anywhere near as unpleasant compared to the cooler itself but the rear plate on the card isn’t without issues either and here’s why; there is no cutout directly above the GPU which is just going to trap heat, XFX could have made this plate useful so easily by not drilling so many holes in it and using thermal pads placed on the back of the memory to transfer heat away thus aiding cooling and taking some strain off of the components and primary heatsink. As some of you may be thinking it, no, I wouldn’t also add thermal pads for the inductors or MOSFETs to the back of the card as then you would be in a situation of thermally overloading the rear plate, GDDR gets rather hot you know under load.

Aside from adding some rigidity then this rear plate is absolutely useless and will hinder cooling rather than help. XFX aren’t even courteous enough to have all of the screws for the plate easily accessible nor have they made the plate independent of the primary heatsink to make it easily removable for a quick and easy cooling improvement.

I’ll spare a moment to mention that there was a warranty sticker on one of the mounting screws this is not cool at all preventing users from performing basic maintenance such as dust removal as well as thermal paste and pad replacement. However in the US these stickers are illegal and in the UK and the EU these stickers mean absolutely nothing if the fault has nothing to do with any basic maintenance carried out at which point you can quote your statutory rights. Arguing that point some manufacturers can intentionally be difficult though so I would advise you go and buy a scalpel set which are very cheap, these sets make the removal of warranty stickers like this without any signs of interference no trouble at all, just pop the sticker back on after any maintenance and there is literally no way the manufacturer can prove or know you ever did anything. Here is an informative link for anyone living in the US, EU, or UK regarding warranty stickers.



Removing the heatsink we get a good look at fin density, thermal materials and overall build quality, that plate for the memory looks suspect but we will get to that in a minute. Thermal materials look good at least with solid contact, those copper pipes are pretty heavy duty so no issues there, but that fin density is very poor though and you can note the damage to the heatsink in the bottom right corner of the image, we’ll get to that soon.



Having a closer look of that plate for the memory and, well, how were you planning to make this plate an effective part of the primary heatsink XFX, using hopes and dreams? No thermal epoxy or paste to transfer heat from the plate to the main heatsink, and to add further insult the plate is made of steel or nickel (it’s magnetic) not aluminium. Steel is an extremely poor conductor of heat with a rating of just 50.2W/mK, Nickel is better but still extremely poor coming in at 90W/mK. If this plate were even cheap aluminium you’re looking at increasing those figures to 205W/mK. What boggles the mind is there are two perfectly good materials for conducting heat that this plate is sitting on top of, copper and aluminium. All you had to do here XFX to fix this is;

1. Fill the right hand side gap and bottom gap across the heatpipes with a thermal pad.
2. Use an aluminium plate instead of a nickel or steel plate.
3. Put some thermal paste on the underside of the memory plate that contacts the primary heatsink to transfer the generated heat effectively to minimise hotspots.

See how easy that is, XFX? You’ve only changed production costs by a marginal amount at worst. Swapping to an aluminium plate instead of steel or nickel might even make it cheaper despite having to use extra thermal materials.

Did all of the XFX engineers suffer a sudden blow to the head mid way through designing this cooler? First the rear plate, now this. It’s the only reason I can think of for things this absurdly stupid happening.

I’ll show you all an example of how to fix this instead of just giving steps because out of the box what XFX has done is only going to completely toast those GDDR5 memory modules, this is so bad I’d say it is not fit for purpose.



Here we have everything in place to largely fix this thermal solution, the pads you use will need to be 1mm thick and good at compacting down, if you don’t like the thought of heat from the GPU transferring heat to the memory plate you can leave those pads on the heatpipes off but conversely you are giving less surface area for the memory to transfer it’s heat away which is ultimately worse. We are still stuck with the steel or nickel plate with this solution but there isn’t much we can do about that unless you also happen to have access to aluminium sheets and professional workshop cutting tools to outright replace it.



And here we have the plate back in and screwed down, just cut and wipe away the small amount of excess and viola, even with that steel or nickel plate you’ve likely just shaved about 10c off of those memory temps. I also replaced the VRM thermal pad and cleaned the surface area as well as fixed the smashed fins while I was doing this, something we will get a close up look at next.



Here are those smashed fins disconnected from the array and this wasn’t the only area this had happened. This is NEVER ACCEPTABLE XFX! Not even as a one off fluke, ever. This is absolutely horrendous quality control.

Mercifully the final stop on this heatsink horror show is the fans used, CF1010U12S, they are 95mm 4-pin fans with a current of 0.45 operating on 12v DC, standard fans you’d find on budget RX590 models such as the Powercolor Red Dragon, and depressingly, the “standout” highlight of the XFX definition on what a cooler is.

You might also be interested to find out that the XFX RX590 is just a rebadged VisionTek RX590, or XFX are at least using the same abhorrent cooler as VisionTek are, here, see for yourself;






Looking the PCB over the first thing we come to is the phase layout and what voltage controller is being used, in this case it is the NCP81022 which is at best without the use of doublers a 4+1 regulator, immediately with this controller and number of phases on the XFX RX590 we know this VRM is using phase doublers, we’ll get to those.

For the primary phases I was surprised to find the rather good On-Semi 3035 MOSFETs, they are 50A monsters with a switching frequency of up to 1.5MHz, and junction temperature range of -40c to +125c. Given the rest of the components I’m surprised this card isn’t using 40A or 45A MOSFETs, it would have made a lot more sense because as far as current handling capabilities go these 3035s are too good from a quality standpoint for the rest of the components.

Moving to the other end of the PCB we have dual encapsulated ferrite chokes and QPI 3098M and 3092 MOSFETs, they have a maximum rating of 128A and 56A respectively and an operating temperature of -55c to +150c, these too seem rather overkill when looked at next to most of the other components.

Flipping the card over we have the phase doublers they are On-Semi 81162s capable of 3 levels of PWM signals to the drivers and short delays, no surprises here. There’s nothing inherently wrong with doublers in my book the problem I have is that I consider this VRM very heavily compromised. Given the number of phases, the use of phase doublers combined with the quality of voltage regulator and power draw of the RX590 under load, yeah, this VRM isn’t inspiring confidence in me due to the number of phases and voltage regulator. I wouldn’t be anywhere near as concerned if the phase doublers were dropped and a 3567B or 35217 voltage regulator were being used.

The revision of this card is A.0, now, I don’t know how XFX handle their PCB revisions but if we assume a Hex system (which is a reasonable assumption given the revision) then we have 0-9 (0 being an engineering sample) and A-F, meaning this “A.0” revision could be as much as the tenth (not including the ES) revision. Cripes if that’s the case XFX need to buy a dictionary and look up the word “consistent” then read and remember it’s meaning.

I’ll be honest for me this PCB looks like somebody just stuffed their hand in to a bag of random components blindfolded and threw them at the PCB, and with whatever stuck, went “That’ll do! Roll it out!”

Last but by no means least is the memory this XFX RX590 uses, it’s Micron D9VVR. Some people might bemoan Micron but those people are wrong out of the box Micron memory has the best stock optimised timings (still not good per-se, but better than Samsung and Hynix) while also overclocking very well just like Hynix ICs and typically with Micron you can set critical timings a little tighter compared to Hynix. No complaints here.

Now we move to the SECOND XFX RX590, this card I went and bought out of my own pocket just to finish this it is an older (and much superior) v1.0 card I happen to get so here is a quick look at that.



The older revision XFX RX590 is substantially better quality as far as the PCB goes, starting at the back end there is a single encapsulated ferrite choke instead of two on the later revision with the High and Low FETs being Sinopower SM4373s and 4377s, these are middle of the road with operating temperatures between -55c to +150c and rated for 70A and 50A, adequate but nothing special, you might want to put some low profile heatsinks on them. Moving further forward the voltage regulator is the excellent International Rectifier 35217 configurable up to 8+0 it seems to be a modernised replacement for the IR3567B on investigation as such there’s no phase doubling going on here leading to exceptional voltage stability when measured with my DMM and HWInfo64 even at an update rate of every 0.25 seconds. Next to the inductors are 50A International Rectifier 3578s, these things are pretty beastly and really cool, figuratively and literally, as they have essentially a heatsink built in to the package which allows for direct contact to more efficiently cool them. The memory used on the card is the same Micron D9VVR.

This is the card used for testing going forward, and you’ll find out why that is soon.


Architecture Specs & Test Setup

Not something I usually feel the need to do now, but let’s take a quick look at a chart.



No doubt at this point you are wondering why I’ve included this chart. It is because I think there is something not talked about that absolutely should, namely the benefits and negatives if you were to consider an RX5500XT instead of an RX590 or even RX580. What nobody seems to talk about is that the RX5500XT out of the box is just a more power efficient RX590 that brute forces it’s way to performance somewhere between a 580 and 590 with raw core and memory speed, yes that’s right, the RX 5500XT most of the time cannot outpace a stock RX590, let alone OC models and manually OC’d cards. Many people forget, or just don’t mention, that the RX590 is itself rather friendly to undervolting particularly if you are fine with taking a small core speed hit and optimising those memory timings to make up the deficit of the slightly lower core speed and then some. This is something that will also be looked at, further to this point you can buy an RX590 for about £160 now where a 5500XT will cost you about £200 for a good 8GB model and is slower. AMD would likely rather nobody remembers or knows these things, but I do and I remember, with the way of the world right now being thrifty and spending smart if you are looking for a capable GPU for your needs without spending a small fortune is more important than ever now.

Out of the box the XFX RX590 has a very high core clock of 1.58GHz, otherwise there isn’t anything special or different from AMDs guideline specification. I did note though that on the XFX site, and I’ll quote the snippet from them here, it reads ”… We’ll keep working to improve your experience with free, regular software updates.” Is that so XFX? Because last I checked AMD did all the software work not you so stop trying to take credit for something you contribute absolutely nothing to, just a bit of a misleading statement there which could easily confuse inexperienced users.

With all that said and done time for the test system, one day I may reveal it in the light but that day is not today.

CPU: AMD Zen+ 2700X @ 4.1GHz 1.26v
Mainboard: MSI X470 Gaming Pro Carbon
RAM: 2x16GB Crucial Ballistix RGB 3200MHz @ 16-18-16-35 1T
GPU: XFX 8GB RX590
Storage: 250GB Hynix SL301 SATA SSD, WD 120GB M.2 SATA SSD (OS drive), 2TB Seagate Barracuda
Opticals: 24x Lite-On iHAS324 DVD-RW, 16x HP BH40N Blu-Ray
Sound: Xonar DX 7.1, Realtek ALC1220
PSU: EVGA 1000w Supernova G2
OS: Windows 10 Pro x64 (latest ISO) and all updates
Case: NZXT Phantom 530


Benchmarks

*poof* what’s that smell? What’s wrong Blackstar why have you just powered off entirely? Oh boy… so, remember how I’ve been referring to one of the XFX RX590s as the FIRST card? Well this is why, 10 minutes in to benchmarking it the card literally committed suicide during the 3DMark Fire Strike test. On investigation of the issue it turned out power delivery from the PCI-E 6 pin connection failed, the test system literally refused to power on at all with that connection in place and removing it running with just the 8 pin of course lead to no display. Well I did say I didn’t like the setup of the power delivery and questioned how the card would manage under load, turns out not very well, not very well at all.

Fortunately there is a happy ending here Blackstar survived with seemingly no trauma, he’s told me though an XFX product is never coming anywhere near him again after this, I don’t blame him.

Back to the benchmarks.

All benchmarks here were run at maximum in-game settings with the exception of Metro Exodus which was run on the “Ultra” setting as settings above this just dial things up to 11 and / or enable RTX features. Tomb Raider had Ambient Occlusion manually set to HDAO instead of HBAO+, due to HBAO+ being an nvidia specific form of AO, even though it will run on RTG hardware.



For consistency the driver used to benchmark the RX580 was used when benchmarking the RX590, which were the 19.11.3 drivers. To make things a little more interesting the XFX RX590 is up against a Sapphire Nitro RX580 8GB @1425MHz with a memory clock of 8600MHz (effective) and my custom optimised timings for Hynix memory (which are very, very optimised).

Not surprisingly there is the thinnest of hairs between the optimised RX580 and stock RX590, but none the less the RX590 is a mite faster despite the RX580s best efforts. You’re probably wondering right about now why you would get an RX590 if you can get performance this close with an RX580 even one that can’t get above 1425MHz on the core, we will get to that.



Looking at the gaming benchmarks you’ll see there really isn’t much room between an optimised RX580 and stock RX590 here either however Tomb Raider and Alien Isolation both show they like that extra core speed the RX590 offers.

Metro Exodus results particularly are poor however, I did run the test again with the latest 20.8.3 drivers and there was no change, I wouldn’t say this problem is on the GPU or the drivers but rather more down to a poorly optimised game, or one at the very least that forces you in to settings because the graphical options versatility it once had with the Metro Redux titles is now gone, a very bizarre decision to say the least and Exodus doesn’t really look any better than the Redux titles but runs about 3x worse. Good job 4A, good job.

Gears 5 with the new drivers also exhibited very poor performance compared to 19.11.3, frame rate with the exact same settings and resolution is 10FPS worse that literally is like stepping down from a 590 to 470, I told RTG about this issue 3 months ago. 3 months, RTG.


Thermals, Power Consumption & Overclocking

These results were measured using the more aggressive fan profile (vBIOS switch flipped away from the 6 pin connector) while the card was left to run the FFXV benchmark @4k with maximum settings for 30 minutes. Noise levels were taken with a digital sound meter in an environment with no external noise sources, everything is fully controlled, and due to the rear plate being an essential component for the VRM part of the primary heatsink we have to rely on HWInfo64 rather than my DT8380 temperature gun for temperature readings.

Starting with coming around more to why you would choose an RX590 over the 580 let’s take a look at core power numbers.



I opted for a smidgen over AMD recommended clocks here but if you wanted to drop down a little further that would be another noticeable drop in power, required voltage, and heat output. For me 1.5GHz would have been much more realistic as an AMD guideline and still a whopping +160-120MHz over most RX580 cards while consuming much less core power, requiring less voltage, and generating much less heat. Am I the only one thinking the Radeon Technologies Group threw the power envelope out the window for no good reason with the RX590? They have been guilty of doing this for some time now.

I’d wager you’re thinking of the RX590 much more favourably after these results aren’t you? Let’s look at that thermal performance and see what undervolting does.



With total system noise levels at stock clocks and voltage things are brain splittingly loud it was like a banshee screeching constantly and I fast developed a headache from it as a result. It should be noted that system fans are 140mm low RPM high airflow, and thus, are whisper silent. To say the XFX cooler was obnoxiously loud would be an understatement and load temperatures were terrible for the amount of noise being generated.

Moving to the “overclocked” results they were significantly better but still too loud. If you have absolutely no other choice than to get this XFX RX590 I would strongly recommend setting a GPU frequency of about 1.53GHz and undervolting for all it is worth you’ll be able to pull the performance pants up and still gain a little with optimised memory timings.

Speaking of overclocking, it’s time to see the “overclocked” results of this XFX RX590. I had to somewhat reinvent the term overclocking for this one. I stuck to the undervolt of 1.087v and pushed the core as high as I could get with that, idle core voltage was also changed to 0.7v, for whatever reason XFX even overvolt that to 0.75v. Memory I did not shy away from with the thermal improvements I made and ended up at 2200MHz with no Error Detection and Correction (EDC) collisions after a 3 hour test. I also spent time optimising the Micron memory timings, as stated earlier they aren’t too bad out the box but there is still a lot of room for improvement which by the time I was done boosted total actual bandwidth to about 235GB/s, on 2200MHz and the theoretical maximum of 281.6GB/s that works out to an efficiency of around 84%.



There is nothing to note here at all “overclocking” the XFX RX590 shows nothing outside of minimal gains to slight run to run variance changes, even the +200MHz on the memory doesn’t seem to have made any difference. The optimised memory timings however have bought equilibrium to performance though allowing for that lower core clock, and hence lower power draw, voltage, and generated heat so while performance improvements are minimal at best I do think a lot of people will very much appreciate the non performance benefits that are attainable with an approach like this all without having to sacrifice out of the box performance levels.

This is no excuse for you XFX, I was simply forced to do things this way because the HSF is awful. Against my better judgment and recommendation for this XFX RX590 I also pushed the core as high as I could without the card completely melting down using a voltage of 1.125v so let’s see how that changes things.



Overclocking the XFX RX590 once again showed little gain in AC Odyssey at 1440p with the ultra quality preset, one tweak changing the volumetric clouds option to “High” instead of “Ultra High” showed a significant gain however of about 20% showing how harmful settings that are there just as feature creep rather than providing an actual noticeable visual benefit can be. Switching Odysseys Anti-Aliasing to “Low” also saw a sizeable frame rate increase of a further 17% and indeed the lower level of TAA looked better to me as the final rendered image is less blurry.

I will never understand why devs feel the need to add feature creep like this nobody likes their frame rate being needlessly wasted regardless of the class of GPU they have.


Conclusion

You’re going to get two conclusions for the price of one today, starting with the A.0 revision XFX RX590 that blew up, the HSF is poorly made, is loud, and the card runs hot as hell as a result. The component quality on the PCB was also largely poor with terrible voltage stability the only real exception of quality were the MOSFETs. If you have this particular revision of the card masquerading itself as one of much better quality than it actually is return it because it is utter crap not worthy of a single glance, let alone your money. I’m surprised you can open the packaging without finding a piece of card with “graphics card” printed on it in place of an actual graphics card penny pinching is so extreme here and this also serves as a prime example of what you are in danger of receiving in exchange for your hard earned cash when AIBs are left to run amuck with no set baseline, in this case by AMD/RTG.

Avoid A.0 revision XFX RX590s is the moral of the story here it goes without saying this is absolutely unacceptable from XFX and if you happen to have an A.0 revision XFX RX590 and you can return it, do so.

As for the V1.0 version of the card it is a strange animal, component quality is largely good with the exception of the Sinopower MOSFETs which are just very average. Both vBIOSes on the card were one extreme or the other, quieter but the card will toast itself ultra fast or like nails on a chalk board loud and the card won’t toast itself quite as fast. These two extremes boiled down to one vBIOS having a temperature target of 80c and the other being 70c. Would it have been so hard XFX to set a temperature target of 73c and called it a day for balance?

The next problem with this card, and you knew this was going to come up, is the absolutely piss poor HSF, from top to bottom it has been made with the cheapest quality components XFX thought they could get away with, the copper plate and heatpipes are only there by way of necessity aside from that fin density is the worst I’ve seen on a heatsink in 22 years relative to the thermal load it is meant to handle, don’t even think about trying to OC with this cooler, just don’t, the card will burn up and give you a migraine while it does it. If I had to put a wattage rating on this cooler based on all empirical evidence I’d sit it at about 160w at best, totally inadequate even for the stock configuration which saw the GPU drawing up to 245w core power. In short the HSF is woefully ill equipped to deal with the thermal load of this RX590 the only way forward with it is to undervolt.

Then we get to how XFX decided to “cool” the GDDR, yet another abysmal design choice by using a nickel or steel plate rather than an aluminium one and the transgressions do not stop there with no thermal paste or thermal epoxy under the plate to transfer heat to the primary heatsink and an uneven base with a chunk missing which I can only assume XFX thought would magically make the memory stay cool with a “breathing space”, that’s not how thermodynamics work, XFX. The absolutely absurd thing here is even with the materials that are used you can substantially improve memory cooling by placing some TIM between the memory plate and main heatsink, then fill the gaps on the mounting area of the heatsink with some thermal pads. You could then have placed some thermal pads on the back of the PCB where the memory is and used those for contact with the back plate and thus simultaneously fixing a large chunk of the coolers thermal problems and making the rear plate actually useful and functional not merely there for show. Given how cheap these improvements would be to make the only conclusion anyone can come to is XFX penny pinching to the extreme.

This penny pinching has deservedly bitten XFX in the arse XFX have their own binning process so know full well the range of ASIC quality on a particular model of card and without a doubt simply thought they could get away with a sub-standard cooler. Another problem with the cooler is how the fans work, they would spin up far faster than they needed to before settling down again for absolutely no reason, sometimes even when booting to the desktop the fans would enter “panic mode”, spin up to 70-80%, then settle down. On a GPU that measured an idle temperature of under 30c. I tried four other Sapphire Nitros and Powercolor Red Dragon GPUs and all of these worked fine landing this problem squarely at the feet of XFX and a problem with their vBIOS or choice of fan controller.

XFX not content with a cooler that has ugly performance also built one visually ugly almost like one of those dodgy Chinese fake GPU listings we’ve all seen. Why on earth XFX would want to slap that tacky logo (which itself looks like the victim of a miss-print with the fat “X” and skinny “FX”) to the side of this card to let people know who manufacture it is beyond me, I’d be thoroughly embarrassed by this design effort it looks like some especially immature for their age thirteen year old who pronounces “fat” as “phat” designed it. The cooler is easily the ugliest thing I’ve ever seen, and not just ugly, but the kind of ugly where it hit every branch of the ugly tree on the way down. At least the hunchback of Notre Dame and Joseph Carey Merrick has another companion at the bottom of that tree now. $/£5 extra on manufacturing a decent cooler for this XFX RX590 would have made the world of difference, you can forgive an ugly cooler if it does the job it needs to well, not one that is both ineffective and ugly. How XFX have you gone from the sleek, effective, fairly easy to maintain heatsink assembly of the RX480 GTR Black Edition or even the RX580 Double Dissipation to this visually and functionally abomination of a cooler.

You can just imagine the marketing meeting for this card now can’t you? “Yeah man I got it! We can name it the “Fatboy” that’ll get around the poor looks, poor cooling performance and bad molding issue.” “That’s a phat name, boy! ”PhatBoiii!.” Utterly repulsive, repugnant, and juvenile naming.

Where the card does have some redemption is with the high ASIC quality and Micron memory both cards had, the first card that died in 10 minutes of benchmarking had an ASIC of 74.6% while the card I bought to finish this has an ASIC of 76.7%. Remove the stock cooler and treat the card to something it is deserving of and you would be spoilt for choice with what to do with it, undervolt and have it run super cool and quiet or overclock it to the moon, while in the case of the latter it would still be power hungry the high ASIC quality would at least ensure a high clock at a comparatively low voltage.

In terms of performance the RX590 does also show it has some muscle behind it, capable of that 60FPS standard in modern titles at 1080p ultra settings or close to it a lot of the time and 80-90+FPS if you go in and dial down a few feature creep options. At 1440p the RX590 also shows it can hang in there at maximum settings in modern titles but again manually going in and toning down the feature creep options will see substantial gains increasing frame rate to 50-60+FPS even here a lot of the time.

In short, in 2020, the RX590 is not just relevant still as a gaming GPU it is a competent 1440p card. You won’t get hundreds of FPS, but you will get a respectable ~60FPS with only a few minor adjustments, that’s an awful lot of performance for around £160. Just don’t get the XFX RX590 unless you have absolutely no other choice and even then buy second hand, don’t encourage the recent behavior from XFX by giving them your cash.



When ordinary failure just isn’t enough, call XFX

Wow, yes, what a mess that is with the memory spreader and lack of proper thermal connection to anything!! (love that, breathing space )
That’s too funny about the “Free, regular software updates” too!!! I hope they see that!

Damn, so magic smoke on round one, never a good sign! Did XFX send you this “first” sample?
If yes, you should request exact same variant replacement, so you can see if that is a design flaw, or just random fluke that it smoked out on you first round of testing.

@Lost_N_BIOS Stuff I review I either get from sources, the manufacturer, or I outright buy myself. In this case it was sources not from XFX the first card. The card looked bad to me from the start so I didn’t ask anyone at XFX because I know they would have thrown their toys out the pram when they saw the review and would have tried to “sway” me in to dropping certain comments and words (yes this is an actual thing that happens with some manufacturers) to make it more positive. You can polish a turd all you want but it’s still a turd. XFX can count themselves lucky I dove as deep as I did I really tried to find a road to redemption with the stock configuration and it just doesn’t exist you can make it better quite easily but the user should not have to take these steps themselves it should already be done by the manufacturer and the warranty sticker fiasco is just insulting.

The power delivery with the A.0 is a concern for all cards made like this, you have a controller that has to be configured for 3 phases to then add phase doublers to drive the 6 phases on the PCB. It’s a fake 6 phase design in lamens terms the NCP81022 cannot drive all 6 phases which is why you need the doublers to split to the other phases the doublers will then try to keep everything balanced, on high power draw devices this is just a bad idea as it leads to very unstable voltages that can bounce around by as much as about 83mv compared to a true 6 phase which will vary by about 32mv. On something like core voltage to a GPU that is a huge difference.

I know pricing on these components too, an IR3567B on farnell.com is £3.09, farnell don’t sell the NPC81022 so I had to go to mouser.co.uk there the NPC81022 is £1.74 and the 81162 doublers are 63p each bringing the total cost to £3.63. Theres also bulk pricing to consider here which I couldn’t get but I doubt the NPC controller and doublers work out cheaper in bulk vs. the 3567B in bulk. XFX are literally spending more on an inferior design.

@ket - Yes, I assumed they’d try to sway your words a bit, especially if you talked to them about card #1 first. But, I thought maybe you’d get around that via “random fluke” excuse while asking for another sample to finish review, but since it’s not from them then no chance in that route.
Yes, warranty stickers are a plague on everything, and I suppose they all still do this because they know they can deny RMA/Service with this excuse and many wont know any better and chalk things up as a loss.
To me, this is a bad business decision, even though it may save them some $$ in the RMA dept, because you know that user then swears off company and tells everyone that will listen to do the same too.

Do you think the power delivery layout is the reason it fried, or lack of cooling overall, or both? Crazy, they pay more for less!!

@Lost_N_BIOS I’d say the card fizzed out due to the power delivery yes, something just flat out blew because of a sub-standard design for a card that draws a lot of power.

wow so beauty such expensive