AMD RX Vega 64: The Tom's Hardware Liquid Cooled Edition

[kinggremlin]

According to Overclockers UK:

" (...) the good news is AMD are rebating early launch sales to allow us to hit £449.99 on the stand alone black card which has no games. This is a launch only price which AMD at present are saying will be withdrawn in the near future, when if it happens is unknown, but remember do not be shocked if the price jumps nearly £100 in a few days. This time around there is no early adopter tax, quite the opposite on the stand alone black card, so do be quick."

So the $500 msrp was basically complete BS by AMD simply used as a marketing tactic to make launch reviews look better. You're never going to see a mythical $500 Vega64 again.

https://www.techpowerup.com/236177/amds-rx-vega-launch-prices-might-be-just-smoke-and-mirrors

 

[shrikecurse]

Too much self promotion!

'Perhaps because Tom's Hardware is a little more discerning with awards than most publications, AMD didn't send over one of its liquid-cooled cards for review.'

Any proof on that opinion? Any actual information on the motives behind amd's choice in sending you a non liquid cooled card? Cause it seems to me you are using this random event to present your methodology as superior to unnamed others?! No?

Happens more times throughout the article, basically every time it says: maybe others got the good and we got the bad, maybe others got special samples and we didn't, maybe others test wrong and we are right, maybe others messed it up used old drivers utilized results they shouldn't have etc... while we never do that!

Just tell us what you find - explaining it using your personal opinions with no proof is not right methodology is it? Is this a review comparison? Cause its filled with jabs at other reviewers and it keeps pointing out imaginary possibilities as possible mistakes! wth is this? Basically you are trying to explain differences by way of mistakes others did without actually knowing if those mistakes happened!!!!!

Didn't like that!

So you took a suspected inferior sample, not specifically chosen or sent to you for liquid cooling or its particularly high OC potential, decided to liquid cool and OC this 'possibly inferior sample' and got disappointed by the results?

 

[FormatC]

@shrikecurse
This is simply a misinterpretation from your side.

To be honest: what is better? To say, my card is representive for all other Vega cards, that means Vega is crap - or to make it more relative, because I have only one card tersted? One site with a video channel published 1980 MHz OC, but I know, that this was simply false. They used an older driver to show us a (paid?) sensation, theoretical DPM-targets instead of real clock rates. Is this better? If you have only one card and the result is so worse as in my test, it is a question of honor, to search the reasons why. Maybe my GPU is a potato chipy, maybe not.

But I'm pissed on from all this YouTube crap without any proof. Click baiting. AMD send a lot of samples to the so-called influencers and not to all media. The reason is clear: better hundreds of clappers with fictive awards than in-depth reviews, that may show the disadvantages of a product.

Didn't like that!

 

[shrikecurse]

@formatc I have not seen this youtube video but i do not doubt people who might have will piss on this review cause they don't like it.

I have not read another review, this is the only one. My comments were based on this review alone! I was not aware of others, their findings etc.

All i know is i came to read 1 review of this product and it mentioned reviews and videos and benchmarks and findings of others who i do not know and didn't want to know. None of this information changed anything about the cards performance - which is what it is.

Simply saying: the card disappointed - then thats it the card disappointed! I could have gotten that without hypothetical reasons to explain other results which i didn't even know!

And i stick by my claim that when you review a product - other reviews do not exist until you are done! Then a review comparison can go in and explain the differences sure! But you shouldnt even be aware of other reviews while doing your own and they shouldn't be in not even as mentions!

So you say to someone: this is a great product! But if you have read that review that said its crap - they were wrong! Yes indeed this is a valuable item! But if you saw a video saying it isnt - thats an old test! So please consider this excellent value for performance item! But if you have read a reviews that says it isnt then maybe there got a different sample!

Come on this is ridiculous - you go to buy a washing machine and the person assisting you lists off possible cons based on other people? How when? Made no sense to me!

I read the article carefully - got no beef with your findings! I am not considering this card or a change to my build!

 

[monsta]

Massive fail from AMD

 

[Virtual_Singularity]

Thanks for the article, testing, conclusion, etc. Much appreciated. Fwiw, much as I dislike pointing this out, methinks its (still) too early to properly judge consumer Vega variants properly, because the sad truth is what little there are of them were rushed into release.

This ongoing dgpu market turmoil, due to mining as well as the hype surrounding it, makes this a most inopportune time to rush consumer Vega to release, most especially for the consumer who just wanted a decent card to game with & waited patiently. It'd be nice to be able to know, with any degree of certainty, that AMD's latest offerings will see significant optimization in time for the holidays. That maybe AMD may even have something that'll give NV *real* high end competition, for a decent price, to boot. Won't bet on it, but ya, it would be very nice...

 

[10tacle]

But I'm pissed on from all this YouTube crap without any proof. Click baiting. AMD send a lot of samples to the so-called influencers and not to all media. The reason is clear: better hundreds of clappers with fictive awards than in-depth reviews, that may show the disadvantages of a product.

Yeah I am regularly correcting people who try and link someone's Youtube "benchmarks" on a review. You can make up anything in the virtual world, but you can only trust reliable credited outlets (like this). And I have a hard time believing AMD wouldn't test every Vega sent to the major hardware review sites like Tom's to weed out a lemon. I mean seriously. The same company who sent the majors a custom kit for Ryzen reviews with their own website names on them would fail like that? That makes no strategic sense.

In any event, you are always going to have complainers out there. Especially those who don't like what they see like their favorite GPU brand not living up to the hype standards and/or the review not giving a 100% glowing reaction.

 

[cmsvmylo]

All the kids and noobs buying all stock of vega with this horrible blower style cooler. And for what? So they can get gtx 1070-1080 levels of performance after 15 months of "wait for Vega". Perhaps vega 56 has more value but people with common sense should wait for custom cards next month. I too was hoping that Vega would be a true opponent for nvidia, so I put my purchase on hold. After the double disappointment of performance and pricing (expecting vega 56 to cost at least 600 euros in my country) I bought an Asus strix gtx 1070 from ebay for 340 euros. Looking forward on a double upgrade on an ultra wide monitor and a graphics card next spring, when volta comes out. Or should I wait for Navi?

 

[Virtual_Singularity]

Navi! Maybe Navi will be what some people thought Vega would be (I jest).

I remember hearing legends about genuinely competitive high end AMD dgpus, at unbeatable prices. They went by the name Vega, no relation to Susanne Vega afaik, (just sayin').

These legends were said to be so fragile, anything more than a whisper & they'd vanish. Scratch that, I'm confusing a quote from a character in an overrated flick about a despotic Roman emperor & a denigrated general with one from Lisa Su. When she promised 2016 would be a great year, & that by Spring of 2017 or thereabouts there'd be no doubt that "AMD is back".

The hype made it sound as though when Vega came out, NV would have nothing on AMD, & also something about maintaining a "close second" to Intel, & some other stuff about goals 3 years down the road re dominating graphics in all market segments. =)

Bah, just poking a bit of fun @ AMD. They've made noticeable progress since that interview with Su in 2016. And it's possible consumer Vega variants may turn out how they were supposed to be. In time. Possibly. Err, maybe.

 

[mitch074]

Funny how this reminds me of the first version of GCN - too hot, too power-hungry, too slow in current games... Fast-forward a couple years, GCN 1.0 was still kicking. GCN 4.0 (Polaris) comes out: too hot, too power-hungry, too slow in current games... Fast forward a single year, you just CAN'T buy a new Polaris card, they're all sold out because they're the least expensive, most powerful GPU for mining. Those who manage to game with them report that they're duking it out with the competition.

We'll see in 12 months if Vega really is a lemon. I'm not saying it's not; it's just that, for years now, any time AMD puts a new architecture out, it sucks at first; then, with driver revisions and optimizations making their ways into games, they go from "it sucks" to "hey, not bad" to "wow, interesting" quite regularly (with a "WTF happened?! 45% better performance with a new API and the latest driver?" thrown in, even).

 

[Valantar]

I'm a bit confused by the lack of detail in the cooling solution you are adding. You don't provide any details in terms of radiator thickness or material, which immediately leads me to question whether the 120mm is actually comparable to the rather thick copper rad AMD uses. The same goes for fans - AMD uses a 3000rpm Nidec Gentle Typhoon, widely regarded as among the very best (if not /the/ best) high-performance radiator fans out there. Now, I'm not saying that your solution is necessarily inferior, but the lack of detail in the article all the while stating that it's comparable seems ... thin. At best, you're not communicating clearly, at worst, it's disingenuous.

Now, I've never been anywhere near a Vega card, but from what I've seen the radiator and fan are the same as on my Fury X, which had a 38mm rad (IIRC) and the Nidec fan. Of course, mine had pump whine up the wazoo and has now been integrated into a custom loop, but temperatures were never an issue.

 

[FormatC]

I'm a bit confused by the lack of detail in the cooling solution you are adding

First:
It is a prototype and not an official product test of an AiO. You can google the product to get the specs, but after my tests and suggestions the company will change the radiator size to 240. And - it is one of the advantages of this product - you can change without any skills the radiators, because all parts are pre-filled.

Second:
My intention is not to write any advertising. I had it here in my lab and I tested and used it. But I'm not a part of Alphacools marketing team. I wrote about the rpm and radiator size and on the pictures you can see, that it is not this flat crap, used for all this Asetek clones. Alphacool is using pure copper for all parts, that are in contact with water.

Third:
I have FuryX here, two of them. The water inside was going up to 60°C and higher - useless. Simply take a look at our launch review and the PCB temperatures I measured. This is nothing to use it in a PC.

 

[Rob1C]

@FORMATC - Thanks for doing this Review using the Card you got, working as long as you did, under the roof you chose.

Thanks for mentioning how your experience with the Card and Software differed from other Reviews you read and offering your advice that we can take or leave. Not your fault that the Card needs a lot of Watts for a less than stellar overclock or that the price jumped to $700.

People should take all the opinions they care to read and only flog the Reviewer where it's clear that something is going on - why discourage someone from trying to help. IF you could say 90% of the other Reviews say you are wrong and here's one Link then it would be different but when you get into personal attacks (is where you work and test this stuff upscale or your mom's garage, and how do you know this or that) but don't have something better to offer you discourage people from working either for free or inadequate rewards.

We don't need fewer Reviewers and aren't desperate for better ones, this is a great site. Don't rake them over the coals.

[Addressed to more than one person, and not specific to this Article alone, nor this Site; just a general observation about appreciation for getting things for free.]

PS: If the Features/Buttons recently broken got fixed there'd be one less thing irking us.

 

[bit_user]

basically every time it says: maybe others got the good and we got the bad, maybe others got special samples and we didn't, maybe others test wrong and we are right, maybe others messed it up used old drivers utilized results they shouldn't have etc... while we never do that!

Huh. Each of those caveats is something appropriate, on-topic, and worthwhile for the reader to know. I would have had no idea about the driver issue, and one shouldn't assume readers automatically know how much variation to expect in overclockability of GPUs.

Just tell us what you find - explaining it using your personal opinions with no proof is not right methodology is it? Is this a review comparison? Cause its filled with jabs at other reviewers and it keeps pointing out imaginary possibilities as possible mistakes! wth is this? Basically you are trying to explain differences by way of mistakes others did without actually knowing if those mistakes happened!!!!!

I'm not saying the article is perfect, but you're being too harsh.

So you took a suspected inferior sample, not specifically chosen or sent to you for liquid cooling or its particularly high OC potential, decided to liquid cool and OC this 'possibly inferior sample' and got disappointed by the results?

It's a worthwhile experiment, for several reasons. First, it explores to what extent Vega's performance might be limited by thermal runaway. This would slightly exonerate AMD, if they designed a GPU that could clock much faster but is being held back by worse-than-expected power efficiency.

Second, it tests a scenario people actually do. It's not uncommon to buy an air-cooled GPU and convert it to use liquid cooling. So, this article does a service in the sense that it provides one datapoint indicating how much is to be gained by this and how much cooling you're likely to need, if you do it.

So, I happen to disagree. Furthermore, testing this stuff involves many hours of grunt work. For that reason alone, I would be a bit forgiving of an article's tone. The important points are the methodology and the data.

 

[FormatC]

It is normally not my style to comment the job of other guys, but this time I was really annoyed.

Why? It is not too much to redo all tests with newer drivers, if a proof shows me different values. Ok, AMD delivered it really late, but not too late. And it is always a question of honor, to do all things and measuring as exact as possible and not to spend all the time in hairstyling, video editing and some special effects. YouTube is a good media, but the YouTubers are mostly a bad joke (with exceptions, sure) or paid influencers. I know, that Nvidia has also a special PR, only to pamper all this influencers and show them the "right" direction...

This 1980 MHz lie is only one example from a lot of. If a product is showing not the expected performance, a reviewer has to proof it twice, what was going wrong and where is the reason why. A bad single sample or generally an bad product? Whoever writes a public rating, bears a great responsibility to the manufacturer and much more to the potential customers.

It makes no sense to over-hype or bash something. We have to stay as objective as we can. This is sometime painful for all this fanboys, I know. But on the other side it is even more helpful for all this guys, working hard for their money, to decide, what the will buy (or not).

Maybe I wrote it with too many emotions, but I feel me really sick, if I see, that a vendor delivers samples to all this
frivolous guys to prevent a product before in-depth tests and to earn fictive awards for simply nothing. Vega is not matured, we need better drivers and it takes a lot of time for all the developers to understand this architecture. But at the moment it is simply not a recommendation for the current price tag.

 

[bit_user]

Massive fail from AMD

This is premature, IMO.

Out of the box, Vega disappoints. However, I think it's too early to judge, from one early sample, much about the GPU's OC potential. And even if it is representative of what we can expect, I wouldn't call a GPU that's cost-competitive and works reliably a "massive fail".

And not to excuse its gaming performance, but I'll note that its compute performance is looking a bit brighter:

http://www.anandtech.com/show/11717/the-amd-radeon-rx-vega-64-and-56-review/17


I didn't even find a benchmark of its deep learning performance, but that should easily trounce Titan Xp. So, whatever you say about these gaming products, I definitely think it's too early to judge the GPU, itself.

 

[FormatC]

I have a lot of good and closer contacts to the R&D guys of a few AIB and I see the issues and problems to build good and working custom designs. AMD made a bad job and confused the AIB with different packages, missing informations and a lot of delays. If you have a schedule, you must keep it. If not, the whole thing will be a disaster. I look at this product also like an engineer. I know all the single steps from BOM (bill of materials) to the start of mass production in the factories, also the costs and the schedule of this AIBs. To be honest - I never saw such a chaos before.

And not to excuse its gaming performance, but I'll note that its compute performance is looking a bit brighter:

It is a gaming card, nothing else. For deep learning, AMD gave us the Vega FE or MI25.

 

[Karadjgne]

I find it refreshing. While on one side, it's nice to see what a component 'can do', just like most YouTube videos, it's gotta be taken with a grain of salt since you can never trust that you'll get a unit off the shelf, but a cherry-picked perfect specimen. If I were to buy a Vega, with my luck, I'd get a card from the same batch as the one tested here. So if the card 'disappointed' cuz it could be a bad card, well that's what many will get. At least it's an honest card. Honestly, I'm glad to see shortcomings, I'd rather see worst case scenario, and anything better is a bonus. My 3570k won't OC past 4.3GHz, no matter what, so seeing all the reviews where ppl hit 4.9GHz etc kinda makes me wanna cry 'foul', it's cherry-picked.

So Tom's test couldn't repeat reddits numbers.. Hah. You'll find most end users can't repeat YouTube videos numbers either. On any component.

 

[10tacle]

A tech site has leaked info on an ASUS Strix RX 64 variant giving us first light on AIB partner offerings. If this is true, it's not looking like they'll have much of an improvement in performance and only consume MORE power (it's a Denmark site so you may need a browser translator for English):

https://www.computerbase.de/2017-08/asus-radeon-rx-vega-64-vorserie-test/2/

 

[bit_user]

And not to excuse its gaming performance, but I'll note that its compute performance is looking a bit brighter:

It is a gaming card, nothing else. For deep learning, AMD gave us the Vega FE or MI25.

I was actually trying to draw a distinction between the Vega gaming products and the GPU, itself. I thought I made that very clear. You conveniently omitted the part where I said that.

The underlying GPU was designed with multiple goals.

 

[footman]

I just received my Vega 64 Air and am still waiting on my waterblock to put it under water next week. I have a 360 rad and 240 rad in my system and a stock clocked Ryzen 1600x. This is my first AMD card since crossfire 290x's and have been excited as I have a new 27in Freesync 2560x1440 144hz IPS monitor and fully appreciate the shortcomings of the Vega card compared to Nvidias 1070, 1080 and 1080ti.

Straight away I can ping AMD for terrible decision to sell Vega air with the blower HSF it comes with. No-one should buy this version of Vega unless you have no intention of using the stock cooler like me. It is loud and can't keep the card from throttling even at 100% fan speed. Hopefully the AIB designs will be better designed from a cooling perspective. Once you get past the terrible stock cooler the card itself is well designed the VRM's are well built and designed to easily handle the high power requirements.

My focus with this card so far has been to get it to run it at 1630mhz, which is the DPM7 setting that can be seen in Watt Tool. I am not so bothered about overclocking on air. I have had some success, but I am still not happy with some of the tools that are available to monitor clocks, they are pretty buggy.

Once I have the card under water I will report back, I have been using the default bios setting on the card and then applying a 50% power with fan at 100% and have seen best results with this, however I am sure that power is through the roof, so my next steps will be to use Watt Tool to underclock the DPM6 and DPM7 settings a little...

 

[footman]

Reports of high overclocks are likely to be inaccurate:
http://www.gamersnexus.net/news-pc/3026-psa-vega-frequency-reporting-bug-in-software

 

[Memnarchon]

This 1980 MHz lie is only one example from a lot of. If a product is showing not the expected performance, a reviewer has to proof it twice, what was going wrong and where is the reason why. A bad single sample or generally an bad product? Whoever writes a public rating, bears a great responsibility to the manufacturer and much more to the potential customers.

They already admit it that it was an error and they blame AMD for this.

(translated) We are talking about a limitation of overclocking in these graphics cards and that the working frequencies are smaller than those we show and it is true that these RX Vega 64 work at a base frequency of 1247MHz and a Boost mode of 1546MHz, including Sapphire It reflects this in its website, but in the benchmarks made, the graph Sapphire RX Vega 64 appears in the base 1630MHz and allows us to reach the 1980MHz. It is likely that the frequency data is erroneous because of AMD's BETA drivers, but in this case it would not be our fault, but AMD itself for launching a graphics card without definitive drivers.

Link to translated site.
And link to original.

 

[FormatC]

@memnarchon:
AMD informed ALL testers with samples days before launch about this issue and shared also the beta6 drivers that shows the right values, also before launch. This were a lot of sites and Youtubers and all this guys were well-informed. Don't say, it was a mistake, I call it pure click-baiting and a lie.

 

[CaptainTom]

These results seem quite different than mine. Maybe I just have a golden sample, but I have no issues getting this thing to 1752 MHz (core, undervolted) / 1085 MHz (Memory, stock voltage).

I know the clock is applying too because each increase gives me higher hasrates with Etherum. Although most games seem to no be making use of the extra horsepower in 1080p@144Hz, I think it is mostly cpu bottle-necking.