1. Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Water Cooled Vega 64 Overclocked And Tested By TomsHardware

Discussion in 'AMD / ATI' started by Th3pwn3r, Aug 18, 2017.

  1. Th3pwn3r

    Th3pwn3r

    Joined:
    Aug 6, 2009
    Messages:
    356 (0.12/day)
    Thanks Received:
    64
    Location:
    Chicago, Illinois
  2. Rehmanpa

    Joined:
    May 19, 2017
    Messages:
    310 (2.38/day)
    Thanks Received:
    93
    Wow, that really sucks. I wonder how vega 56 does with water cooling, especially with a much cheaper price point.
     
  3. Eric3988

    Eric3988

    Joined:
    Dec 21, 2005
    Messages:
    330 (0.08/day)
    Thanks Received:
    50
    Location:
    Great state of Texas
    That's if Vega 56 is the price we've been told...
     
    10 Year Member at TPU
  4. xkm1948

    xkm1948

    Joined:
    Mar 18, 2008
    Messages:
    2,340 (0.67/day)
    Thanks Received:
    1,399
    The single 120mm radiator is barely enough for FuryX, not to mention the power monster Vega.
     
    ZeppMan217 and Vya Domus say thanks.
  5. Th3pwn3r

    Th3pwn3r

    Joined:
    Aug 6, 2009
    Messages:
    356 (0.12/day)
    Thanks Received:
    64
    Location:
    Chicago, Illinois
    Well, the take home is that this water cooling kit is probably quite a bit better than what's included from AMD. It has some really nice fans pushing a good volume of air through the radiator and they're probably a lot more quiet . However, Toms was a little more to honest and didn't sugar coat it. You could clearly tell Igor was disappointed.
     
  6. fullinfusion

    fullinfusion Vanguard Beta Tester

    Joined:
    Jan 11, 2008
    Messages:
    9,456 (2.67/day)
    Thanks Received:
    3,292
    Toms blow bro!

    Christ it's a pathetic site for any info imho.
     
    Boosnie, Cybrnook2002, Melvis and 5 others say thanks.
  7. HD64G

    HD64G

    Joined:
    Apr 30, 2011
    Messages:
    745 (0.32/day)
    Thanks Received:
    672
    Location:
    Greece
    As we know by now, Vega10 gpus can be heavily undervoltaged. That alone allows much better clocks and performance for lower consumption. The reviews tend to ignore that. Gamers nexus made a live video for Vega 56 on air on tube a few days ago about that and went well:



    So, how much better will a water cooled Vega 64 will do is profound.
     
    DeathtoGnomes and Vya Domus say thanks.
  8. the54thvoid

    the54thvoid

    Joined:
    Dec 14, 2009
    Messages:
    6,483 (2.28/day)
    Thanks Received:
    5,755
    Location:
    Glasgow - home of formal profanity
    If you mean it will perform astronimcally high, you'd be wrong. Hexus had an invite to see the Asus Strix version, already a working model. It ran at clocks of 1900+ on the core but it gained a 5-10% perf increase, with a large uplift in power usage. Vega 64 is not a fantastic development. It's too big and heavy to run fast enough but at 1900. that's actually faster than a 1080ti FE after throttling.

    Asus Strix AMD Vega 64

    [​IMG]

    In March 2017, 1080ti FE (no overclocking).

    [​IMG]

    Vega power use

    [​IMG]

    1080ti FE power use

    [​IMG]

    A range of custom cooled 1080ti and a poorly cooled Titan Xp

    [​IMG]

    Even the Auros Extreme OC only pulls 353w max power.


    You know why @xkm1948 gets a hard time? Because he was such a furious early defender of the Fury X which had all the same hype Vega is getting. But what happened? He felt let down and the promised land of fine wine was BS, spread by AMD supporters who can't seem to accept, at release, the card is about 5-10% max where it will be in the future (including customs). Vega isn't going to get much better and for the time spent 'improving' on Fiji, the end result is disappointing, given how good Fiji actually was at release.

    People need to accept that the card is out, it's there for all to see and there is no magic. I'm going to hear responses about the tech that AMD have put into it and that once developers start using it but hey, that happened two years ago with Fiji. We buy in the present for items that deliver now. Buying on hope that it will deliver something not promised is a big gamble and for people like xkm 1948, that never paid off.
     
    terroralpha, wolf, Th3pwn3r and 2 others say thanks.
  9. Vya Domus

    Vya Domus

    Joined:
    Jan 8, 2017
    Messages:
    1,225 (4.69/day)
    Thanks Received:
    613
    It did not , Fiji had nothing particularly special compared to the previous GCN iterations , if there was something to be done for Fiji it would have applied to all other previous cards and even to Polaris. I know you and others don't want to accept it but Vega is very different from previous architectures. And no , that does not mean it will get a million times better , it's just the reality of things.

    Not everyone does that , and this way of going about doing things should not be taken as an absolute. Historically I bought things that were supposed to be not that great in the present but where promising for the future, and it did pay off. Of course that does not mean it will always happen but don't completely discard this idea.

    If I was in the market to buy a 500$ card ( assuming you can find Vega for 500$ :( ) , I would most certainly pick Vega 64 over the 1080 , Pascal has shown what it can do and there is nothing left in the tank for it. I would much rather pick something that has a chance to get better.
     
    Last edited: Aug 19, 2017
    ZeppMan217 and HD64G say thanks.
  10. the54thvoid

    the54thvoid

    Joined:
    Dec 14, 2009
    Messages:
    6,483 (2.28/day)
    Thanks Received:
    5,755
    Location:
    Glasgow - home of formal profanity
    I'm not an engineer so I can't argue how different Vega is from Fiji (did you design it?) but given it has the same cores, the implication is it's still based on it. If altering core structures and instructions is new architecture then people ought to recognise Pascal as new. Vega is an iteration of GCN. You read enough reviews, that's the consensus. GCN with tinkering. It's not ground up, you want that, look at Zen. I would argue people like you (for some reason because you're not one of those red types) can't accept Vega is not a ground up design. You need to prove it is because on paper, it's very similar. You won't be able to because it's still based on the compute heavy design of GCN, which is just after all a handy name, not a strict protocol.

    Anyway, my point was, Vega 64 will not miraculously get better. Not how different it is. Donor sure why you went down that path to counter my point. If anything, you have reinforced it based on, it might get better assumptions.

    This is a circular argument you cannot prove now. So come back in 2 years when Vega 64 is 30% faster. I'm not having a discussion on 'ifs' and 'buts' when Navi will be out by then, a new AI heavy design.
     
  11. RejZoR

    RejZoR

    Joined:
    Oct 2, 2004
    Messages:
    11,896 (2.51/day)
    Thanks Received:
    5,508
    Location:
    Europe
    Vega does NOT have the same core. Just because the units arrangement is the same, that doesn't mean it's the same core jesus...
     
    10 Year Member at TPU
  12. radrok

    radrok

    Joined:
    Oct 26, 2011
    Messages:
    3,121 (1.44/day)
    Thanks Received:
    893
    Basically anything owned by Purch.
     
  13. Toothless

    Toothless

    Joined:
    Mar 26, 2014
    Messages:
    4,966 (3.88/day)
    Thanks Received:
    3,756
    Location:
    Washington, USA
    A couple of coworkers and I were talking about the outrageous power consumption Vega has. They both were pretty disappointed with it and figured a 1080 was better,
     
    xkm1948 says thanks.
    Crunching for Team TPU
  14. Vya Domus

    Vya Domus

    Joined:
    Jan 8, 2017
    Messages:
    1,225 (4.69/day)
    Thanks Received:
    613
    You are talking about things in very absolute terms , which is why I cannot agree with you. I never once said Vega is nothing like anything RTG ever made , I said it is considerably different compared to previous iterations. Come on man , you have right here on TPU an article about it , we don't need to be engineers to fully understand some of this stuff. Pascal certainly had improvements and changes from Maxwell , have they been as big as the jump from Polaris to Vega ? Objectively they haven't been.

    I NEVER said it will get miraculously better or by 30% or whatever , please point me to the comment where I said. What I said is that it definitely has power left untapped , whether or not RTG will ever be able to make use of it is not guaranteed.

    Also make up your mind , you say this is a circular argument that I cannot prove ( I never tried to man , read my comments again ) but previously you claimed you are 100% sure Vega will not get better , isn't that a circular argument that you cannot prove as well ?
     
    Last edited: Aug 19, 2017
  15. HD64G

    HD64G

    Joined:
    Apr 30, 2011
    Messages:
    745 (0.32/day)
    Thanks Received:
    672
    Location:
    Greece
    But the first 2 graphs clearly show an increase of 10% both for min and average FPS just by putting a better cooler on it that keeps thermals in check and allow a higher sustainable clock. Even better than what happened to RX480 when comparing stock vs custom designs. And consumption will get in check with a bit undervoltaging also. Not too much of a hussle for any enthusiast me thinks, eh? As for clocks at 1900GHz, it might show that on monitor program but it doesn't produce the performance of it because of power limitations. Polaris max boost is like this also.

    To sum up, without driver optimisations, optimum usage of new features that never existed before on a GPU, and just by a custom design, RX64 gets ahead of custom 1080s. And reaches the middle between 1080 and 1080Ti on average for DX12 and Vulcan games Exactly what I was expecting as a firset step of imeddiate improvement. Here come the sw side to make things (much?) better. We have seen it on EVERY AMD GPU launch for the past 5 gens. This time might surprise many how much potential exist in Vega arch for new and future games when the new features are used. Call me optimistic on this.

    And to make things clear: I am glad that many guys here already have 1080s or 1080Tis and enjoy high FPS on recent games, especially if they bought them once the prices got down. But, why so many have to bash anything new that cannot reach their expectations at launch, especially when AMD history show that it is normal and in 2-3 months things get better or even much better sometimes?

    As for Fury X, it got to the point to be equal on average to 1070 at 4K even with only 4GB HBM that most considered a handicap on that resolution. And on same games it is and need to lower graph quality. Imho, Fury X only serious handicaps were the low oc potential and its initial price due to expensive HBM and water cooling cost. As for sw for specific workloads it is very special thing to predict for a product when bought if not working from the launch date. By that, I don't refuse https://www.techpowerup.com/forums/members/50521/ the right to protest for it. Bitterness though should have a limit, or it gets into us and make us harm. Why to make our life awful because of pc parts afterall? Enjoy life all! :toast::lovetpu:
     
    DeathtoGnomes says thanks.
  16. the54thvoid

    the54thvoid

    Joined:
    Dec 14, 2009
    Messages:
    6,483 (2.28/day)
    Thanks Received:
    5,755
    Location:
    Glasgow - home of formal profanity
    Like I said, I'm not an engineer or a reviewer but at least Anandtech aren't as naive as some people on TPU

    http://www.anandtech.com/show/11717/the-amd-radeon-rx-vega-64-and-56-review/3

    Go argue with that web site because you know more than them. As for the TP{U review, I'd argue w1zzard took the popular path and didnt want to describe it as a GCN variant but... it is.


    Bit-Tech

    @RejZoR - of course it's not the same core - GP100 and GP104 are not the same core but they are both Pascal. Jesus...

    Vega is still based on GCN. Lots added but still very much GCN.

    And again - GCN is the architecural basis - it doesn't mean it's the same chip. Lots of improvements and new implementations but still, GCN.
     
    Last edited: Aug 19, 2017
  17. Vya Domus

    Vya Domus

    Joined:
    Jan 8, 2017
    Messages:
    1,225 (4.69/day)
    Thanks Received:
    613

    What are we even talking about anymore ? Whether or not AMD wants to name their architecture GCN or whatever ? You do know what an execution model means in this context ? It's about how the code is interpreted at the compiler level when it is converted into whatever machine code Vega/GCN uses such that the way things get done follow a certain order. No wonder that needs to remain the same , it is fundamental for compatibility. The hardware implementation can vary wildly while still maintaining the same model. That's why you can still run ancient 30 years old code onto modern CPU even though they might handle things internally in a very different way.

    But you know what OK , Vega is Fiji/GCN 3.0. Screw anything written in that very article you posted that talks exactly about what is new and different.

    We just cannot find a common ground , so lets agree to disagree.
     
  18. the54thvoid

    the54thvoid

    Joined:
    Dec 14, 2009
    Messages:
    6,483 (2.28/day)
    Thanks Received:
    5,755
    Location:
    Glasgow - home of formal profanity
    From the very article i quoted.

    Vega is a development on from but using the base of GCN. They have tweaked it massively but it was the base they used. If that means it is a new architecture, okay. We can agree but my semantical autism rejects completely the notion we can call something a new architecture when it is using the old one as a base to build upon.

    This isn't like when AMD went from VLIW to GCN. Again, a fanatastic piece from Anandtech:

    http://www.anandtech.com/show/4455/amds-graphics-core-next-preview-amd-architects-for-compute/3

    I'm digressing now but this is what I mean when i say architecture change. GCN was created to maximise GPU compute against the very weak VLIW. Ironically at the same time Fermi was boiling over and Nvidia reduced compute to focus on massive efficiency gains (and later bifurcating the GPU plan into Gx100 and Gx102). GCN took the reigns of compute and VEga continues this same pattern and design similarity. The changes made to GCN over time, up to now iwth Vega are very much the same as the architectural improvements Nvidia has worked on since Fermi. Still similar arch but massively under the hood improvements.

    This is what I mean when I talk about architecture. A change in direction to address a design issue (gaming to compute, VLIW to GCN).

    Anyway - I hate arguing so you can be right in your definition in which case you need to also defend Pascal from folk that say it is just a faster Maxwell (it has architectural tweaks too). I will be right in my strict definition of architecture.

    And all this from a post about water cooling.
     
    xkm1948 says thanks.
  19. Vya Domus

    Vya Domus

    Joined:
    Jan 8, 2017
    Messages:
    1,225 (4.69/day)
    Thanks Received:
    613
    So then Zen and every new architecture from 1978 that is based onto x86 isn't really new is it ? I mean all these new CPUs have been tweaked massively but they have the same base , don't they ?

    You might stand by your definition , but that's not how the industry looks at this. Manufacturers are always going to maintain a base level of compatibility so the whole thing doesn't turn into a clusterfuck , which means there will never really be a completely "new architecture" because they always have to adhere to old designs.

    The way you think about this has two outcomes , either everything is new or nothing is. Troublesome in my opinion ,but , to each his own I suppose.
     
    Last edited: Aug 19, 2017
  20. r9

    r9

    Joined:
    Jul 28, 2008
    Messages:
    2,518 (0.75/day)
    Thanks Received:
    513
    Months before release of Vega somebody can't remember who, member of this forum made rough prediction based on rx 480 taking into consideration stuff like double the shaders and better memory that it will do same as gtx 1080.
    People give him bs quoting AMD marketing propaganda.
    Maybe we should revisit the old posts. :D
    I was that guy I would quote my self and rub it those people faces. lol

    No matter how you cut it they compared the Fury 60% higher clocks only 30% faster.
    They definitely made a change, for worse.
     
    Last edited by a moderator: Aug 21, 2017
    xkm1948 says thanks.
  21. jaggerwild

    Joined:
    Oct 30, 2008
    Messages:
    1,199 (0.37/day)
    Thanks Received:
    565

    SPOT ON LOLZ! @ TOMS fail site!
     
  22. Vya Domus

    Vya Domus

    Joined:
    Jan 8, 2017
    Messages:
    1,225 (4.69/day)
    Thanks Received:
    613
    You must be the 20th person that thinks I am arguing that Vega is incredible or something. OK :)

    Are we all truly blind ? You think I did not see the benchmarks ?
     
  23. r9

    r9

    Joined:
    Jul 28, 2008
    Messages:
    2,518 (0.75/day)
    Thanks Received:
    513
    Don't know. Are you, do you ? lol

    Look I really wanted for Vega to be success, but its AMD's biggest fail.
    Huge chip, expensive memory and power hungry.

    On the subject of power consumption they did exactly what they did with with Ryzen.
    Once they were ready they saw the numbers and had to make a decisions sell bad overclocker or poor chip.
    They made the right choice they bumped the clocks highest they could leaving no headroom for oc but that was fine cuz the stock number looked ok.
    Same thing here they are running at higher volts and frequency just so they can at least beat the 1080.
     
    xkm1948 says thanks.
  24. Vya Domus

    Vya Domus

    Joined:
    Jan 8, 2017
    Messages:
    1,225 (4.69/day)
    Thanks Received:
    613
    So why did you feel the need to say that, if you don't know ? :kookoo:

    Describes any comment about this subject really.
     
  25. RejZoR

    RejZoR

    Joined:
    Oct 2, 2004
    Messages:
    11,896 (2.51/day)
    Thanks Received:
    5,508
    Location:
    Europe
    @the54thvoid

    No, GPxxx are very much similar cores as they are based of the same thing, just different basic elements arrangement and clocks. Vega is NOTHING like Fiji. Same arrangement of basic units, but the rest in between, underneath and above is totally different.
     
    Vya Domus says thanks.
    10 Year Member at TPU

Currently Active Users Viewing This Thread: 1 (0 members and 1 guest)