Fury X voltage adjustment now available

[Silverforce11]

The problem with that TPU test is that they tested it with only one game, which completely don't react to AMD hardware. I would like to see what OC does in Far Cry 4, for example.

No, their testing show vcore OC scales poorly without memory OC. Together, it scales linearly, almost perfect.

 

[Glo.]

Me, unfortunately, neither. I just don't think that OC scaling on GCN cards is that perfect. What I mean is that it should bring more performance, from frequency gained. Maybe its just one game, maybe not.

Also its worth noting the small difference between Fury and Fury X GPUs in games despite the fact that Fury is cut down and downclocked version of full Fiji chip.

 

[Silverforce11]

Me, unfortunately, neither. I just don't think that OC scaling on GCN cards is that perfect. What I mean is that it should bring more performance, from frequency gained. Maybe its just one game, maybe not.

Also its worth noting the small difference between Fury and Fury X GPUs in games despite the fact that Fury is cut down and downclocked version of full Fiji chip.

How much more gains do you want? 9% for a 10% OC is pretty good.

http://forums.anandtech.com/showpost.php?p=37587841&postcount=41

The only unexpected result is that it needs vram OC to scale since we all would have thought, that's a ton of bandwidth already.

 

[tential]

How much more gains do you want? 9% for a 10% OC is pretty good.

http://forums.anandtech.com/showpost.php?p=37587841&postcount=41

The only unexpected result is that it needs vram OC to scale since we all would have thought, that's a ton of bandwidth already.

How much do you think an OCers dream is? And tell me that amount.

That amount is enough.

 

[railven]

How much do you think an OCers dream is? And tell me that amount.

That amount is enough.

I think you're asking the wrong person, because he already said the most important part of volt control is undervolting.

So while we're foaming at the mouths to go forward, he wants to go backwards. And seeing as he lives in Australia where the khw is I believe already double what mine is - I don't blame him.

 

[tential]

I think you're asking the wrong person, because he already said the most important part of volt control is undervolting.

So while we're foaming at the mouths to go forward, he wants to go backwards. And seeing as he lives in Australia where the khw is I believe already double what mine is - I don't blame him.

I understand what you're saying...
But AMD said it's an OCer's dream. If AMD is saying the OCer's dream is to undervolt cards then I'll know to never trust any OC information from them again.

 

[bystander36]

I understand what you're saying...
But AMD said it's an OCer's dream. If AMD is saying the OCer's dream is to undervolt cards then I'll know to never trust any OC information from them again.

He was also going on about how unfair it was to compare OC Fury (X) to the 980 (Ti) because of the lack of unlocked voltage for AMD. I guess it didn't make much difference after all, not that those comparisons had unlocked voltage on Nvidia.

As always, if you are a fan of either side, you find the bright side of any situation.

 

[tential]

He was also going on about how unfair it was to compare OC Fury (X) to the 980 (Ti) because of the lack of unlocked voltage for AMD. I guess it didn't make much difference after all, not that those comparisons had unlocked voltage on Nvidia.

As always, if you are a fan of either side, you find the bright side of any situation.

I think it was unfair to compare both without unlocked voltage. Definitely. You need to see that to make an informed decision. Now that I see it, I can make an informed decision.

I'm extremely unimpressed with Fury X OC after voltage is unlocked.

Either way, I'm curious how AMD can stand by their "OCers dream" given this scenario?

At least it's quiet though.

 

[railven]

I understand what you're saying...
But AMD said it's an OCer's dream. If AMD is saying the OCer's dream is to undervolt cards then I'll know to never trust any OC information from them again.

With the power consumption that was revealed, I'm pretty sure everyone is gonna take that as a load question.

I thought my HD 7970 was an OCing beast when it hit ~22% OC on stock volts alone. Then I tinkered with a little OV I was able to hit 1225 with +25mv. That was my, at the time, OC'ers dream.

I didn't have a means to test power consumption then, but it was an Accelero XTreme so temps were mid 60s which I was perfectly fine with.

 

[Silverforce11]

I've said several times before the vcore tool was out, Fiji should reach 1.2 to 1.25ghz with vcore, but power use will skyrocket. It behaves exactly as prior GCN SKUs do. Some of you act all surprised but anyone who has experience with GCN vcore mods knows it.

Fury X at best has about 15% OC headroom. It will lose to an OC 980Ti at 1440p. It will lose at 4K as well in single GPU configs. But it will still beat it in multi-GPU configs 4K.

With the undervolt and OC at 1125mhz saving power (as per TPU's results), in multi-GPU 4K, it will beat 980Ti while being more power efficient. So, not too bad for that high end usage scenario.

I still wouldn't buy it, needed a bigger performance leap to move from R290s.

 

[Silverforce11]

I just got 2x R290X ASUS custom models for a friend to upgrade and replace my 7950 rig. I wanted to test to see how well it underclocks.. here's the results, with Unigine Valley runs & GPU-Z monitoring.

First run, stock, no changes to power limit. It's a bit higher than my R290 but not much.

At -50% power limit. Note the average power drop from ~172W to 105W (%? I cant be bothered doing maths atm heh). Max power also drop from 226W to ~170W.

Performance drop? About 11%. Clocks ~880mhz down from 1ghz.

GPUZ of R290X

In games, with the -50% power limit, the fans don't even ramp up from idle speeds. Note that any -vcore on top of this, improves the results further.

 

[Kalessian]

^ That's actually kinda nice to know thinking about it from an HTPC perspective

 

[Makaveli]

^ That's actually kinda nice to know thinking about it from an HTPC perspective

agreed great post Silverforce11

 

[Headfoot]

Yeah very versatile architecture in GCN. I'm sure thats partially why the consoles went with GCN given how skittish both Sony and MS are about power consumption after the last gen of consoles. Looks like if you crank it down you can preserve a good portion of the performance but lose a lot of consumption. Provided of course this scaling holds true for Pitcairn and Bonaire as well as Hawaii (and I dont see why it wouldn't).

Great post.

 

[Headfoot]

Bottom line, Fury / X is a solid chip, good performance, decent overclocking. Hitting 1.2 ghz on average, historically speaking is still pretty impressive. I remember when getting chips to 1ghz on the 5850/70 was totally groundbreaking.

The problem is just that the 980 Ti is has faster performance and much better overclocking for the same price/not much more.

I wouldn't say its a case of Fury/X fail more of stolen thunder and extremely good execution by nVidia. If it was Titan X vs Fury X it would be different story. The 980 Ti was just an excellent preemptive strike. The 980 was a turd at its price. The 970 was good but the 290 was so much cheaper for nearly the same perf. The 960 is also eclipsed by the 290. But aftermarket 980 Ti is unmatched, truly the star of Maxwell 2.

Both architectures are immensely fascinating for the observer though. Seeing HBM in play is incredibly cool. And it's insanely impressive that nVidia can have a chip as wide as the 980 Ti hit 1600 with some regularity.

 

[Headfoot]

I look forward to the results! GPU Z is solid, but Kill A Watt would be even better. It'll be interesting to see how the GPUZ predicted usage differs from the wall measurement with the 50% powertune target

 

[chimaxi83]

I look forward to the results! GPU Z is solid, but Kill A Watt would be even better. It'll be interesting to see how the GPUZ predicted usage differs from the wall measurement with the 50% powertune target

Yeah I'll check it out later after work. Should be interesting. I've been undervolting lately because Crossfire heat is a very real thing, wonder what my power usage differences are.

I gotta be honest though. You have words directly from Joe Macri at AMD like "we built these for the overclockers", "you're gonna be able to overclock this thing like no tomorrow", "this is an overclockers dream". Exact quotes regarding Fury. Those claims really don't align with the reality. The overclock results are decent, until you factor in the associated power usage. It's just ok, nothing like what Joe made it seem like. Maybe I just misunderstood.

 

[Headfoot]

The overclocking is decent, 9% perf for 10% overclock. It looks worse compared to the 290 because Fury X starts at 1050 stock, if it were at 947 again it would be bigger % wise. Either way I agree it was really a bad call for them to talk up the overclocking so much when it was pretty much identical max OC to Hawaii which they didnt talk up at all

 

[raghu78]

How much AMD have fallen from the launch of HD 7970 way back in Jan 2012. The fundamental problem has been AMD's perf/transistor has been reducing. The HD 7970 was 4.3 billion transistors and today we see the Fury X with 8.9 billion transistors and HBM have lesser perf/transistor. Fury X cannot double the performance of R9 280X / HD 7970 Ghz but the transistor budget has more than doubled. Nvidia on the other hand has been able to keep better perf/transistor through architectural innovation. This lower perf/transistor has forced AMD to go for denser designs which has also affected the overclocking headroom of their chips. With 8.9 billion transistors Fury X is 596 sq mm while GTX Titan with 8 billion transistors is 601 sq mm. Right now AMD have a architectural problem and more importantly a R&D problem due to their massive drop in revenues and continual losses. It looks more and more that we will end up with a Nvidia GPU monopoly. Its frustrating to see this happen but the truth is that AMD is suffering for the mistakes of their CPU division and that wretched Bulldozer architecture.

I gotta be honest though. You have words directly from Joe Macri at AMD like "we built these for the overclockers", "you're gonna be able to overclock this thing like no tomorrow", "this is an overclockers dream". Exact quotes regarding Fury. Those claims really don't align with the reality. The overclock results are decent, until you factor in the associated power usage. It's just ok, nothing like what Joe made it seem like. Maybe I just misunderstood.

imo the overclock results are pathetic. Heck the R9 290X had better OC headroom than Fury X. Its now clear that AMD have lost the plot when it comes to designing competitive GPUs with good OC headroom. The transistor density has been consistently increasing from HD 7970 to Fury X and that has meant lower max OCs. The original HD 7970 had the best OC headroom and could easily hit 1300 Mhz with voltage overclocking and watercooling. Thats a 375 Mhz increase or more than 40% OC headroom. Fury X is pathetic. Its clocked much higher at 1050 Mhz and cannot hit 1200 Mhz. I was a strong AMD supporter but now I am of the opinion that this is a fundamentally sick company. I am not optimistic about its future anymore.

 

[railven]

The overclocking is decent, 9% perf for 10% overclock. It looks worse compared to the 290 because Fury X starts at 1050 stock, if it were at 947 again it would be bigger % wise. Either way I agree it was really a bad call for them to talk up the overclocking so much when it was pretty much identical max OC to Hawaii which they didnt talk up at all

I recall they did talk up Tahiti overclocking headroom. In fact, if I remember right it was featured right at the launch articles...

http://www.techpowerup.com/reviews/AMD/HD_7970/31.html

The overclocks listed in this section were achieved with the default fan and voltage settings as defined in the VGA BIOS. Please note that every single sample overclocks differently, that's why our results here can only serve as a guideline for what you can expect from your card.



Maximum stable clocks of our card are 1075 MHz core (16% overclock) and 1715 MHz Memory (25% overclock).

Both overclocks are spectacular. It's amazing to see over 15% overclocking potential on a brand-new GPU design, using a brand new production process. Over 1700 MHz memory clock is actually the highest memory speed we have ever seen!

Personally, I think Hawaii and now we're seeing Fiji, ran away with power consumption. Hawaii's lack luster thermal point in quiet mode versus Uber modes insane blower and Fiji just not having the tools to "spread its legs."

More so, I believe AMD knew Hawaii and Fiji were pigs.

 

[Elfear]

How much do you think an OCers dream is? And tell me that amount.

That amount is enough.

Overclocking on Fiji is pretty lackluster but you and Silverforce are talking about two different things. You're talking about max oc potential and he's talking about scaling (i.e. how much performance you gain from overclocking; a 10% clock increase would give a 10% performance increase with perfect scaling). Scaling looks to be really good with Fiji as long as memory is overclocked as well.

 

[Rvenger]

Thread cleaned and reopened.

Let's get this back on topic and stop the sarcasm and arguing. Remember this is the AMD subforum, intentionally bashing AMD in a trolling manner will earn you an infraction/vacation.

My apologies to the members who wanted to have a civil discussion.


Rvenger

Anandtech Super Moderator

 

[Makaveli]

Fury X cannot double the performance of R9 280X / HD 7970 Ghz

At what resolution are we talking about here?

1080p? I would think at 1440p and above the gap should be close to double?

 

[tential]

Overclocking on Fiji is pretty lackluster but you and Silverforce are talking about two different things. You're talking about max oc potential and he's talking about scaling (i.e. how much performance you gain from overclocking; a 10% clock increase would give a 10% performance increase with perfect scaling). Scaling looks to be really good with Fiji as long as memory is overclocked as well.

Ok, well I'd like to know the max OC potential of the card. It's watercooled and was advertised as an OCers dream. If the card scales on memory too then perhaps we need another review that has more games and shows a person who has spent the time to figure out what a good OC is as this card clearly needs memory too?

I'm just saying a lot of things don't add up and I really wish we could get some indepth reviews to figure these things out.

Why does Fury scale off HBM memory increases when it was supposed to have so much memory bandwidth to not even need OCing anyway as it'd be a pointless endeavor?

What OC's was AMD getting on their end to believe this to be an OCer's dream?

What are the 8K teaser shots of bioshock infinite? Is Fury X capable of this? Even just theoretically because maybe then crossfire Fury XX (or whatever it's going to be called), can handle it?

Since we do have people interested in undervolting, despite my complete disinterest in it, it's clear people are interested in and definitely should be looked into more as well.

Did Techpowerup have a time constraint when writing that Fury X OC article? Why are they the ONLY ones with an article up and only 1 game tested? It's been 4 days since is no one else interested in testing this?

Etc.

 

[poofyhairguy]

Performance drop? About 11%. Clocks ~880mhz down from 1ghz.

Wow. Kinda like made at home Fury Nano. Makes me wonder what the clocks of the dual GPU card will be. That card will stick at the top of most benchmarks for a while.