Fury X voltage adjustment now available

[tential]

The interposer is just a method of attaching it to the board. This is more an issue with the GPU's internals and their ability to clock. The board's VRM should be more than enough to handle the electrical requirements; the cooling is "sufficient" (at the very least better than typical air cooling); and the voltage can now be turned up.

I'm quite surprised that HBM overclocking has any effect at all. What the hell is it doing with all of that bandwidth? Though, as mentioned above, could it be a latency problem?



This is souring AMD's image for me. I stopped following the manufacturers and latest tech back in 2011 or so and only really started looking closely again in March of this year. This gave me a fairly clean slate with which to evaluate them again.

Thus far Nvidia came out with the Titan X and 980 Ti, somewhat quietly for each one. No outlandish claims. AMD came out with the Fury X quite loudly with lots of claims that I have yet to see realized. From a marketing standpoint, AMD has been a hype machine.

It's hurt my perception to the point where I'm not getting an r9 290 priced below 200 lol. I could get fury x performance for 400 (in the games I play) and have a locked 60 fps minimum but amd just has disappointed me so hard that I'm going to wait. I'll wait til next year and see how both companies match up.

Amd having no card capable of 4k down sampling from within drivers is killing me too. I don't care if the cards can't do it (I'm not playing the Witcher 3 sorry so for me down sampling on cheaper cards is useful when I'm playing older titles).

I don't want a gtx 970 but the more I think about it, it's the only card that fits my requirements right now. I don't want to pay that price though it's not worth it the performance I want is possible cheaper it's amd blocking 4k down sampling on the 290x/290. That reason alone is enough for me to not get their cards it's ridiculous the only 4k down sampling cards are either tonga or fury...

Amd clearly doesn't care about the things I do as a user so if they don't get their act together next year I'm doubling down and going flagship nvidia. Since I wanted fury x after all and the 980ti is far more close to what I want performance wise without the wce and if amd can't give me what I want maybe nvidia/Intel will be the only pc game in town soon

 

[soccerballtux]

Lol no. It was already rumored the 980ti was coming before we saw and fury x perf leaks. In fact no one truly knew fury x as amd kept an extremely tight lid on perf leaks.

980ti came out because that's what nvidia does. They follow up the premium card with something just as fast but cheaper.

Even if 980ti came out 1 month after fury x we still wouldn't rave about fury for that month. Look at how this release happened lol.... 980ti could have releases 1 month after fury x and still been amazing since fury x didn't even have voltage mod oc results up.

Spin it however you want this launch sucked. Just sucked. And I went into this launch expecting amd to do just ok..... Not this though.... Honestly the blunders during this launch were so basic they weren't on my radar for ways amd could screw up.

well, you're half off the mark from what I was talking about. If 980Ti hadn't come out, AMD would be back on top. It wouldn't have been a blunder, it would have simply been 'new tech, doesn't overclock, but it's 40% faster than a 980, TitanX performance for $400 cheaper etc., so who cares--next gen it will overclock, but for now it's an awesome buy'

I think you're the one doing the spinning here. And anyways, do you not think NVidia could have a way to get information aside from AMD leaks?

 

[soccerballtux]

It's hurt my perception to the point where I'm not getting an r9 290 priced below 200 lol. I could get fury x performance for 400 (in the games I play) and have a locked 60 fps minimum but amd just has disappointed me so hard that I'm going to wait. I'll wait til next year and see how both companies match up.

Amd having no card capable of 4k down sampling from within drivers is killing me too. I don't care if the cards can't do it (I'm not playing the Witcher 3 sorry so for me down sampling on cheaper cards is useful when I'm playing older titles).

I don't want a gtx 970 but the more I think about it, it's the only card that fits my requirements right now. I don't want to pay that price though it's not worth it the performance I want is possible cheaper it's amd blocking 4k down sampling on the 290x/290. That reason alone is enough for me to not get their cards it's ridiculous the only 4k down sampling cards are either tonga or fury...

Amd clearly doesn't care about the things I do as a user so if they don't get their act together next year I'm doubling down and going flagship nvidia. Since I wanted fury x after all and the 980ti is far more close to what I want performance wise without the wce and if amd can't give me what I want maybe nvidia/Intel will be the only pc game in town soon

NVidia cards have been aging very poorly lately, and you'd have to trust NVidia isn't going to just nuke the 970 performance in a year with a driver update. Coming from the company that did this in the first place to begin with, and that routinely does stuff like this, that's a pretty foolish decision of trust to make.

you're kinda a bad sport to support a company that doesn't play fair. If they didn't, AMD might not be in the spot they're in now...

 

[poofyhairguy]

you're kinda a bad sport to support a company that doesn't play fair.

Even if that company has the product he wants?

 

[soccerballtux]

Even if that company has the product he wants?

plenty of women want an endangered species fur coat. many people want things that are bad for the world

 

[Sabrewings]

you're kinda a bad sport to support a company that doesn't play fair. If they didn't, AMD might not be in the spot they're in now...

Oh, yes. It's the consumer's fault and not AMD's. Damn consumers, always consumering.

 

[stahlhart]

plenty of women want an endangered species fur coat. many people want things that are bad for the world

Stop derailing this discussion.
-- stahlhart

 

[Elixer]

Drat.. ignore this.

 

[Headfoot]

Very strange how much mileage they are getting out of the supposedly overkill bandwidth of HBM. Id be interested to see if it really is latency related.

AMD should not have played up "overclockers dream." It overclocks to the same spot as Hawaii which needs to be said... a little over 1200. Which is respectable. For some reason though it seems to scale much worse than Hawaii did.

Either this thing really is limited in DX11 or it was too unbalanced of a design for their first max sized die.

 

[poofyhairguy]

Oh, yes. It's the consumer's fault and not AMD's. Damn consumers, always consumering.

I am not trying to blame AMD for anything. They have to make guesses and take risks like anyone else.

To me (especially now that we know the underclock potential) Fury looks like a part I design if I am trying to milk the GPU compute crowd. Given all its power and watercooling if we were still back in a GPU mining phase this card would be king. Seeing as how it was probably designed back when AMD was selling a ton of tahiti's for mining it makes sense that it has compute power that gets bottlenecked elsewhere. Now that we know that overclocking (or ramping down the FP64 to 1/16) can't overcome the bottleneck and unleash the raw compute power for gaming we are left to hope that future games can do it by design.

Nvidia's Maxwell came from mobile. If it wasn't for Tegra and all that diddling around with SoCs then maybe right now we are playing on improved Kepler instead. They took a risk to move into the mobile market, and even though that didn't pay off in mobile (few major design wins) that is paying off in the GPU market. You just never know.

I think whatever follows Fury will be a real winner. AMD took some big steps that will pay off when the node shrink comes.

 

[3DVagabond]

You're right, but I was paraphrasing the article. Looks like wizard did a ninja edit, because at the end where he says "Looking at the numbers, I'm not sure if a 150W power draw increase, just to get an extra 3 FPS, is worth it for most gamers," it did say 200W.

That's cool. Easy enough to do. I quit reading the commentary well before the end of the article.

It doesn't surprise me with the way their articles have been lately.

 

[tential]

NVidia cards have been aging very poorly lately, and you'd have to trust NVidia isn't going to just nuke the 970 performance in a year with a driver update. Coming from the company that did this in the first place to begin with, and that routinely does stuff like this, that's a pretty foolish decision of trust to make.

you're kinda a bad sport to support a company that doesn't play fair. If they didn't, AMD might not be in the spot they're in now...

I'll expand later you've made a lot of good points I'm just a little busy but yes I agree with the 970 it's why i purchased nothing.

I can't support amd removing 4k downsample on the 290x and the nvidia side is the 970....i mean I've written enough about how poor a value choice that card is to fill 3 books. I can't now go get one for 1 freaking feature. It just sucks that in order to downsample from 4k I can't buy a card I want. I have to pick between cards I don't want.... If I could pick up a used fury at 300-350 next year though the I don't care haha. Actually a used r9 nano would at a good price would make me happy. If the r9 nano did actually come out at the 450 mark I'd buy it probably. Depending on crossfire performance I dunno why that card in crossfire excites me a lot. Performing at oc(power consumption) levels of high end cards while consuming far less power with amds amazing cf scaling.... Yes it's very enticing!

 

[railven]

So I finally got to read the whole article.

It didn't mention anything about noise, so I'll assume he got a good pump. Woof, that radiator can cool! Is making me feel more confident about going with the water cooler hybrid option.

I was starting to wonder if this is perhaps due to non-official OCing methods? Granted, not that most OCing methods are official, but they often have some type of vendor support. (Does W1zard work for someone? I think I read some where in the thread/comments about a Trixx software update/option).

Well, I guess now we wait and see how the DX12 library and Win10 work out. So far every bullet point has been crossed out. That is just embarrassing. That many watts and barely a blip. AMD needs to fire their marketing team, for real this time. All of them. Overclocker's dream should have never slipped out. Unless their definition of an overclocker is "look at all them watts!"

Stick a CLC on Hawaii XT/Grenada XT (whatever is the official name), unlock everything and sell it for $400-450. You will win the "I don't care about power consumption" crowd without even trying. At least there the extra wattage are doing something.

 

[x3sphere]

I was starting to wonder if this is perhaps due to non-official OCing methods? Granted, not that most OCing methods are official, but they often have some type of vendor support. (Does W1zard work for someone? I think I read some where in the thread/comments about a Trixx software update/option).

He works on the Trixx software for Sapphire yes.

 

[railven]

He works on the Trixx software for Sapphire yes.

Ahhh guess it makes it as official as can be. Makes me wonder if they'll release this to users. Not sure if putting high power/heat which it seem he was even against is worth it.

Edit seeing in the thread he said software be out to users soon.

 

[Silverforce11]

Some of you guys have funny expectations, GCN in every single form so far as typically reached 1.2ghz with some good samples hitting 1.25ghz with vcore OC. This isn't actually surprising at all, its the upper limits of GCN. It scales good as long as the memory is OC along with the core, that's the surprising thing since other Tahiti & Hawaii did not exhibited this behavior. My R290s got linear perf gains just on a core OC since one of them the vram wouldn't even OC by a tiny bit (elpida ram).

I think the scaling will be just fine in most games like other GCN GPUs have shown, but 1.2ghz is a realistic expectations with vcore, expecting 1.25ghz is like expecting every 980/Ti to get 1.5ghz+ which plenty of reviews don't even get. It's never been worth it for myself to push 7850 (50% OC) 7950 (45% OC), 7970 (35% OC), R290s (25% OC) with a high OC with +vcore for daily usage, the power use goes up really fast making the perf/w horrendous.

The importance of vcore mod is the undervolt ability, which seems to behave exactly the same as other GCN GPUs as well, with even a possibility to run higher clocks at less vcore to use less power for a slight gain in performance. The efficiency gains are excellent, -50W, +5% performance. If we're assuming its 250W gaming load, it's now 200W, 5% faster, it just got ~30% more perf/w.

Edit: The scaling is actually fine, as long as vram is boosted:

Stock -> 1160mhz = ~10% OC
Perf 50.9 fps -> 55.1 fps = ~9% perf gains.

Thus, best results are either an OC with no +vcore or a smaller OC with -vcore. Either way, it's not a massive overclocker like Maxwell 2 BUT it can gain massive perf/w, so it can beat the 980Ti/Titan X on that metric (been awhile, can NV GPUs undervolt?). Funny ain't it?! lol

 

[hawtdawg]

So, there is clearly some sort of issue with the HMB here. Is there anyway to test and see if the bandwidth is actually anywhere close to what is claimed?

 

[tential]

So, there is clearly some sort of issue with the HMB here. Is there anyway to test and see if the bandwidth is actually anywhere close to what is claimed?

It's clear AMD doesn't know their own product.

This is not an OC monster.
This is a card that scales with HBM OCing (So saying HBM OC was not included because it wouldn't be useful was a lie).

If NVidia did this you'd be calling for their heads...

 

[3DVagabond]

It's clear AMD doesn't know their own product.

This is not an OC monster.
This is a card that scales with HBM OCing (So saying HBM OC was not included because it wouldn't be useful was a lie).

If NVidia did this you'd be calling for their heads...

It's interesting how one random quote from one AMD employee has made the rounds so much and, surprise surprise, we have someone accusing AMD of lying, not knowing their own product.

Just about as amazing as one article where one game is benched at one resolution and settings has defined the O/C performance of the card. This is just the beginning. It's yet to be seen how it turns out.

 

[railven]

It's clear AMD doesn't know their own product.

This is not an OC monster.
This is a card that scales with HBM OCing (So saying HBM OC was not included because it wouldn't be useful was a lie).

If NVidia did this you'd be calling for their heads...

EDIT #2: Whooops, forgot what subsection I'm in. Nevermind.

OT: The power consumption for the performance gain is ridiculous. Why I asked if it would look better on other games? If W1zard works for Sapphire, I'd assume he were bias towards AMD?

 

[soccerballtux]

If NVidia did this you'd be calling for their heads...

no...not really...I think you're inflating things here.

I'll say I think it's halfway towards 'calling for their heads'--it was a mistake-- but it most definitely was NOT premeditated, which is what is required before 'calling for heads' is appropriate. IMO.

 

[soccerballtux]

Some of you guys have funny expectations, GCN in every single form so far as typically reached 1.2ghz with some good samples hitting 1.25ghz with vcore OC. This isn't actually surprising at all, its the upper limits of GCN. It scales good as long as the memory is OC along with the core, that's the surprising thing since other Tahiti & Hawaii did not exhibited this behavior. My R290s got linear perf gains just on a core OC since one of them the vram wouldn't even OC by a tiny bit (elpida ram).

I think the scaling will be just fine in most games like other GCN GPUs have shown, but 1.2ghz is a realistic expectations with vcore, expecting 1.25ghz is like expecting every 980/Ti to get 1.5ghz+ which plenty of reviews don't even get. It's never been worth it for myself to push 7850 (50% OC) 7950 (45% OC), 7970 (35% OC), R290s (25% OC) with a high OC with +vcore for daily usage, the power use goes up really fast making the perf/w horrendous.

The importance of vcore mod is the undervolt ability, which seems to behave exactly the same as other GCN GPUs as well, with even a possibility to run higher clocks at less vcore to use less power for a slight gain in performance. The efficiency gains are excellent, -50W, +5% performance. If we're assuming its 250W gaming load, it's now 200W, 5% faster, it just got ~30% more perf/w.

Edit: The scaling is actually fine, as long as vram is boosted:

Stock -> 1160mhz = ~10% OC
Perf 50.9 fps -> 55.1 fps = ~9% perf gains.

Thus, best results are either an OC with no +vcore or a smaller OC with -vcore. Either way, it's not a massive overclocker like Maxwell 2 BUT it can gain massive perf/w, so it can beat the 980Ti/Titan X on that metric (been awhile, can NV GPUs undervolt?). Funny ain't it?! lol

so the scaling isn't horrible, but I think we were all expecting to get 30% overclocks. And that is horrible.

 

[Silverforce11]

so the scaling isn't horrible, but I think we were all expecting to get 30% overclocks. And that is horrible.

Why would you expect 30% OC?

You didn't pay attention to every other GCN GPUs out there? They peak at 1.25ghz. They don't peak at 1.4ghz. Thats completely unrealistic expectations.

I posted about this on launch, I said with vcore, it should get 1.2 to 1.25ghz and that's what happened. It's GCN, its from TSMC, why would this one be much better than Tahiti or Hawaii?

 

[Silverforce11]

It's clear AMD doesn't know their own product.

This is not an OC monster.
This is a card that scales with HBM OCing (So saying HBM OC was not included because it wouldn't be useful was a lie).

If NVidia did this you'd be calling for their heads...

It is an OC monster. Look at how much power it eats up with that OC.

 

[railven]

Edit: The scaling is actually fine, as long as vram is boosted:

Stock -> 1160mhz = ~10% OC
Perf 50.9 fps -> 55.1 fps = ~9% perf gains.

Thus, best results are either an OC with no +vcore or a smaller OC with -vcore. Either way, it's not a massive overclocker like Maxwell 2 BUT it can gain massive perf/w, so it can beat the 980Ti/Titan X on that metric (been awhile, can NV GPUs undervolt?). Funny ain't it?! lol

EDIT: Derp, you said it right there long as VRAM is boosted too. Ignore the rest


I assumed you included the Memory OC in there? Or are we just talking about core? Because just core, at 1160 it's closer to 53.5 if you ask me.

53.5/50.9 == ~5% gain.

Seems that other half the final ~9% comes from the memory OC.

But man, at those volts for that perf gain, woof.