R9 290 series specifications

[sushiwarrior]

There shouldn't be any "boost" states, there's a smooth gradient of clocks (anything is possible) using 7 DPM states and getting the highest clock by maximizing the percentage of time spent in each DPM state.

Top state is something like 1050-1070 on 290X, but it won't hit that without a hike in the power budget (temps are the limit during steady state, but not for max boost).

 

[ShintaiDK]

That is EXACTLY WHAT I DESCRIBED EARLIER. 13mhz bin adjustments when you pass temperature or TDP thresholds. You're still above boost clocks even per that chart. You remember what the stock GTX 680 boost clock was, right?
This is what I posted earlier:



The GTX 680 stock boost clock is 1058mhz. Unless your card outright FAILS (eg the fans suddenly stop working) it will be above the boost clockspeeds, this applies to both the Kepler and the GHZ 7970. Again, your assertion that the 280X is slower than the 7970GE is wrong. I really feel like you're obfuscating the facts just for arguing sake, to be honest. If you monitor your GPU clockspeeds with EVGA precision with the ingame overlay feature, i'm sure you know what's up.

Who said it was a stock? And look at the second chart. Boost speeds are simply not something that works for prolonged gaming for regular setups.

 

[thilanliyan]

But as I understand, we can just ignore everything else as performance will be at boost speed (or above) unless the game/program don't need the horsepower.

To me it seems like it is the opposite. As I understand it if the game/scene is really demanding, the temp/power draw increases, and so the clock goes DOWN, just when the higher clocks are really needed. Am I missing something?

 

[blackened23]

There shouldn't be any "boost" states, there's a smooth gradient of clocks (anything is possible) using 7 DPM states and getting the highest clock by maximizing the percentage of time spent in each DPM state.

Top state is something like 1050-1070 on 290X, but it won't hit that without a hike in the power budget (temps are the limit during steady state, but not for max boost).

So what you're saying is that AMD made their boost more like nvidia's? Because what you're saying did not and does not happen with the 7970GHz - the 7970GHz goes to the boost speed clock and that's it. Done. Stays there.

 

[ShintaiDK]

Agreed, it confuses me too.

But as I understand, we can just ignore everything else as performance will be at boost speed (or above) unless the game/program don't need the horsepower.

Unfortunately its not how it works. Example with the Titan and Metro 2033. First ~2 minutes it runs full boost, next ~2 it runs reduced boost and from there on its baseclock only.

 

[thilanliyan]

Unfortunately its not how it works. Example with the Titan and Metro 2033. First ~2 minutes it runs full boost, next ~2 it runs reduced boost and from there on its baseclock only.

Yeah that's as I understand it works as well. How useless boost really is then.

 

[blackened23]

Unfortunately its not how it works. Example with the Titan and Metro 2033. First ~2 minutes it runs full boost, next ~2 it runs reduced boost and from there on its baseclock only.

I still think you're obfuscating facts. With Kepler GPUs, the boost you get in games is always, generally speaking, way higher than advertised boost. My GPU has an advertised boost of 1020mhz. In games it shoots past 1100mhz. It may dynamically adjust downwards from there, but it is ALWAYS above the advertised boost. Again, if you use your ingame overlay you would be well aware of this - i'm sure that your GPU, as well, boosts well above advertised. It may dynamically adjust downwards from there but it will always be above the advertised boost. Which is good for performance, obviously.

I have never, ever seen my Kepler GPUs go below advertised boost speeds. And I use the overlay in all of my games. What DOES happen is the boost from the get-go of launching a 3d application is way higher than what your GPU advertises. I'm sure other Kepler users here can collaborate that fact.

The point here, no offense, I think you're flat out wrong when stating the 280X is slower than the 7970GE. It should be a very slight smidgen faster. The speed increase over the 7970 will be unimpressive for sure, but nonetheless it will be faster in actual games. That is, unless sushiwarrior is correct in that AMD dramatically changed their version of Boost.

 

[ShintaiDK]

So what you're saying is that AMD made their boost more like nvidia's? Because what you're saying did not and does not happen with the 7970GHz - the 7970GHz goes to the boost speed clock and that's it. Done. Stays there.

We'll see tomorrow if what you're saying has any basis in reality. I greatly look forward to seeing if you're the worlds biggest troll or if you work for AMD...

But your HD7970 is not reporting how it boosts. AMD made sure that wouldnt happen to avoid the situation I can show with nVidia.

 

[blackened23]

But your HD7970 is not reporting how it boosts. AMD made sure that wouldnt happen to avoid the situation I can show with nVidia.

Erm? You can monitor clockspeeds 24/7 with MSI afterburner or EVGA precision with both Kepler and AMD GPUs. There is never a situation where you WONT know the clockspeeds unless you willfully just don't want to see it. No idea what you're talking about here. Watching clockspeeds 24/7 within 3d applications is not an issue. You can always see what your "Boost" is doing.

You should use the overlay feature of EVGA precision. If you use an ingame overlay with a 7970GE it will stick at 1050mhz all the time. Similarly, using an ingame overlay with a Kepler card will show it being at boost speeds or higher 100% of the time, unless your fans fail. I think you're confused by the fact the Kepler on average boosts way higher than what is advertised on the box - every kepler GPU i've owned has done this. Make no mistake, this is a GOOD thing. It's more performance for free - the Kepler boosts to the highest sustainable clockspeed possible, generally speaking, even if it is higher than "stock" or what your GPU advertises. Even though it may dynamically adjust downward by 1-2 bins (it never adjusts more than that, in my experience - 1 bin is 13mhz , BTW) you will still be well above the advertised boost speeds.

 

[rainy]

We'll see tomorrow if what you're saying has any basis in reality. I greatly look forward to seeing if you're the worlds biggest troll or if you work for AMD...

For sure ShintaiDK doesn't work for AMD, he bashing them on regular basis - just look at history of his posts.

Infraction issued for callout.
-- stahlhart

 

[sushiwarrior]

So what you're saying is that AMD made their boost more like nvidia's? Because what you're saying did not and does not happen with the 7970GHz - the 7970GHz goes to the boost speed clock and that's it. Done. Stays there.

We'll see tomorrow if what you're saying has any basis in reality. I greatly look forward to seeing if you're the worlds biggest troll or if you work for AMD...

Not quite like Nvidia's, it is a bit more advanced I would say. Takes WAY more than just current temp and power into account (current die activity, predicted temperature, other "estimations" like that). It also is much more fine grained than working in 13MHz bins.

Unfortunately its not how it works. Example with the Titan and Metro 2033. First ~2 minutes it runs full boost, next ~2 it runs reduced boost and from there on its baseclock only.

More or less similar with the 290X, boost capability is entirely dependant on what you set the fan speed limit and power draw limit to, not really on silicon. Give it +x% power limit and x% max fan speed and it can stay in top clock bin indefinitely.

I still think you're obfuscating facts. With Kepler GPUs, the boost you get in games is always, generally speaking, way higher than advertised boost. My GPU has an advertised boost of 1020mhz. In games it shoots past 1100mhz. It may dynamically adjust downwards from there, but it is ALWAYS above the advertised boost. Again, if you use your ingame overlay you would be well aware of this - i'm sure that your GPU, as well, boosts well above advertised. It may dynamically adjust downwards from there but it will always be above the advertised boost. Which is good for performance, obviously.

I have never, ever seen my Kepler GPUs go below advertised boost speeds. And I use the overlay in all of my games. What DOES happen is the boost from the get-go of launching a 3d application is way higher than what your GPU advertises. I'm sure other Kepler users here can collaborate that fact.

The point here, no offense, I think you're flat out wrong when stating the 280X is slower than the 7970GE. It should be a very slight smidgen faster. The speed increase over the 7970 will be unimpressive for sure, but nonetheless it will be faster in actual games. That is, unless sushiwarrior is correct in that AMD dramatically changed their version of Boost. We'll see tomorrow if he has good information or not - I have strong doubts on that, but we'll see.

I can't say for sure but Tahiti XTL was floating around, which is a cut down (256 bit) Tahiti, 280X could be the XTL. However I consider that somewhat unlikely.

 

[blackened23]

For sure ShintaiDK doesn't work for AMD, he bashing them in regular basis - just look at history of his posts.

I was referring to Sushi. He mentioned that AMD basically changed their boost algorithm entirely. We'll see, I suppose.

I don't know if he means it changed with just the 290 or if that also applies to the 280X.

 

[Face2Face]

Overclocking current AMD parts is pretty much equivalent to old school GPU overclocking, of course with the added concern of vrm temps, something to my knowledge current Geforce owners (600 & 700 series) don't need to worry too much about. I haven't owned a GPU boost 1.0 or 2.0 card, but I am a fan on how it operates from what I have seen, especially with the added features of GPU boost 2.0 . I am interested to see if AMD changed their boost tech with the newer series of cards.

Edit - From what Sushi is saying, it sounds like it will....??

 

[blackened23]

Overclocking current AMD parts is pretty much equivalent to old school GPU overclocking, of course with the added concern of vrm temps, something to my knowledge current Geforce owners don't need to worry too much about. I haven't owned a GPU boost 1.0 or 2.0 card, but I am a fan on how it operates from what I have seen, especially with the added features of GPU boost 2.0 . I am interested to see if AMD changed their boost tech with the newer series of cards.

This I definitely agree with. Nvidia's GPU Boost 2.0 (and even 1.0) is flat out way better and more versatile than AMD's GPU boost. I was a skeptic initially but i'm a big fan now - the Kepler will generally automatically boost to 95% of it's highest performance potential, and you may be able to OC higher from there. I love it.

It also has a side effect of creating better efficiency and preventing card failures, which is nice in the long run, as well.

 

[ShintaiDK]

Yeah that's as I understand it works as well. How useless boost really is then.

While the effect varies for endusers. One cant stop thinking that benchmarks would usually be run within that timeframe.

 

[ShintaiDK]

Erm? You can monitor clockspeeds 24/7 with MSI afterburner or EVGA precision with both Kepler and AMD GPUs. There is never a situation where you WONT know the clockspeeds unless you willfully just don't want to see it. No idea what you're talking about here. Watching clockspeeds 24/7 within 3d applications is not an issue. You can always see what your "Boost" is doing.

You should use the overlay feature of EVGA precision. If you use an ingame overlay with a 7970GE it will stick at 1050mhz all the time. Similarly, using an ingame overlay with a Kepler card will show it being at boost speeds or higher 100% of the time, unless your fans fail. I think you're confused by the fact the Kepler on average boosts way higher than what is advertised on the box - every kepler GPU i've owned has done this. Make no mistake, this is a GOOD thing. It's more performance for free - the Kepler boosts to the highest sustainable clockspeed possible, generally speaking, even if it is higher than "stock" or what your GPU advertises. Even though it may dynamically adjust downward by 1-2 bins (it never adjusts more than that, in my experience - 1 bin is 13mhz , BTW) you will still be well above the advertised boost speeds.

Show me an example where an AMD card doesnt go between full boost or no boost for the reported frequency. But rather 4Mhz incrementals as AMD stated.

 

[GlacierFreeze]

I was referring to Sushi. He mentioned that AMD basically changed their boost algorithm entirely. We'll see, I suppose.

I don't know if he means it changed with just the 290 or if that also applies to the 280X.

It was covered about a couple months ago, in official AMD released slides on the new gen stuff. Yes, they changed their boost states and how they work quite a bit. Some likened it to be "more like" Nvidia compared to how it was done in 7000 series.

Not sure if it'll be in 280x or not, since they're basically rebadges. Wouldn't doubt they implement it in new series to make them more interesting than what they're replacing. We'll have to wait and see for sure, unless he confirms. And for a while, Sushi has posted lots of good info. He's far from a troll. One of the few people that knows what he's talking about.

 

[thilanliyan]

It also has a side effect of creating better efficiency and preventing card failures, which is nice in the long run, as well.

I have been watercooling, overvolting, and overclocking ATI/nV video cards since my X1800XL, and I've not once killed a card that way. Admittedly I don't tinker as much as I used to but I hope we don't completely lose the ability to control voltage/clocks like we've been able to.

 

[wand3r3r]

I agree with Shintai here. Drop the boost bs which is ideal for benchmarks but deceptive in reality. I'd take the old set clocks any day as I can determine what's stable and keep it steady.

 

[blackened23]

Show me an example where an AMD card doesnt go between full boost or no boost for the reported frequency. But rather 4Mhz incrementals as AMD stated.

I don't have a 7970 anymore to test (sold them a long time ago) but that is how boost worked when it was introduced with AMD cards. If it changed recently, it wasn't publicized. I just disagree with you in the strongest terms about base clock being relevant - I mean, the 280X is not an impressive jump over the 7970GE at all. We can both agree on that. But it should be a slight smidgen faster because of the boost clock unless AMD DRAMATICALLY changed how their boost behaves. In the past, their boost basically used 100% full speed in 3d apps and it stuck there. Boost clock really is the only clockspeed that matters unless AMD dramatically changed it. That is possible, but I don't know if AMD changed it only on the 290 series or if that applies to the 280 series as well.

Kepler boost is different in that your boost will, on average, be way higher than what your card advertises and it MAY adjust downwards by 13-26mhz, but you'll still be above your advertised boost IF the 3d application requires the horsepower. In this respect, nvidia's GPU Boost is about a million times better than AMD's boost. Nvidia's boost increases GPU longevity and prevents user error/card failure. AMD's boost does not do this (unless it was changed).

Anyway, we'll see tomorrow, I suppose.

 

[ShintaiDK]

Why would they spec the card as 870/1070 and not 1000/1070 for example if it was so certain? It just doesnt add up.

And as I showed with the Titan, play 4 minutes of Metro 2033 and all your boost is gone.

 

[blackened23]

If I had to guess, to lower 2d power consumption. The base clock *is* used quite frequently on the desktop in 2d apps.

I'm just speculating, though, like I said I don't know if AMD's boost has changed.

 

[thilanliyan]

I agree with Shintai here. Drop the boost bs which is ideal for benchmarks but deceptive in reality. I'd take the old set clocks any day as I can determine what's stable and keep it steady.

QFT, but these days people who like to tweak are not the lowest common denominator nor the largest market, which nV and AMD have to cater to I think.

 

[SiliconWars]

I agree with Shintai here. Drop the boost bs which is ideal for benchmarks but deceptive in reality. I'd take the old set clocks any day as I can determine what's stable and keep it steady.

That cat is already out of the bag, no putting it back in I'm afraid.

 

[boxleitnerb]

What I'm hoping for with boost technology is that at some point specific parts of the GPU are boosted while others are not (or even are set to lower clocks). If the GPU notices that there is an ALU bottlenecked situation, it might increase ALU clocks but at the same time lower memory clocks, ROP clocks, TMU clocks etc.