AMD: Southern Islands (7*** series) Speculation/ Rumour Thread

[boxleitnerb]

Obviously this is a different scene as the fps are quite different. A "game" consists of many scenes where limits may change on a moments notice. Another thing that this review lacks: Information. What was benched, how?

Btw, at these fps, microstutter is almost irrelevant anyway, especially if you don't like tearing and play with vsync on. 40-80 would be the range where it really matters in my opinion.

 

[jimbo75]

You're trying really hard without convincing me boxleitnerb.

Once again, this is not cpu-limited - the scaling isn't great but it rarely is to the 3rd and 4th cards. The important thing here *is* the lack of microstutter shown with 3 and 4 cards.

I would frankly take this information - regardless of how lacking it is in certain aspects - long before I'd trust anything out of [H], nevermind a review that includes their sponsor's cards losing on raw fps yet winning on "smoothness".

 

[boxleitnerb]

When the settings are demanding enough, scaling with 3 and 4 GPUs is quite good usually. So a CPU bottleneck is a reasonable explanation. Tom's should have at least checked by overclocking the CPU and looking at the fps increase.

I've been using 580 SLI for almost 18 months now and I surely know what a CPU bottleneck looks like, thank you. And that looks like one to me - it usually is with fps this high. Any way you look at it -> if the GPU load is low, the game engine dictates the output of the fps and that is much more even than what AFR can do. Has nothing to do with CF, it would be the same way with SLI, too. But alas, Tom's didn't look at that either.

Look here:



Introducing a CPU bottleneck by reducing CPU performance can smooth frametimes considerably.

What do you have against [H]? Just because someone takes an unconventional approach to benchmarking and correctly realizes that fps aren't everything with AFR, they are "sponsored"? Should they lie if SLI feels better and say it doesn't? I have never ever heard that CF is smoother than SLI. However, the other way around I have read it quite frequently.

 

[jimbo75]

That's because SLI is generally smoother - with 2 cards, and that's what is generally benchmarked.

 

[boxleitnerb]

There is no reason why this should be different with 3 or 4 GPUs - if demanding settings are used (like SGSSAA for example). Anyway, we are slightly off topic.

 

[Kaotik]

Which lag ?

If you are talking about Frame metering and purposly introduced averaging, then 20ms-20ms-20ms will feel smoother compared to 10ms-40ms-10ms.

If you are talking about input lag - input has nothing to do with rendering, and perceived input lag will again be more consistent in 1st case, and 2nd scenario will feel jerkier and not more responsive. Unless you magically chose to move your mouse inside that 10ms period.

Your statement that input lag has nothing to do with rendering is completely false - in fact it has everything to do with rendering - the longer it takes to render the frame and send it to the screen, the longer it takes your input to have effect on what happens on screen (and no, of course it isn't the only thing affecting and adding input lag))

You can't magically cut frame rendering times and just "average it", you can add lag to the frames being ready "too early" which is what nVidia does - you can't go from 10-40-10 to 20-20-20, but you can go to 20-40-20 or 30-40-30 or even 40-40-40 - the middle 40 is the one that can't get any faster than it is. And yes, it adds input lag to those frames too, even if would appear smoother input response in the end, it would be slower at those frames.
There's one big problem with this, though - aside from the input lag it creates, it also makes any benchmarks done with FRAPS and possibly others invalid. FRAPS inserts it's timestamp before the lag is added, meaning FRAPS thinks the frame got ready much sooner than it did.

Here's the techreport piece http://techreport.com/review/21516/inside-the-second-a-new-look-at-game-benchmarking/11

 

[ECH]

True, but [H] has always made it their task to actually play the games they benchmark (their playable setting approach). They also have a long history of testing this way and of testing SLI and CF.

I don't trust someone who makes such basic mistakes as THG as Gipsel correctly noted.

So in the end CF didn't experience any issues in the games tested do to the higher frame rates (as nothing specific was provided). However, noted that if frame rates were lower there could be issues.

 

[HMBR]

it seems there are many 7000 series sold with defective components, causing "black screens" and other instabilities

http://www.behardware.com/news/lire/27-10-2012/

 

[ECH]

it seems there are many 7000 series sold with defective components, causing "black screens" and other instabilities

http://www.behardware.com/news/lire/27-10-2012/

The article doesn't say 7000 series but the 7870.
And as noted in the article they reference small capacitors as the potential problem.
Pic

But the amendments made at the end of the article indicate that vendors are aware of the issue and they've either corrected the problem or looking into correcting it.

 

[HMBR]

The article doesn't say 7000 series but the 7870.
And as noted in the article they reference small capacitors as the potential problem.
Pic

But the amendments made at the end of the article indicate that vendors are aware of the issue and they've either corrected the problem or looking into correcting it.

well, 7870 is part of the 7000 series,
the effects of the problem are quite big, in most Internet forums you will find many users complaining about the issue, and it's not something new

http://forums.guru3d.com/showthread.php?t=364498&page=16

 

[Dave Baumann]

No, that particular issue was specific to 7870.

 

[HMBR]

No, that particular issue was specific to 7870.

it seems to be limited to the 7870, which translates to many 7000 cards,
but looking at the forums posts, I've seen some users of other cards complaining of similar problems (7970s even...) which of course can be caused by an unrelated problem, but still...
what I find strange is that this problem is clearly not new, and there are still a number of new cards being sold with the inadequate components (capacitors), and some people claim to have received cards from the RMA with the exact same problem!
selling bad cards like this is pretty bad, particularly for Sapphire.

 

[sonen]

Your statement that input lag has nothing to do with rendering is completely false - in fact it has everything to do with rendering - the longer it takes to render the frame and send it to the screen, the longer it takes your input to have effect on what happens on screen (and no, of course it isn't the only thing affecting and adding input lag))

You can't magically cut frame rendering times and just "average it", you can add lag to the frames being ready "too early" which is what nVidia does - you can't go from 10-40-10 to 20-20-20, but you can go to 20-40-20 or 30-40-30 or even 40-40-40 - the middle 40 is the one that can't get any faster than it is. And yes, it adds input lag to those frames too, even if would appear smoother input response in the end, it would be slower at those frames.
There's one big problem with this, though - aside from the input lag it creates, it also makes any benchmarks done with FRAPS and possibly others invalid. FRAPS inserts it's timestamp before the lag is added, meaning FRAPS thinks the frame got ready much sooner than it did.

Here's the techreport piece http://techreport.com/review/21516/inside-the-second-a-new-look-at-game-benchmarking/11

It's still not "input lag". I called it _perceived_ input lag, maybe we can call it rendering lag.
Also Nvidia's FM might not be able to convert 10-40-10 to 20-20-20, but I doubt 30-40-30 is best it can do. Namely I wouldn't be so sure they can't do nothing to minimize frame drops, and are simply smoothing frame times.

 

[Psycho]

what I find strange is that this problem is clearly not new, and there are still a number of new cards being sold with the inadequate components (capacitors), and some people claim to have received cards from the RMA with the exact same problem!
selling bad cards like this is pretty bad, particularly for Sapphire.

The articles says that AMD identified/fixed it long ago, but somehow Sapphire didn't get the memo (made the changes), unlike the rest. And it has been looking like a sapphire 7870 problem all along (mine, from more or less first batch, doesn't have any problems though). Of course you will always see "ghost reports" in such forum threads..

 

[Ethatron]

Also Nvidia's FM might not be able to convert 10-40-10 to 20-20-20 ...

It's 20-30-10, if you delay the first frame by 10ms then you lower the "lag" of the consecusive frame by 10ms. The sum of all times is constant (otherwise they would have invented a time-accelerator, or play back games at slow-mo).
That's the whole idea. Instead of instantanious display, they keep frames in a buffer and try to display them at locally equidistant intervals. The interval is unrelated to v-sync, so it's another form of frame-buffering, not the same.

You can track the changes of velocity of the frame-production statistically and use observations of general acceleration and decceleration to filter out the outliers. That means intelligent "l-sync" (lag-sync) would try to optimize every frame's moment of display to be the same time-span away from it's previous and succesive frame's moment. Think it like the automatic smoothing of a radial spring, the spring has this operation "built-in".

This can work more or less well if you contextualize (fit) the predictor on related API-calls (number of triangles pushed, number of outstanding GPU-to-CPU transfers, etc.), it can work even better if you give the developers the possibility to hint at a specific profile to be used at any moment (rough rollercoaster, smooth, super-overdraw, etc.).

Of course this de-phases the "shutter-moment" (the moment the frame is rendered) from the display moment, and may be irritating, so it better not be much time.
On the other hand it could give rise to a hardware-form of motion-blur, if the l-sync buffers would be passed to a RAMDAC capable of merging excess-frames (but in the presence of micro-stuffer) into a single coherent motion-blurred "moment".

 

[DeF]

The articles says that AMD identified/fixed it long ago, but somehow Sapphire didn't get the memo (made the changes), unlike the rest. And it has been looking like a sapphire 7870 problem all along (mine, from more or less first batch, doesn't have any problems though). Of course you will always see "ghost reports" in such forum threads..

Well i bought my Sapphire 7870 about month ago and had exact issues described in the article, plus some additional problems with clocks being not stable on load (droping to 2d) and weird voltage values readouts in gpu-z (negative voltage values). I decided to return my card and get 7870 from gigabyte instead. Now everything is running perfectly fine.

 

[iMacmatician]

From HT4U: "AMD brings Tahiti LE in November - new Radeon HD 7800-model" (original).

AMD apparently intends to re-start a leftovers. Tahiti GPUs, which are neither as Radeon HD 7970 (Tahiti XT) , or HD 7950 (Tahiti Pro) can be used to get the stamp of Tahiti LE and will be released as a new graphics card product in the market. The kick-off is according to our internal information in mid November, so in less than two weeks.

As our sources report, AMD is thinking but not a naming à la Radeon HD 7930th Rather, this Tahiti-LE GPU appear as Radeon HD 7800 model and from the performance point be faster than a Radeon HD 7870 OC. It would also be a Radeon HD 7870 Edition Boost, but this represents only speculation on our part.

3DCenter speculates whether this 7890 would have 2 GB GDDR5 on a 256-bit bus or 1.5 GB GDDR5 on a 384-bit bus, and if it will have fewer SPs than the 7950 but higher clocks.

EDIT: I'm not sure if this is speculation or if they have a source, but Arab PC World mentions specs of the 7890:

• 1536 SPs @ 1000 MHz ("Expected to come by 7890 bear card Logo GHz," and they don't mention boost), 96 TMUs, 32 ROPs,
• 256-bit bus, 1200 MHz GDDR5, possibly 2 or 3 GB,
• 6 + 6 pin power,
• 95% the performance of the 7950 and 105% the performance of the 7870.

 

[lanek]

From HT4U: "AMD brings Tahiti LE in November - new Radeon HD 7800-model" (original).



3DCenter speculates whether this 7890 would have 2 GB GDDR5 on a 256-bit bus or 1.5 GB GDDR5 on a 384-bit bus, and if it will have fewer SPs than the 7950 but higher clocks.

Could be a good midlife kicker for clean tahiti cores left, specially looking the HD8000 midrange cards will not appear before 5-6 months.

 

[HMBR]

I fail to see much room left for a card between the 7870 and the 7950

I think the gap from the 7770 to the 7850 needs some attention.

 

[Alexko]

I fail to see much room left for a card between the 7870 and the 7950

I think the gap from the 7770 to the 7850 needs some attention.

If they're discontinuing the regular 7950 to leave only the Boost edition, it might make sense. Besides, I think this is more about getting rid of partially defective chips than about filling an empty niche.