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

Don't be naive. The intentions are clear, why else the AMD optimized setting at all, and at default even? Again: The control slider itself is not the issue, the default setting is. And as I said, not every cheat has to be secret. Sure, some sites would call them on it, but my guess is, most won't and use the driver defaults as they have always done. Even at the time when AMDs AF was shimmery and inferior.

The default setting has never had a profile, so it's always set to provide what the application asks for.

It's funny you should bring up shimmering, as this was caused by AMD providing what the application asks for, where Nvidia changed that in order to provide what Nvidia wanted to provide, not what the developer wanted.

Again, AMD provided their customers with a choice instead of arbitrarily changing image quality as Nvidia does.

As for being "naive", do you really think that Nvidia hasn't been doing behind the scene "optimisations" for years? Just they never let the customer know, have any control over it, or disable it as AMD does.
 
The default setting has never had a profile, so it's always set to provide what the application asks for.

It's funny you should bring up shimmering, as this was caused by AMD providing what the application asks for, where Nvidia changed that in order to provide what Nvidia wanted to provide, not what the developer wanted.

Again, AMD provided their customers with a choice instead of arbitrarily changing image quality as Nvidia does.

As for being "naive", do you really think that Nvidia hasn't been doing behind the scene "optimisations" for years? Just they never let the customer know, have any control over it, or disable it as AMD does.

I thought the negative LOD bios setting in the Nvidia control panel would allow you to have it the same way AMD does, with some shimmering depending on if the dev has requested negative LOD bias or not?
 
I thought the negative LOD bios setting in the Nvidia control panel would allow you to have it the same way AMD does, with some shimmering depending on if the dev has requested negative LOD bias or not?

No, the LOD clamp is the other way, forcing lower texture res than asked for, and therefore less shimmer, more blurriness (and more performance).
 
The default setting has never had a profile, so it's always set to provide what the application asks for.

It's funny you should bring up shimmering, as this was caused by AMD providing what the application asks for, where Nvidia changed that in order to provide what Nvidia wanted to provide, not what the developer wanted.

Again, AMD provided their customers with a choice instead of arbitrarily changing image quality as Nvidia does.

As for being "naive", do you really think that Nvidia hasn't been doing behind the scene "optimisations" for years? Just they never let the customer know, have any control over it, or disable it as AMD does.

Source for the AF claim? As far as I know this was a hardware bug in the texture filtering units. This has been discussed to death.
Where does Nvidia change the image quality arbitrarily? HawX was a application bug. They don't do it with tessellation or AF. Also I find it unconstructive to divert from the topic at hand by saying "but Nvidia does optimizations too". Maybe, maybe not. But that is beside the point here. When it comes to optimizations, no matter what kind they are, my alarm siren always goes on. And it always is for the benefit of the IHV and for winning benchmarks, never for the customer. Some may call this view overly pessimistic, I call it realistic.

If you want to make a close with FP16 demotion.. you are in the wrong way, due to this tricks who was working only on some games ( basically old games outside Dirt1 ) ... Nvidia have start use FP16 demotion just on the 460 release.... (what is funny is the FP16 demotion is a technic described by Nvidia at first and you can find it in their whitepaper )

At this moment reviewer have start use Catalyst AI on HQ level who remove all AMD optimisation per games ( who is not completely justified as most of the optimisation are justified .. ).. You can watch all review they mention they put the catalyst AI on the max level...

At the same time, nvidia dont let you choose if you want use or not use thoses optimisations ( like FP16 demotion ), they are invisible and you cant disable them .. I just want to recall HawX episode, where the MSAA was magically transformed in 4xCSAA instead of 4xMSAA for Nvidia cards ( who ofc let a big difference in term of performance. ) ( one example in one hundreds ) .. (developpers have appologize for this error, saying it was a bug in the settings control ..

As for HawX:
http://www.kitguru.net/components/g...ick-stam-addresses-hawx-cheating-allegations/
Sounds reasonable. If you believe it or not, well that's your call.

I wasn't talking about FP16, but if you bring it up, back it up. A quick search shows that is is not Nvidia but AMD that does this. Most reviewers still bench with driver defaults, go look around in the web.
 
Last edited by a moderator:
As far as I know this was a hardware bug in the texture filtering units. This has been discussed to death.
Where?

Also I find it unconstructive to divert from the topic at hand by saying "but Nvidia does optimizations too".
I find unconstructive to discuss issues unrelated to HD 7xxx series in HD 7xxx-related thread. Could you be so kind and continue this discussion in a more appropriate thread, please?
 
I thought the negative LOD bios setting in the Nvidia control panel would allow you to have it the same way AMD does, with some shimmering depending on if the dev has requested negative LOD bias or not?

Actually no, what was happening was that Nvidia was choosing to apply either bilinear, brilinear, or trilinear rather than trilinear or bilinear if the program specifically asked for it in conjunction with anistropic filtering depending on Nvidia's algorithms.

AMD cards had the shimmering issue because they always provided the bilinear or trilinear filtering that the program asked for.

Hence shimmering on AMD cards, IIRC, was due to the program asking for bilinear filtering and hence AMD cards providing bilinear filtering in conjunction with anistropic filtering. While Nvidia cards would ignore the request for bilinear filtering and instead us brilinear or trilinear filtering.

AMD has since changed it so that they have an option to ignore program requested bilinear just like Nvidia has been doing by default. Hence the AF quality is similar between the companies now.

Regards,
SB
 
Where?


I find unconstructive to discuss issues unrelated to HD 7xxx series in HD 7xxx-related thread. Could you be so kind and continue this discussion in a more appropriate thread, please?

For example at 3Dcenter.
Fair enough.

@the post above:
No. To my knowledge that is completely untrue. It definitely was bug. Ask the experts around here (which I am not, but I've followed this discussion extensively).
 
Last edited by a moderator:
HD 6870 owner here and I appreciate being able to play Crysis 2 with tessellation on and playable frame-rates thanks to the adjustment available in the CCC.

Also, IIRC, aren't even the Jersey barriers tessellated to a degree whereby no discernible visual improvements are gained over a much less tessellated version?

If my memory of that is correct then, imo, that is a decisive fact.

AMD did the right thing in addressing such a sad situation.
 
The texture thing BZB is talking about is the default min / mag settings where our drivers have always even set to developer controlled (per spec) and historically NV's were fixed.
 
Or there's lower demand for AMD products.

I think they should go all-in with further significant discounts.

20q105v.jpg
 
I wasn't talking about FP16, but if you bring it up, back it up. A quick search shows that is is not Nvidia but AMD that does this. Most reviewers still bench with driver defaults, go look around in the web.

You turn that on with "Surface Format Optimization", there is no hidden or unturnoffable cheating going on in AMD-drivers.
In addition, surfaces are only optimized for which either the faster format has no precision-change (dropping alpha, or changing to int fe.) or if the loss of precision makes [for human perception] zero difference in the final 8bit backbuffer. You may have a bit-flip here and there, but's not percievable at all in the normal noise of the image.
Actually the surface-format optimization is so conservative, I consider it the "friendly correction of the game's render-pipeline - because of insufficient profiling from the side of the developer - to the common benefit of gamers", and not to show off in benchmarks.

Sorry, OT: the 7xxx series probably does less of those as some higher precision formats are now as fast as lower precision formats from before. :^P
 
No, the LOD clamp is the other way, forcing lower texture res than asked for, and therefore less shimmer, more blurriness (and more performance).
Are you sure? Because that's not what the tooltip says. Do you have some linkys/citations or such to back this statement up, because this is the first time ever I hear this claim...
 
Negative lod bias (in general) means using a lower (ie higher res) than default mipmap level (IOW using a scale of more than ~1 texel per pixel), for a sharper image (maybe too sharp leading to aliasing), and therefore requiring more bandwidth (=> less performance). I doubt it's very common though, but there are places you would want that.

The "negative lod bias - allow" should be the default setting (and seems it is now?), ie doing what the application asks for, while the clamp setting supposely means that the lod bias will be clamped to be positive. I really don't see how the 'allow' setting would give more performance as claimed by the tooltip.
 
You turn that on with "Surface Format Optimization", there is no hidden or unturnoffable cheating going on in AMD-drivers.

Very minor correction: It is turned on by default and you can turn it off.

The driver option doesn't disable tessellation, it just reduces it to reasonable levels.

Actually, when you uncheck the box "AMD optimized", you can select a tessellation level below 2x - in the german version this setting is labelled as "Aus" (off). I don't know if this skips tessellation altogether or if it only reduces geometry expansion to 0 - i.e. I am not sure, if hull and domain shaders have to run on the original geometry anyway or if they can be skipped as well.
 
Actually, when you uncheck the box "AMD optimized", you can select a tessellation level below 2x - in the german version this setting is labelled as "Aus" (off). I don't know if this skips tessellation altogether or if it only reduces geometry expansion to 0 - i.e. I am not sure, if hull and domain shaders have to run on the original geometry anyway or if they can be skipped as well.
I doubt they can be skipped by default as you can modify quite a lot in these shader stages. Skipping them makes it very likely to break the data flow for the rendering.
 
Wasn't it the 125 MHz clock difference what largely separated 7970 from 7950 - given AMDs still sub-par scaling with # of functional units? From that perspective, I simply cannot believe that they would basically kill off their 7970 and 7970 GHz-Edition with a 7950 GHz-Edition.

As mentioned the cooler screen with GHZ logo was from the 7870 GHZ.. The only reason i could see is if they want to update the 7950 cores with same PT Boost.. ( finally 7950 are many times core who will not forcibly goes on 7970 lineup, so if it is enabled on 7970GHZ, i dont see any reason it will not on future 7950. Or maybe they want use the allready made 7970 cores on it.. )
 
Are you kidding me? :)
If you do a contest of any kind, do it right. For example if you run, you run - you don't shorten the length of the track just to be "faster". If you want to run 100m, run 100, not 90 or 80 or 50. In my mind this is absolutely stupid and defeats the purpose of benchmarking completely. Fastest score yes. But be honest about your methods.

Since yesterday was a holiday over here and I clearly had too much time on my hands, and in the spirit of the upcoming Olympic games (but with tongue firmly in cheek), I present :

"The Decathlon. A true(?) story"

----

Once upon a time there was a small country "A" who decided they wanted to host a fantastic sporting spectacle, as it would contribute to their country's economy and world recognition. On consideration they decided that they would host a decathlon as it would allow the spectators to see the widest variety of events. Unfortunately as they worked on their preparations they discovered that the costs involved were spiraling and becoming a real strain. If they could just find a strong partner then it would really help.

At this point they were approached by a large country "B" who said that in return for some minor considerations they would agree to share some of the costs of promoting the games. This seemed like a great idea to the hosts who felt that with this assistance they would now be able to make a really great event. In addition, since country "A" was so inexperienced in setting up a large event like this, "B" offered to send some of their "Decathlon Organization Specialists" (or DOSsers for short) to help arrange the events and make sure that this was a truly "cutting edge" decathlon. This also seemed like a great deal to "A" at the time and they gave "B" lots of freedom in helping them to set up what would now, surely, be the best decathlon ever.

The time came for the games, but on arriving to compete country "C" took a look at some very recently revised event details and they were horrified. They came to the hosts and mentioned that they had always been rather concerned with "B"s close involvement in the event, since "B"s athletes were also competing in it, but now that they had arrived it appeared that things were worse than they had feared. Even in their worst imaginings they really hadn't been expecting to be presented with a decathlon where several of the traditional events were completely missing and had now been replaced by five separate rounds of the high jump.

It was known that country "B"s athletes were particularly good at jumping, it being ingrained in their DNA by the shouting of their leaders. On checking the rules more closely "C" had also discovered that not only was the high jump being repeated multiple times, but the minimum qualifying bar had been raised to a height which only country
"B"s athletes had ever been known to clear in the past. Anyone who couldn't clear this height would therefore get no points in those events at all. "C" considered their athletes to be very good at the high jump, but the combination of the two changes would make it highly unlikely that they could make up the probable points gap to "B"s athletes after the five high jump stages, regardless of how good they might be in all of the remaining events.

Country "B"'s officials stated that their research showed that the only reason anybody really came to watch a decathlon was to see the athletes jumping not just very high, but _really, really_ high, so by making these modifications they had simply been ensuring that this would be the best decathlon of all time.

After a long and uncomfortable pause "C"s officials then asked why the 1500 metres was among the events that had been removed - they clearly recalled that in the previous year "B" had been telling everyone that it was a tremendously important event and that their decathletes were the best at it. In such a "cutting edge" decathlon it seemed like a strange omission, and this year "C"s athletes had been training hard and were known to be _really_ strong in the 1500 metres.

"B"s officials looked a little bit uncomfortable, but stated that while they still thought the 1500 metres was very important under the right circumstances, on reflection they had decided that it was really best left to the specialist athletes and that it didn't have much of a place in a decathlon. On being pressed on the matter they admitted that it was possible that they might change their minds again later in the year when some new athletes were coming up through the ranks.

In the end, since this was an independent event with no third party oversight it was clear that the hosts and "B" were free to set up the events however they chose, and there was no-one truly independent to whom any appeal could be lodged. Country "C" stated that they really didn't think that this was the sort of sporting event that the Greeks had originally had in mind, but they continued to participate since it was clear that their supporters would be pretty unhappy if they didn't get to see their athletes perform having paid for their tickets.

Unsurprisingly the athletes from Country "B" ended up taking both the gold and silver medals in the games, and their spectators went home very happy having thoroughly enjoyed the event. Country "C"s spectators were confused and a lot less happy, and they were left wondering why there needed to be quite so much mindless jumping, and what had happened to all the other events in which they knew their athletes excelled.

The press enjoyed the event a great deal and thought the overall organization had been exceptional, although a few of them did make the comment that there had perhaps been "a few more high jumps than were strictly necessary."

Country "A" was very happy to have hosted a successful games, and any controversy over the unusual arrangements didn't seem to have affected their ticket sales in the slightest.

----

The following year the event was held again, and a strange thing happened. Suddenly country "C"s athletes seemed to be much more competitive in the high jump events. Country "B" was surprised and alarmed by the situation, and after holding a detailed investigation they discovered that "C"s athletes were apparently using a new kind of shoe with four times the spring of the traditional shoes. They complained that such a technological change
really didn't seem very sporting as their athletes were still competing using the original shoes.

On being questioned about this "C" stated that they really felt that last year their spectators had been deprived of the true spectacle of seeing their athletes perform due to all the remarkable last minute rule changes, and that they had worked very hard to come up with a solution for this year so that their supporters could get a lot more enjoyment out of an event which they paid good money to watch. As had been made very clear to "C"s officials the previous year this was an independent event with no oversight, so on investigation they had found that as a result there didn't actually appear to be any official written rules about what shoes the athletes had to wear when competing. "C"'s officials then pointed out that from the stands and on television it was impossible to tell which athletes were wearing which shoes, so now everyone could just watch the athletes from both nations performing five great high jump events.

"C"s supporters were definitely able to enjoy the event much more this time around. "B"s athletes continued to perform just as well as they had before, but in the end the results and the medals ceremony were somewhat different.

The press commented on all the peculiar changes and controversy - it was obvious that "C"s spectators were now enjoying the games much more, although it did appear that some of "B"s supporters were not so happy with the changes. "C"s officials mentioned that they were allowing their spectators to vote on whether or not their athletes should use the new shoes, but they didn't see much chance that any of them would really want to go back to using the old shoes when given the choice.

It was rumored that an official from "B" had been heard to mumble "Next time we sponsor a decathlon there will be at least nine high jump events, and if they are very lucky we'll let them keep the pole vault."

Meanwhile, strange things were also happening elsewhere. More and more independent decathlons were popping up with different sponsors, and each one seemed to have a different mix of events, scoring or both. "B"s athletic federation were definitely upset by some of them where the 1500 metres now featured prominently, grumbling that perhaps the organizers hadn't got the memo that the 1500 metres was "so last year" and that this year it should be high jumps all the way.

After a while it really became very difficult to determine whose athletes were the best at the decathlon, as it was now so dependent on just which decathlon you went to see. Unfortunately by this time no-one was even able to remember what the original set of events was that composed a traditional decathlon, or how they should be scored...


----

Anyway - enough off topic from me. I hope the moderators will choose not to be too hard on me, and I now return you to your regularly scheduled arguments... :)
 
Back
Top