New Info on Enduro Problem

I asked Chris Angelini at Tom's Hardware if he could contact AMD about the root cause of the Enduro utilization issue. Here is the response:

"As promised, an answer straight from AMD

'In response to your reader's question, the issue at hand is unrelated to a copy engine. It is actually because of some coding related to how the CPU "feeds" the GPU, and needs some additional fine tuning. He is, however, correct in that we do employ a copy engine to make discrete and integrated graphics work together in a switchable graphics configuration. The discrete GPU passes the rendered frame to the iGPU while it works on the next frame, and the iGPU passes the frame on to the display.'

All the best,
Chris"

This confirmation connects to the hypothesis I came up with last month, found here: http://forum.notebookreview.com/sag...-logs-no-questions-please-27.html#post8858223

It feels great to hear some info from AMD!

 

Great info. Fine tuning of coding sounds good in terms of getting a functional hot fix. I just hope that this hot fix comes soon...

 

Thanks for the info! The more they let everyone know the closer they are to a fix (hopefully)

 

Its actually great news, finally the hardware bottleneck theory falls apart. I read your old post (at the time you posted it), its actually a very detailed explanation of about the same, great insight btw.
Being noone to say anything about this i have always supported the notion of a software issue (therefore driver fixable), it just makes a lot more sense taking all facts and data into account.

 

and many people blamed 3610qm slow that time....

 

I thought the new driver was suppose to be out already... =\

 

They say in the next couple of days. I'm hoping it will show up in the next day or two while my laptop is still at Mythlogic being built so it gets loaded with the new driver straight away. We shall see!

 

Tranta, have you got a link of this? its a bit odd as AMD have publicly stated to Jarred from Anandtech that the root of the issue was transferring of data from dGPU to iGPU over PCIe. Neither does it fit with any testing we've done, where reducing CPU speed from 3.8Ghz to 2.8Ghz or vice versa yielded very little difference in utilization. also there are benchmarks that do very little CPU work (like kombustor or 3dmark06) but show bad utilization, and on the opposite side you have games that use CPU much more (metro 2033, max payne 3, the witcher 2) yet their utilization is pretty good...
either way, if this is not a bandwidth issue then AMD really have no excuse for not being able to fix this problem completely. Hopefully that fix comes out this week so we can finally see what results we get from 3 months of their hard work.

 

Here is a screen shot of the newest message that I quoted in the first post:



I am not sure where Jarred ever said AMD told him that. We knew that the problem was related to the PCIe bus and a balance issue between data going in and data going out, according to what AMD originally told Chris.
BTW, I never shared Chris's first message which stated that the issue was known to be PCIe bandwidth related, and that AMD could fix the issue with software changes: "We now from AMD that the issue is, in fact, one of limited bandwidth across the bus. AMD claims this issue will be solved via software. So, if it's a hardware problem, it's possible that support for the hardware isn't in place yet via the driver.

I'll ask AMD, however, whether they have a similar feature as Nvidia in place, and if it not being properly "turned on" is what's causing the performance hit.

All the best!
Chris"

Chris sent me that on the 4th. He asked AMD about it for me, and then yesterday he sent me AMD's new response.

So, the problem is bandwidth related (or how the data flowing through it is being managed, more precisely), and the driver code fix will make the CPU to GPU transfer more effective/efficient with this bandwidth. The copy engine is confirmed, so even if the iGPU were apart of the problem (which it likely is in some way), AMD still would be able to fix the problem. The 680m performs great and runs through the same PCIe with a copy engine, so if you were to say the 7970m is hardware limited, then so is the 680m. They are now both at the same level with the confirmation of the AMD copy engine, so if there is a limit, it will be the same, very minimal limit as with the 680m.

The first hotfix won't completely solve the issue, as it is apparent perfecting this data balance is very tough. It isn't a simple "do this," but more like trial and error and going through lots of code and making it more efficient. If we see progress with the hotfix, then bam, progress. Mainly, realize that you and I aren't apart of the driver team, and that there is a LOT we still don't understand.

 

Let it go arcticjoe, sometimes one gets so commited with a theory that cant see past your own nose.... The facts and recent AMD comments, besides logic and common sense, show that there is no hardware bottleneck. At least non different than the one Nvidia 680M has to get through.

It really sounds like you dont want a fix to come, as someone said in a past thread.

And yes, AMD did say the hotfix will be public this week, we still have two days to go... if not it aint so bad as well, better they tune it up as best as possible, dont wanna be patching 20 times.

 

but i have let it go in the post you've quoted...? if this is the reason why my card under-performs then so be it, i dont mind what it is as long as AMD can fix it. my guess about bandwidth issue was made based on what facts we had known at the time, if I got it wrong then nothing really changes... there are no prizes to be won here, all i care about is having a fully functioning hardware on my laptop.
and once again, i had never stated it was a hardware bottleneck - in fact i have specifically said that it could be a software one, yet in every thread you seem to jump in with "GOOD NEWS GUYS!!!1 .. its not a hardware bottleneck". nobody cares what type of bottleneck it is, any type of constraint that prevents a system from running at full speed will be regarded a bottleneck.

and just because 50% of my posts do not contain words "GOOD NEWS" it doesnt mean that I dont want the fix to come.. I just wish people kept their heads level and not act like AMD are doing us this amazing act of kindness by releasing a working driver, 7 months after their card was released. I will admit I am a little bitter about the whole thing, having sold my 6970m based p170hm for half the price i paid for it to buy a machine that seems to be inferior in performance, plus not being able to use it properly for half a year has left me somewhat disillusioned with AMD and their support.

anyway, i feel like a broken record, but ... lets wait for the fix and see what performance is like, k? thx

 

And it is still AN email from a secondary source (however credible it might be). At the end of the day, AMD will get no points from me on how they handled the situation. It caused a lot of headaches not knowing what the problem is or if its even fixable. Its about time we get some kind of a fix.

 

Completely agree! It completely baffles me that anyone can be defending AMD at this point.

Fix or no fix the damage for me has been done. I will go out of my way to never purchase another ATI/AMD product.

 

The only certain case when I would never buy another AMD product is if they go bankrupt, which would be a sad day for all PC gamers out there.

I quess you are into casual gaming, since after Nvidia has failed you too, you won't have many options left...

 

I am not defending AMD, but I am suggesting that at this point, some of you guys are biting the hand that feeds a bit too hard. We are still putting the necessary pressure on Anandtech/Tom's Hardware and AMD, and that will and has kept AMD in check.

Why this is good news? Because it IS good news. Knowing that the issue is for sure not a hardware bottleneck is a great relief. Like I said, the copy engine is confirmed to exist. The 7970m has the same limits as the 680m, and all it is missing is good driver base code. The driver fix will be released soon. Any other talk about AMD flucking things up is similar to complaining about Germany causing the Holocaust. Time moves on. That is the PAST, AMD is making new drivers for us. There is no better scenario at the moment. Would you rather AMD NOT be making new drivers? Then it would be AMD with the mistake in early 2012, and then continuous denial! Stop complaining about progress. Get over the past, and just be happy that the present and future are looking good for us 7000m owners.

 

The only real problem I've had with this whole scenario is that AMD didn't communicate with their customers. When AMD releases the patch it still doesn't change the fact that they handled the issue poorly (with only short term profits in mind) and against all business logic. Not knowing was the worst and i could have waited for the fix a lot longer, had I known one was coming.

That said, I am very happy that drivers are coming, but it is no wonder people are still upset.

 

ok, back to reality dude : the fix hasnt been released yet, you havent seen it, you dont know how well it will work, you dont know how long before AMD make further improvements or if there will be any.
nobody here is complaining about progress, people are complaining about lack of it instead. even if the new driver drops today, its still many months overdue.

 

I never said the fix is out. Look at Jarred's numbers; if you are telling me the results of his testing do not represent progress, we might as well just stop arguing now. No, we do not know the overall performance behind the upcoming hotfix, but Jarred's testing shows that it is more than likely to be a large step up. And guess what, if AMD DOESN'T continue make more improvements, then we will keep pressuring them now won't we? Isn't that how we got AMD to respond in the first place? I never said that we should stop putting pressure.

And do not make me repeat this one more time. The driver took so long because THINGS LIKE THIS TAKE TIME. You don't have the experience or knowledge to tell AMD that they can do it faster. I have already said this a million times, AMD made a mistake by not testing thoroughly in early 2012, and when they discovered the problem, that is when they wen't into motion. I WAS angry that AMD didn't have this solved before they released the 7000m series, but what is the point of continuing to be angry after getting a positive response? You have the choice to be happy. If you want to choose to be angry at the past, then fine, be angry, but that is your loss.

And Montage, AMD is communicating with its customers. I sent a support ticket last month and they responded telling me that the 12.10 driver would improve utilization fore the 7000m series.

 

That's hardly what I meant with "communicating with the customers"... That was last month when we already knew a patch was coming, and they already had the thing well under construction. I mean how AMD initially responded to the issue by trying to hush things down and keeping totally silent. They didn't officially say a word for months, which isn't exactly cool IMHO.

 

I'll give you that, but you have to admit AMD has opened up quite a bit recently. My point is that the situation has flipped flopped from disastrous to positive, and the positive change has resulted in me being positive as well.

 

.
we both know Jarred's numbers are somewhat flawed - lower resolution can be discounted as i cant imagine anyone ever playing at less than 1080p with their laptop and @ 1080p avg increase in performance was only 7%...which is only 3fps at 40fps; he only measured max fps when the root of enduro issue is the fps dips; in Batman AC max fps actually dropped 21%; ... now i am not dismissing the fact that there is progress being made, but seriously cant we wait to see the end result before we all hail AMD for their amazing work?

look around, people are already recommending 7970m to other posters here despite it currently being a faulty product.

i realise that everything takes time, but some companies are a lot slower than others when it comes to driver releases and AMD are definitely one of the slower ones. couple that with the fact that they largely ignored the issue for the first 4 months, despite numerous attempts to bring it to their attention and you cant say that they have anyone else but themselves to blame for the flak that the're currently taking... so no, i dont have much sympathy for them.

must be awesome to be so easily pleased - i for one like to see the results before being happy about the progress, promises and rumours are not enough.

 

Why do people keep saying he used max values? Some one said that bullcrap at one point out of the blue and now everyone is automatically believing one's uninformed opinion. Jarred never said that, and it is obvious he used average looking at the frame rates. BF3 does not max out in the 40s, yet you are telling me Jarred's 41.1 FPS value at 1080p is a maximum value?

Additionally, the 7% increase includes the decrease from Batman. Why that happened, I don't know, but Skyrim's performance at 1080p increased by 12%, and Civ 5 by 64%, and both games needed the FPS boost (under 60FPS, any game could use more frames). An FPS boost like that for any game shows that what AMD is doing has great potential. Again, I know this driver doesn't show the whole picture, but it makes me excited because it shows the potential of the direction AMD is taking. No, it isn't released and I WILL wait for it to come out to judge it! I keep repeating myself over and over again in different ways and you can't seem to understand. There is no reason to be negative if we are already doing the best we can.

I am pleased, yes. You could say any one that is pleased is easily pleased, so your argument has no foundation. I could say that when you are pleased by a movie, you are easily pleased. Jarred's recordings were in average FPS and showed some large jumps besides Batman. Yeah, that is pleasing. Once again, choose not to be pleased and be pessimistic. If the hotfix ends up sucking, then yeah, that will be it, and from there we continue to keep up with AMD. No negativity is required, yet you reek of it. These are not rumors, and promises made by AMD should be taken seriously. If you can't trust a company that openly tells its customers through support that the next driver will improve utilization (which was already proven through Jarred anyways), then there is no one you can trust. AMD is a business, and they make money by maintaing a customer base. Lying repeatedly will destroy their reputation.

Now, whether the issue will be COMPLETELY solved is another thing. We don't know how far AMD will go, and that is why if and when we run into issues with the hotfix, it will be our duty to report as much as we can to AMD and make sure they know that we won't be happy with a partial fix. I am positive about responding to that likely situation as well. Again, no negativity required. We will be happy for progress, and happy to report and put more pressure in response to bugs and improper performance.

 

today is friday. when it be out?

 

AMD has not actually officially said that the patch will be out this week. Official statement is this month.

 

They are probably running into new problems as they fix others. It is better that the hotfix is stable, so I am fine with waiting a bit more.

 

I believe that it'll be released sometime around windows 8 launch day (October 26th)

 

wow, didn't realize windows 8 was that close.

anyway, here's to hoping amd steps up and gets the issues fixed so this can become a distant memory

 

Speaking of distant memory
Maximum PC | AMD Shares Hit Three-Year Low After Issuing Q3 Sales Warning

 

ouch, rough. lol

 

Tablets are good for watching videos, but are people really using them for normal applications like word processing? Sure, demand for tablets can rise, but tablets can't do EVERYTHING, and people need normal computers do to tasks more efficiently. Maybe computers these days are just becoming so powerful while applications aren't really scaling with the growth of the PC market. You can still use a Core 2 Quad/DX10 GPU/4GB RAM computer setup from 2008 and get by with great response and efficiency, even with video editing and gaming.

 

as much as AMD's handling of this issue has pissed me off & I dont ever see myself buying any of their products again, i sincerely hope they manage to pull through as you need healthy competition in the market to drive innovation and lower prices. Maybe they should stop focusing on their crappy CPU line and work more with their GPUs where their products can still compete. Also, they really should have bought out ARM when they've had the money - mobile market is where money is being made right now and they'd be raking it in.

 

What is sad is that the FX-8150 is barely even competitive value wise. The AMD FX processor compares to an i5 2500k in gaming tests (some times performing worse), and the 2500k is only $30 more on newegg. Hell, owners of the better Phenom IIs still don't have any thing worth upgrading to AMD wise. If you have a Phenom II X4/X6 above 3.0GHz, you are still gaming hard. Intel's i5 from last generation matches AMD's CURRENT best, and both are near each other in price. The only advantage with AMD is the better onboard GPU, but no serious gamer buys the BEST available AMD processor for facebook. They are going to buy a dedicated GPU as a gamer. For the higher end of the gaming market, the onboard GPU only needs to be good enough to drive 1080p video.

In the end, AMD failed in a couple ways. For one, the company decided to go with an architecture that couldn't compete with Intel's best offerings. Second, the company decided to include a more powerful onboard GPU onboard the FX series chips, wasting power that could have been allocated to the CPU. People own dedicated GPUs, and most of those GPUs will be better than any current integrated solution.

Right now, it is apparent AMD is one step behind every one else. They are only now catching on to fixing their switchable graphics, some thing that was done by Nvidia two years ago. AMD is a generation behind in CPU power, and even with their current offerings, the value isn't much greater than that of Intel when looking at gaming performance. It is, however, important to remember that game developers like optimizing for Intel+Nvidia for whatever reason. This industry bias will cause problems for AMD, but AMD still could have fought back harder.

 

Oh yes, this is a godsend. Never wanted to regret going 7970M, and this has strengthened my faith in AMD. Hopefully after I install these drivers after they come out, I won't get BSODs...

 

Are you sure those BSODs are related to the 7970M? I've been using the stock drivers and haven't had a single BSOD or a crash yet.

 

yeah, whilst mine crashes in Max Payne 3, it never BSDOs. I am running the latest driver, but neither drivers have gave me any trouble with system crashes.

 

I've never blue-screened yet, using stock drivers at least.

 

That's one thing AMD does right, their drivers overall very stable, probably among the most stable in any of the computer industry, vs how many they release per year, supporting the # software/games. It's amazing. Which goes for Nvidia as well, both are great for driver support, no other companies match them in that regard.

When I had BSODs on this machine, it was the SSD, but since RMA, that has been rock solid also.

 

The whole idea is that the onboard GPU can be used as a specialist FP core and accelerate a wide range of stuff while being able to drive a decent amount of displays.

The problem was they always needed a way of porting current code at least to help some, then maybe come up with tools to natively code to take advantage of it.

 

I've had one BSOD related to the AMD Driver, according to WhoCrashed at least, but that was only one.

and it was a few months ago. damn, time passes by fast

 

Nothing to see here...

 

I've always wondered what AMD's 'hotfix' branding is supposed to indicate compared to their other driver updates. Anyone know?

Looking at Wikipedia, I can't find a definition that fits what AMD is doing.

 

Hotfix drivers focus on one issue.

 

Now that I think about it, this card has been a whole lot more stable than the 650m in my old laptop. That thing would crash several times a day. There were several games I couldn't get running at all. I haven't had a single issue aside from underutilization in a few games, and it's only caused poor fps in 2.

 

That's what the term usually implies, but looking at prior Catalyst releases it also seems to be used where they mean "beta" and/or "out of schedule release".

 

Hotfix just means it's a driver that is released quickly for a specific reason. Beta drivers are usually something AMD wants to release in the past, excited about. To show off something and usually they were leaked.

For example, if they realize a fix was available for W7/Vista but was missing on XP, they will do a hotfix for XP with the fix. For Enduro, yeah, it would be a hotfix, a fix until a WHQL is available with it. But it's not a beta driver, meaning Hotfix drivers are usually drivers without WHQL certification.

 

Betas don't have WHQL stamps. The only difference between betas and official drivers are often the WHQL stamp (at least that is the case with Nvidia's drivers). As far as I know.