ATI Graphics horror case story: Switch to NVIDA radically lowers CPU utilization

[septimusminimus]

(posting intent: curious if others with ATI accelerators, esp. 4xxx series,  are seeing brutally and unreasonably pegged CPUs, terrible load times and frame rates too) 

I've had an HD4850 accelerator since back when they were fashionable, always used to run Second Life and nothing else. I've used it under Linux, Windows XP/32 and 7/64, using myriad third party clients. Always latest versions of clients plus accelerator drivers, naturally. Regardless, it rarely ran better than the 8600GT I pulled it for; sometimes, it ran much worse. For an extreme case, there was a stretch where I was seeing 10FPS on ultra low settings at 1280x1024 under XP. With modern settings on a 3.X client, the card pegged both cores of my E6300@ 3.95GHZ Wolfdale, leaving my avatar sldiing through a slideshow world. Horrible. Plain horrible. Long-standing, too. Deferred rendering plain slaughtered the system. I was looking for a deal on a new i5 2500K/1155 just so I could sucessfully partciipate here in SL. Silly me!

Now I have the new GTX460 768MB in ($80, free shipping! Night vs. day. The CPU never hits full utilization at the stock 2.8GHZ. Yes. Same Dolphin 3.x client, same other components, same install too.  It's doing 20-35FPS where the old would do 2-15FPS depending upon astral alignment.

Often, the old would perk up if I minimized the window, disabled and re-enabled vertex buffering, killed vertex buffering if under Linux; it would randomly lurch to a standstill too (e.g, 3FPS and declining,) only to pick up a semblance of performance at a later time.)

I expect great criticism here. I'd be reticent to stir the waters, save that my HD4850 has brought me such great frustration for so long. Obviously, you will all bring up arguments surrounding causation vs. correlation, anecdotes to contrary and attacks on the messenger's hardware and system configuration too. I don't care to start a flame war. Not my intent. I just want others to be careful before buying into the used FPS -> dollar bargain H48xx-series 3D accelerators, as well as perhaps considering older notebooks with disrete ATI GPUs (ex 3650, 4350 -> curious to hear other reports).

[septimusminimus]

Move to the main tech forum. Sorry!

[Jenni Darkwatch]

I've had ATI cards and nVidia cards. More recently (i.e. in the last 5-7 years) I stuck with nVidia. For pure tech spec comparisons I find this page rather useful btw: http://www.gpureview.com/show_cards.php?card1=633&card2=566 (link goes to the comparison between your old and new card). Performance comparisons of course have little to do with tech specs necessarily, though I like that the page lists noise and power consumption.

[VRprofessor]

"Obviously, you will all bring up arguments surrounding causation vs. correlation, anecdotes to contrary and attacks on the messenger's hardware and system configuration too. I don't care to start a flame war. Not my intent."

Well, I suppose I could bring up causation v correlation, that is in my job description, but from an information gathering point of view I would be delighted with reasonably good descriptions of peoples' experiences in this case. 

First, I don't think anyone is going to give me the $80k or so it would take to set up a lab and hire a bunch of undergrads for the summer to run  systematic benchmarks on dozen or so computers using a couple of dozen different video cards.  Even using only a couple of PC platforms the cost of video cards and undergrads would be prohibitive. 

That leaves us with anecdotal reports and tech savy folks to critique those reports.  Anecdote #1 should bring forth lots of criticism, especially in the form of alternative causality.  Additional anecdotes could simply be agreeable for reasons of confirmation bias, or they could be all comers telling their story.  In the first case we get bad conclusions, but in the second we can start to eliminate some of the possible alternative causes suggested.  With enough independent anecdotes we can narrow the conclusions fairly nicely. 

As much fun as the benchmarking would be, I wouldn't bother to do it for reasons other than fun.  Mostly because I have read enough of these threads now to conclude that those folks with technical backgrounds have largely vetted the anecdotes and come to the conclusion that nvidia based cards work better for SL than amd/ati cards.  Given that there are still reports of the opposite experience it suggests that there are still some details that are not fully explained, but I think those buying a GPU to support SL for the first time are relatively safe looking at nvidia based cards.

[Perrie Juran]

---

VRprofessor wrote:

"Obviously, you will all bring up arguments surrounding causation vs. correlation, anecdotes to contrary and attacks on the messenger's hardware and system configuration too. I don't care to start a flame war. Not my intent.

"

Well, I suppose I could bring up causation v correlation, that is in my job description, but from an information gathering point of view I would be delighted with reasonably good descriptions of peoples' experiences in this case. 

First, I don't think anyone is going to give me the $80k or so it would take to set up a lab and hire a bunch of undergrads for the summer to run  systematic benchmarks on dozen or so computers using a couple of dozen different video cards.  Even using only a couple of PC platforms the cost of video cards and undergrads would be prohibitive. 

That leaves us with anecdotal reports and tech savy folks to critique those reports.  Anecdote #1 should bring forth lots of criticism, especially in the form of alternative causality.  Additional anecdotes could simply be agreeable for reasons of confirmation bias, or they could be all comers telling their story.  In the first case we get bad conclusions, but in the second we can start to eliminate some of the possible alternative causes suggested.  With enough independent anecdotes we can narrow the conclusions fairly nicely. 

As much fun as the benchmarking would be, I wouldn't bother to do it for reasons other than fun.  Mostly because I have read enough of these threads now to conclude that those folks with technical backgrounds have largely vetted the anecdotes and come to the conclusion that nvidia based cards work better for SL than amd/ati cards.  Given that there are still reports of the opposite experience it suggests that there are still some details that are not fully explained, but I think those buying a GPU to support SL for the first time are relatively safe looking at nvidia based cards.

---

If there is one thing I have learned reading through the Graphic Cards threads it is that sometimes SL can be really finicky.  SL ran fine and dandy on my computer until V3 code was introduced.  Despite several discussions here in this Forum and with a Linden Dev, no one has been able to decipher why V3 gives my computer heartburn while other folks with very similar configurations have no trouble at all.  The Linden Dev even went as far as getting hold of a copy of my Graphics Card and can't duplicate my problem at the Lab!

[Darien Caldwell]

Not sure why this is a Horror Story, or why you expect controversy.

 

A graphics card is a GPU, it's a co-processor of sorts. It takes load of the main CPU. Programs can use this to their advantage to speed up operations. On the flip side, when a GPU doesn't support some functions, they have to be emulated in software by the CPU.

So all you have is a case where you went from a poorly supported GPU that had many functions emulated by the CPU, to a better GPU that could actually take on and accelerate more functions.

A good thing, and completely expected.   What a huge non-controversy.

[Perrie Juran]

---

Darien Caldwell wrote:

Not sure why this is a Horror Story, or why you expect controversy.

 

A graphics card is a GPU, it's a co-processor of sorts. It takes load of the main CPU. Programs can use this to their advantage to speed up operations. On the flip side, when a GPU doesn't support some functions, they have to be emulated in software by the CPU.

So all you have is a case where you went from a poorly supported GPU that had many functions emulated by the CPU, to a better GPU that could actually take on and accelerate more functions.

A good thing, and completely expected.   What a huge non-controversy.

---

Was your reply to me or was it intended for the OP?

Just double checking because I do believe you were responding to the OP.

[Dilbert Dilweg]

I have always been an Nvidia user. My first experience with ATI is that 2 of them went bad on me in one year... Total crap was my first impression of ATI after that, After I got an Nvidia card, things were great. I loved the stability and robust nature of it. I was sold on Nvidia . My last system I built had nvidia motherboard combined with nvidia vid cards. It loved the matchup.

I recently rebuilt my system and have a new FX series 6 core CPU. Which is designed to work well with the AMD Redeon (formerly ATI),, So I am at a real dilemma on which type of vid card to get.. I love the hell out of Nvidia. But I also enjoy AMD products and they now own the radeon. It is a big investment because I am spending probably 600 bucks or so on one. So now I am wondering how well the new AMD radeon runs compared to the ATI version. It is a tough decision and I can't test drive LOL

[Jenni Darkwatch]

I've got an AMD chipset with an nVidia GPU - works fine for me.

[Nyll Bergbahn]

I have an ATI Radeon HD 5850 paired with an Intel i5 750 quad core overclocked 4.1 GHz CPU and get wonderful performance, particulary with Niran's viewer (somewhat odd as he favours and works with Nvidia cards). Sure, there are the odd niggles but Nvidia owners get niggles too, especially with drivers. I can happily roam SL in Niran's equivalent of Ultra with Deferred Rendering on and seldom see my FPS drop below 25. It's usually up at 35+. Add 25/30 fps without Deferred Rendering.

I do appreciate that the general view is Nvidia is better with SL due to better OpenGL support but it is possible to get excellent performance from ATI/AMD as well. MY 5850 is about equal to your Nvidia 460 in general performance terms.

Enabling VBO with LLs mesh viewers however decimates fps down to 5 or so (was ok before mesh viewers). That may have been the problem for you with the 3.x viewers. VBO is fine for me with TPVs including Niran's however.

 

[Dilbert Dilweg]

I understand that and fully aware nvidia works well with it.. I have Nvidia running with it now. But thinking about upgrading video cards.. So I am at a place of making a decision of course which way to go this time around lol

[Brayla Ravenhurst]

I generally don't have problems with my ATI video cards.  The biggest problem with ATI is that sometimes they release really crappy drivers, so upgrade drivers with caution.  Also, avoid super cheap video cards (e.g. sapphire, etc) because their coolers are crappy - unless you are nerdy enough to buy the video card and replace the cooler right away.

 I have the 6xxx series - the 4xxx are rather old.  

[Cincia Singh]

I have the same AMD 6-core CPU on a Gigabyte mobo with 8GB RAM @ 1600 ... I have an EVGA GTX560 Ti (make sure you get the 256 memory bandwidth not the 128) and it works great in SL. I'd like to upgrade to a 660 soon. My grudge against ATI/AMD cards is they frequently have issues with OpenGL which is what drives SL.