Something else I have noticed on reviews that report minimum FPSs is that CF 4870s often have the same minimum FPSs (sometimes even lower minimums) as a single 4870. Isn't that going to lower the quality compared to a more powerful single GPU that has higher minimum FPSs? In other words, if both set-ups have higher maximum FPSs than a human eye will detect, who cares about the 20% to 40% higher maximum of the 4870X2 if it has minimum FPSs that cause more stutter than a single GTX280 experiences.
-
We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.
GT200b to launch at Nvision 08 (Aug 25-27)
Page 5 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Something else I have noticed on reviews that report minimum FPSs is that CF 4870s often have the same minimum FPSs (sometimes even lower minimums) as a single 4870. Isn't that going to lower the quality compared to a more powerful single GPU that has higher minimum FPSs? In other words, if both set-ups have higher maximum FPSs than a human eye will detect, who cares about the 20% to 40% higher maximum of the 4870X2 if it has minimum FPSs that cause more stutter than a single GTX280 experiences.
Absolutely. That would be a prime example of where you're paying more expecting better performance where you need it most, but only getting single card performance. The AVG FPS might be very different though and show an improvement, but only because you were getting higher frame rates in less intensive sequences where you didn't need them to begin with.
The Odorous One
Banned
It's not perceived by me. How you can possibly continue making statements without using multi-gpu is baffling 😕
Your psu has ripple too, can you perceive that?
taltamir
Lifer
Odorous... they accept that you do not perceive it yourself, it is clearly stated that not everyone perceives micro stutter. But you claim that just because you don't perceive it, nobody else does as well.
And I would have perceived my PSU ripple had I had electrical sensitivity (like a shark) and was looking right at it, seeing as humans do not have such a sense, then no, we can't. This has nothing to do with micro-stutter though.
Virge, you warned about getting too heated. Did you mean we should cease discussion of this topic, or did you mean that we should keep such a discussion civil.
And I would have perceived my PSU ripple had I had electrical sensitivity (like a shark) and was looking right at it, seeing as humans do not have such a sense, then no, we can't. This has nothing to do with micro-stutter though.
Virge, you warned about getting too heated. Did you mean we should cease discussion of this topic, or did you mean that we should keep such a discussion civil.
ArchAngel777
Diamond Member
Actually, the move to 55nm does not garentee any clock improvements at all. Even Intel, moving from 65nm to 45nm didn't see that huge of a jump in clocks. Most people were already hitting 3.6 on the E6600's, and now the E8400's can generally hit around 4Ghz... Forecasts were thinking they would hit 4.5+ no problem, but that isn't the case. Once t he 45nm matures, then will they yeild better results. But if moving from 65 to 45 was that small of a jump in clock, then my guess is 65nm to 55nm will be even smaller.
If I had to guess, I'd put the new GT200B at 700Mhz (maybe 750 for O/C) core, 1600ish for Shader and around 2400/2500 for memory. The chips should run a bit cooler clock for clock, but 55nm isn't going to revolutionize the GT200 at all.
Keysplayr
Elite Member
You could be right there. Die shrink of course does not guaranty higher clocks. 9 times out of 10 though, this has been the case. I think your numbers, 700-750 aren't far off from what we'll see.
I wonder about that NVIO chip though. Will they absorb it's functionality into the core, or will it remain discrete. G80 to G92 had some pretty large modifications. Less ROP's, more texture units, smaller bus. Who knows what NV has in store for GT200.
ArchAngel777
Diamond Member
A die shrink can only be good. So I am hoping they can really bring down the costs and increase the performance at the same time. But, I am a bit weary of purchasing a card right now, especially if what nVidia hints at holds true. It seems the minute I purchase an expensive graphics card, one is released shortly after that demolishes it. Or, it takes a huge price cut. I was hoping that GT200 was more than it ended up being, so in many ways I am very glad I decided not to buy it. Just look, it went from $650 to $400 in a month! Crazy times! I think I will be skipping this generation anyway... When I really sit back and honestly evalulate the performance of the 8800GT/GTS/GTX I can't really see a compelling reason to upgrade for 16x10 or 19x12 displays...
That may be true at 1680, but its definitely not the case at 1920. Many recent titles were already taxing my 8800 GTX to the point frames were constantly below 40FPS and teetering on the edge of being unplayable.
Here's a quick rundown of more recent titles:
Group 1
LOTRO DX10
CoH + OF DX10
World In Conflict
Sup Comm + FA
NWN2 + MotB
Mass Effect
Assassin's Creed
The Witcher
Crysis
Gears of War
Huge improvement in gameplay experience, not only frame rates but also enabling higher levels of IQ and detail. In some cases, I'm also able to enable AA but many of these demanding titles were borderline playable on a single 8800 GTX with some settings turned down.
Group 2
Bioshock
COD4
STALKER
Titan's Quest + IT
No major improvement with frame rates, upgrading just pushed FPS to 60 capped or to a more constant, higher frame rate. Upgrade has mainly allowed me to push up AA settings.
Group 3
FEAR
GRAW2
Sins of a Solar Empire
CS:S
Anything older than 2 years
Frames were already high enough it didn't really matter, but enabling higher levels of AA is now a possibility without any perceived performance hit.
On micro-stutter:
Micro-stutter exists in both single GPU and multi-GPU setups, and its generally worse in the latter. People differ in the level of micro-stutter they can discern. Stating it as a general disadvantage for multi-GPU is not really fair, else I could say that anything above 60FPS (or even 30FPS) is wasted as a lot of people couldn't discern it. People like BFG10K, who seem to be sensitive to micro-stutter, are free to shun multi-GPU and any other setups/cards they feel are not to their liking. Even people that make their buying decisions based on color should be free to do so, after all its their money. Guys should just let the ms issue rest: does it exist? yes; will everybody notice it? no; will I notice it? only way to know for sure is if you try it. Anything else IMO is a wasted effort.
On minimum FPS:
A single stated value of minimum fps doesn't tell you enough to make an informed decision. A better resource would be the distribution of the FPS. For all you know the dips might occur once in the game or, at the other end, every other frame (like in micro-stutter). I think an average FPS + variance, though not ideal, is better than the average-min-max set.
To more relevant stuff:
One of the downsides to die shrinking is the reduced surface area for heat dissipation, and I feel this is what may limit the achievable clocks.
It definitely would be interesting to see how it goes for Nvidia, it being their first 55nm, though I was expecting them to jump to 45nm. As AMD has found out in the CPU biz, being a node behind your competition could increase the hurt.
All in all, its a good time for the consumer. Now for Intel to join the fray!
Micro-stutter exists in both single GPU and multi-GPU setups, and its generally worse in the latter. People differ in the level of micro-stutter they can discern. Stating it as a general disadvantage for multi-GPU is not really fair, else I could say that anything above 60FPS (or even 30FPS) is wasted as a lot of people couldn't discern it. People like BFG10K, who seem to be sensitive to micro-stutter, are free to shun multi-GPU and any other setups/cards they feel are not to their liking. Even people that make their buying decisions based on color should be free to do so, after all its their money. Guys should just let the ms issue rest: does it exist? yes; will everybody notice it? no; will I notice it? only way to know for sure is if you try it. Anything else IMO is a wasted effort.
On minimum FPS:
A single stated value of minimum fps doesn't tell you enough to make an informed decision. A better resource would be the distribution of the FPS. For all you know the dips might occur once in the game or, at the other end, every other frame (like in micro-stutter). I think an average FPS + variance, though not ideal, is better than the average-min-max set.
To more relevant stuff:
One of the downsides to die shrinking is the reduced surface area for heat dissipation, and I feel this is what may limit the achievable clocks.
It definitely would be interesting to see how it goes for Nvidia, it being their first 55nm, though I was expecting them to jump to 45nm. As AMD has found out in the CPU biz, being a node behind your competition could increase the hurt.
All in all, its a good time for the consumer. Now for Intel to join the fray!
The first step is acknowledging it does exist of course. I agree that the only way to know if it bothers you is to see it first-hand, but for people to blindly recommend CF/SLI or deny its existence is poor and irresponsible advice. If I committed to multi-GPU and plunked down $400-600 for an "upgrade" only to find I couldn't stand it, I'd be pretty pissed.
I'd prefer FPS vs time graphs and an attached .txt or spreadsheet with frame dumps. Then there'd almost be no need for attached commentary as the graph itself would address many of the biggest issues you see around here. Unfortunately AT has stated they will not do this, so I'll have to continue looking elsewhere for truly relevant info.
There used to be quite a bit of discussion about die size and heat dissipation, but it really hasn't materialized in any negative manner. GT200 isn't their first move to 55nm, they've already cut their teeth with G92 9800 GT and GTX+. Its also good indication NV should be able to eek out higher clocks and hit previous OC'd model speeds even if they don't save on power or temps.
deadseasquirrel
Golden Member
Please define what you mean by "often". Because in the reviews I see, which actually show a 4870 vs 4870CF (and also show MIN fps... there aren't too many that do), one would be hard-pressed to call what you describe as "often".
Let's look here first. They test Bioshock, COD4, UT3, and Crysis across several resolutions. The only game where a CF setup has a lower MIN fps than a single 4870 is UT3. As for CF having the same MIN fps as a 4870, that tends to happen at lower resolutions such as 16x12, where one would be making a very questionable purchase to begin with.
Now, let's look here. They test A LOT more games... 10 total. Only in 2 of them do a 4870CF have lower MIN fps than a single 4870 (GRID and WIC, which they say is due to a driver bug).
Either way, I wouldn't call this as happening "often". But if you have other reviews you can link to that show this happening more, please do. I haven't been able to find too many reviews that show a CF vs a single card AND show min fps too. Few and far between. Maybe we'll see more once the X2 is released.
finbarqs
Diamond Member
remember guys, if the 4870X2 is ANYTHING like the 3870x2, then we're in for an awesome treat! First of all, it runs natively with standard ATI drivers, meaning that we do not need to enable "CF" mode to get it the framerates of CF! Second, DUAL MONITOR Support. Nothing from Nvida based on 2 GPU's can do dual head output, when SLI is enabled. ATI's you can.
Really seems like a lot of folks are grasping at straws for any reason they can to bash multi-gpu setups... Also hafta giggle when people bring up a few fluke minimum framerate issues especially when it's like 1 or 2 fps...
In the vast majority of current popular games and hit new release titles the 4870x2 does very very well. And it hasn't even been released yet so the drivers can only get better. If you want to buy a single gpu card that performs worse in the *vast majority of currently popular titles that people will be playing* because you play some old and/or obscure titles that most people don't, well, cool, do that then...but don't try to use that to bash multi-gpu solutions. That's like bashing a corvette because it doesn't have four seats or something...makes no sense.
And I'm sure at some point after the die shrink there will be a good ol' 280x2 card too...
As multi gpu becomes more and more popular, and it looks like that is the direction ATI as a company is heading, support is going to get much more mature and kinks will be worked out much faster than they are now with them being a niche product.
In the vast majority of current popular games and hit new release titles the 4870x2 does very very well. And it hasn't even been released yet so the drivers can only get better. If you want to buy a single gpu card that performs worse in the *vast majority of currently popular titles that people will be playing* because you play some old and/or obscure titles that most people don't, well, cool, do that then...but don't try to use that to bash multi-gpu solutions. That's like bashing a corvette because it doesn't have four seats or something...makes no sense.
And I'm sure at some point after the die shrink there will be a good ol' 280x2 card too...
As multi gpu becomes more and more popular, and it looks like that is the direction ATI as a company is heading, support is going to get much more mature and kinks will be worked out much faster than they are now with them being a niche product.
Hauk
Platinum Member
Specs according to an SLI forum member..
240 Shaders
- 80 TMU's
- 32 ROP's
- 512 Bit bus (thus likely 1GB of GDDR3 RAM)
- Shader-Ratio back at 2,5 from 2,15
- More transistors as GT200 (with the sole purpose to increase the shader clock)
- Shaderdomain works in 54MHz steps
- 1166 GFlops
- 648/1620 or 675/1674 (core/shader) - these are likely options only
- Probably TDP 236W
He won't reveal his source though..
240 Shaders
- 80 TMU's
- 32 ROP's
- 512 Bit bus (thus likely 1GB of GDDR3 RAM)
- Shader-Ratio back at 2,5 from 2,15
- More transistors as GT200 (with the sole purpose to increase the shader clock)
- Shaderdomain works in 54MHz steps
- 1166 GFlops
- 648/1620 or 675/1674 (core/shader) - these are likely options only
- Probably TDP 236W
He won't reveal his source though..
ArchAngel777
Diamond Member
I don't agree. I think people want the best bang for the buck, and an SLI/Xfire will never provide that until it can scale ~95% in every game without relying on ATi or nVidia writing a driver specifically for it. I can't remember who said it here, but someone said a long time ago when responding to Rollo that 'SLI and bang for the buck' should never be used in the same sentence. Of course, currently, it shouldn't!
The only thing that particular annoys me is all the people who once praised SLI are now singing different tune when ATi finally has the fastest single card, via multi-GPU setup. I know who these people are and so does the majority of the active forum. Best to just ignore their posts in general as you know they cannot ever be honest or truthful, at least, not anymore than your average politician.
In any case, ATi will have the fastest single card via Mult-GPU. nVidia has the fastest single GPU and nVidia has the fastest multi-GPU setup. Those are the facts... Doesn't matter if you don't want to use multi-GPU or not, those are still the facts.
Instead of this tiresome bickering between both sides, why not just admit the facts, then leave it up to your wallet as for what card you decide to go with? Seems like a no-brainer to me. Ya'll got to much time on your hands if you can sit here and argue over graphics cards over and over and over...
bryanW1995
Lifer
I'm not. It will perform just like an overclocked gt 200. More importantly, it will be cheaper for nvidia to produce once they get decent yields out of it. Hopefully, they'll introduce a midrange gt 200 offering at the same time.
bryanW1995
Lifer
It's a pretty safe bet that keys has owned a lot more video cards than you have.
I understand that you want to support ati right now, but all you're doing is making them look bad by getting into a pissing match with chizow. Also, you keep baiting mods like an out-of-control pennsylvania bear hunter. You need to tone it down.
bryanW1995
Lifer
I think that they'd be smarter to switch to gddr5 and 256 bit. As long as gddr5 is plentiful they'll get similar/better performance costs will be quite a bit lower.
LittleNemoNES
Diamond Member
damn I should have waited a month instead of buying my GTX280 😛
evolucion8
Platinum Member
Actually, the HD 4870X2 outperforms the GTX 280 in SLI more often than not. But quite strange considering that a single HD 4870 cannot outperform a 280, seems that the scaling in Crossfire is outstanding (When it works). So nVidia has the fastest single GPU, ATi has the fastest single via Multi GPU card, and both trade blows in Multi GPU setups.
http://www.anandtech.com/video/showdoc.aspx?i=3354&p=4
http://www.hardocp.com/article...wzLCxoZW50aHVzaWFzdA==
There some more other websites, but I'm sleepy as hell. Crysis for some reason doesn't scale well with Crossfire, ATi should fix it soon.
I actually did mention that in my pro-NV rant as a negative bullet point against SLI. 😉 Its been rumored though that NV will be fixing this and adding some other features in their Big Bang II driver in September. Personally I think they're going to push multi-GPU + multi-monitor gaming with extended resolutions like Matrox tried with Parhelia/Hydravision.
In the short-term, it looks like we'll see WHQL PhysX drivers on Aug 5 that extend support to the older 8-series G80s and G92s.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 25K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 24K
-
-
