5850 just as fast as a 5870?

[ugaboga232]

So you still think you know what xfire and sli are but that is not what they are. You are thinking of what lucid hydra promises to be.

 

[Daedalus685]

Ohh, I get it... That is what I thought before.. but noone by that name posted in this post, and he used the same word again...

Wow. Do I need to link to my profiles on other forums that use this exact same screen name?

By the way, when I say minimum frame rates are erratic I mean they are not consistent with bandwidth starving. They occur over very short bursts here ant there, the deviation in the performance of a 5770 is very large, which I'd usually blame on drivers. Even if it is because of the GPU being crappy, there is still a lot of evidence to show that the issue is not with the bandwidth.

 

[AzN]

Azn show us one bit of solid data that the 5800 is memory bandwidth bottlenecked. Until then do not change the discussion into one about fake accounts.

I've already explained this in my previous posts but it's amusing for you.

4870x2 > 5870 same amount of SP, tmu, rop. 5870 with higher clock speed is getting beat in most cases. Before you blame drivers remember where the 5870 architecture came from. All ATI has to do with change few numbers and voila you have same optimzation as RV770.

People with 2 post counts just don't come into the same thread talking about same references and words that cross reference from another. It's so hilarious you guys deny it. Should I come back with another account so I can be cool as you guys and support myself? LOL

 

[Daedalus685]

They put up double gb/s numbers though including techreport, anandtech, and so forth...

They post that the bandwidth is double technically. Which it is. This is not the same thing as actually having access to move double the files around.

If the memory systems remain independent, then why do I not have a 2gb frame buffer when i crossfire my 4890s?

I was not claiming that you said the 5970 has a 512bit bus.. I was simply using it as an example on the usual notation as it was the last crossfire review I have read.

Just because a review likes to say "Now with double the raw power" does not mean it is so. It is just a colloquial journalism thing to do. No harm done unless one takes everything everyone says as the unwavering truth.

 

[Daedalus685]

I've already explained this in my previous posts but it's amusing for you.

4870x2 > 5870 same amount of SP, tmu, rop. 5870 with higher clock speed is getting beat in most cases. Before you blame drivers remember where the 5870 architecture came from. All ATI has to do with change few numbers and voila you have same optimzation as RV770.

People with 2 post counts just don't come into the same thread talking about same references and words that cross reference from another. It's so hilarious you guys deny it. Should I come back with another account so I can be cool as you guys and support myself? LOL

I don't know about these other folks.. but I have already stated who I am, and how you can find more info on me. Perhaps this thread got a lot of attention because of the nonsense begin thrown around in it?

The fact that the 5870 is slower than a 4870x2 proves nothing. It means that something is different, but there is no reason to think it is memory any more than dx11 changes, driver issues, etc etc. In fact we have evidence to indicate it is probably not memory, and more than likely because of architecture (dx11) or drivers (on purpose or not).

You may not appreciate how difficult it is to write a driver... but it is not a Sunday drive.. There were large changes to the 5000 series over the 4000 that drastically change the drivers. But hey, since the 4000 series was just a slight reworking of the 3000, which was a reworking of the 2900, and they even use the same dx level, then the drivers on release of the 4870 must have been perfectly mature right? There certainly couldn't be evidence of driver improvements already posted in this thread as they are pretty much the same architecture.

(Edit, there I even finally updated my avatar to teh same one I use at Tom's...)

 

[ugaboga232]

Ok azn you obviously do not get it. We do not know why the 4870x2 beats the 5870. What we think is that drivers work better in the 2 800 chunks while the 5870 is not fully utilized. Memory bottleneck IS NON EXISTENT. The only other option would be serious hardware problems.

Also I have been a long time lurker. Notice how i had one post and joined in september? You really think I would make a new account just to bother you? Also its december so why would I have another account from november? You are just grasping at straws. Also BFG agrees with us so are we his alts?

Azn please learn what sli and xfire are then we can have meaningful discussions.

Edit: How could we both post at the same time?

Also supposedly 16x AF takes quite a toll on the new cards (5XXX) because of their far better algorithm. That is also a possible candidate for the low performance as every review site uses 16xAF nowadays.

 

[AzN]

They put up double gb/s numbers though including techreport, anandtech, and so forth...

They post that the bandwidth is double technically. Which it is. This is not the same thing as actually having access to move double the files around.

If the memory systems remain independent, then why do I not have a 2gb frame buffer when i crossfire my 4890s?

That's because the card is rendering alternate frames or alternate checkerboards essentially offloading to individual cards. That still wouldn't make 2 4890 cards in 2gb frame buffer.

I was not claiming that you said the 5970 has a 512bit bus.. I was simply using it as an example on the usual notation as it was the last crossfire review I have read.

Just because a review likes to say "Now with double the raw power" does not mean it is so. It is just a colloquial journalism thing to do. No harm done unless one takes everything everyone says as the unwavering truth.

"now double the raw power" is marketing gimmick I agree but when they put up numbers adjacent to single cards they are reviewing it sure does but when I do it's no no for you.

 

[ugaboga232]

Review sites can be wrong. You should learn what you are talking about before you come into this thread. And they put the double bandwidth because it has double bandwidth. But it must support the framebuffer so a lot of the bandwidth is wasted making it a moot point.

 

[Daedalus685]

That's because the card is rendering alternate frames or alternate checkerboards essentially offloading to individual cards. That still wouldn't make 2 4890 cards in 2gb frame buffer.




"now double the raw power" is marketing gimmick I agree but when they put up numbers adjacent to single cards they are reviewing it sure does but when I do it's no no for you.

/facepalm

You simply don't understand what you are talking about azn.. You need to see that there is a difference in the effective bandwidth depending on how you use teh hardware. Also, that AFR has nothing to do with the frame buffer at all.

Yes, each card is rendering a frame, and storing it into a frame buffer.. The frame buffer is stored in the memory of the cards, and is doubled so that it is on each card. Do you know what a frame buffer does? It holds the date to be drawn to the display, since only one card can send display information it has to have the entire buffer.

 

[AzN]

The fact that the 5870 is slower than a 4870x2 proves nothing. It means that something is different, but there is no reason to think it is memory any more than dx11 changes, driver issues, etc etc. In fact we have evidence to indicate it is probably not memory, and more than likely because of architecture (dx11) or drivers (on purpose or not).

It proves everything. Different? Sure... Slightly upgraded architecture to improve performance which it does in some SP hungrier games and still gets beaten by 4870x2 in most cases. So you really don't think 5870 can get more performance with same amount of bandwidth as 4870x2 and maybe beat 4870x2 in its own game? Kind of absurd observation when 5870 gains performance when more bandwidth is applied.


These are simply theoretical peak values, and they are in some cases quite academic. You're rarely going to hit the peak pixel fill rate on one of these cards, for instance, since memory bandwidth will likely limit you first. The GeForces probably won't ever reach their peak dual-issue shader FLOPS numbers, and I doubt the Radeons will achieve more than 80% of their peak compute capacity when executing pixel shaders. On top of all that, we've just thrown in the numbers for the multi-GPU solutions, even though they'll rarely scale linearly.


Yeah that's it.. ATI is purposely giving you crappier performance over 4870x2.

I'm really Genghis Khan who crossed the Pacific to the US so I can rule you all. Sheee don't tell anyone!

 

[ugaboga232]

So you really don't like listening? Core raises performance much more than bandwidth. The SP's of the 5870 just are not being utilized. Whether that is hardware or driver (far more likely) is yet to be seen.

I am going to sleep now but I hope you can read up on what xfire ACTUALLY is. Till then, any discussion with you is pointless.

 

[AzN]

/facepalm

You simply don't understand what you are talking about azn.. You need to see that there is a difference in the effective bandwidth depending on how you use teh hardware. Also, that AFR has nothing to do with the frame buffer at all.

Yes, each card is rendering a frame, and storing it into a frame buffer.. The frame buffer is stored in the memory of the cards, and is doubled so that it is on each card. Do you know what a frame buffer does? It holds the date to be drawn to the display, since only one card can send display information it has to have the entire buffer.

Do you know what alternate frame is Daedalus? In crossfireX you off load the card by alternating frames that's what crossfire does. not raid array you set and forget to get better read performance.

 

[Daedalus685]

I think that the 5870 is not an exact doubling of the 4870x2. As the last time I checked the 4870 doe snot support dx11. I have no idea what kind of hit this has on shader for shader performance. But given what we have seen, it is either an obvious hit, or the drivers just are not up to snuff.

Of course the 5870 gains performance with bandwidth.. but all that says is that the GPU is not a 100% perfect bottleneck.. by the way out side of bottles what is?

 

[ugaboga232]

AZN, YOU HAVE NO CLUE WHAT XFIRE IS. Please do not think that you are right when 3 people have told you that you are not. Learn what it really is then we can talk.

Goodnight

 

[AzN]

So you really don't like listening? Core raises performance much more than bandwidth. The SP's of the 5870 just are not being utilized. Whether that is hardware or driver (far more likely) is yet to be seen.

I am going to sleep now but I hope you can read up on what xfire ACTUALLY is. Till then, any discussion with you is pointless.

It sure is getting utilized by 4870x2 though since it has more bandwidth. While 5870 still shy of 4870x2 performance limited to it's bandwidth with 12% higher core and shader clocks.

 

[Daedalus685]

Do you know what alternate frame is Daedalus? In crossfireX you off load the card by alternating frames that's what crossfire does. not raid array you set and forget to get better read performance.

UHHH...

You have no idea how any of this works...

Crossfire (AFR) splits the GPU rendering up, it does NOT split the rendering up per card. The memory is needed for display of the frames that are computed. Unless you found a crossfire monitor that allows you to plug into both cards at the same time.

Again, as I have already said, the raid analogy was just for the memory.. you clearly did not understand it.. forget the word raid..

Do some reading.. I'll check back tomorrow.

 

[AzN]

I think that the 5870 is not an exact doubling of the 4870x2. As the last time I checked the 4870 doe snot support dx11. I have no idea what kind of hit this has on shader for shader performance. But given what we have seen, it is either an obvious hit, or the drivers just are not up to snuff.

Of course the 5870 gains performance with bandwidth.. but all that says is that the GPU is not a 100% perfect bottleneck.. by the way out side of bottles what is?

5870 is more than double of 4870x2 far as core goes but still shy of 4870x2 performance regardless. It never reaches peak performance because it's it's bandwidth shy while 4870x2 with more than enough reaches further in peak performance beating out 5870. So sorry..

 

[Daedalus685]

It sure is getting utilized by 4870x2 though since it has more bandwidth. While 5870 still shy of 4870x2 performance limited to it's bandwidth with 12% higher core and shader clocks.

Again.. the 5870 is not the 4870x2.. they are different architectures, use different parts of the driver, and are not an apples to apples comparison. No one with any semblance of scene would expect identical performance, only roughly equivalent.. which is what we got.

 

[AzN]

UHHH...

You have no idea how any of this works...

Crossfire (AFR) splits the GPU rendering up, it does NOT split the rendering up per card. The memory is needed for display of the frames that are computed. Unless you found a crossfire monitor that allows you to plug into both cards at the same time.

Again, as I have already said, the raid analogy was just for the memory.. you clearly did not understand it.. forget the word raid..

Do some reading.. I'll check back tomorrow.

now RAID is what you used originally? Or is this another multi-acount troll?

Alternate-frame rendering — The king of all multi-GPU load balancing modes is alternate-frame rendering, or AFR for short. AFR interleaves full frames rendered by the two cards, so that, say, the master card renders odd frames and the slave card renders even ones. This is the preferred load-balancing mode whenever possible, because AFR shows markedly better performance scaling than other modes. Part of the reason for AFR's good performance is the fact that it splits the geometry processing load between the two cards evenly, something no other mode does.

It sure says AFR split the rendering up per card.

 

[AzN]

Again.. the 5870 is not the 4870x2.. they are different architectures, use different parts of the driver, and are not an apples to apples comparison. No one with any semblance of scene would expect identical performance, only roughly equivalent.. which is what we got.

They are not different architecture. It's like calling e8400 different architecture opposed to E6600.

 

[Daedalus685]

They are not different architecture. It's like calling e8400 different architecture opposed to E6600.

No, the other chap in the 5770 used raid originally. But it is a good analogy for the memory.. but you don't understand what it means. (So i guess it is not a good analogy). It only applied to the memory, just like how raid1 does not give you faster writes, neither does crossfire or sli.

As for the architecture.. The 5870 is dx11, so are you trying to tell me you think that the 3870 is dx11 also? I mean they are pretty much teh same architecture right? The 3870 is a lot more like the 4870 than the 5870 is the 4870.. yet drivers still improved for the 4000.

it is not like the e6600 and e8400 at all... Since you know so much about micro architecture, why is it that you think dx11 will not hurt shader scaling?Obviously with proof please.

 

[AzN]

No, the other chap in the 5770 used raid originally. But it is a good analogy for the memory.. but you don't understand what it means. (So i guess it is not a good analogy). It only applied to the memory, just like how raid1 does not give you faster writes, neither does crossfire or sli.

As for the architecture.. The 5870 is dx11, so are you trying to tell me you think that the 3870 is dx11 also? I mean they are pretty much teh same architecture right? The 3870 is a lot more like the 4870 than the 5870 is the 4870.. yet drivers still improved for the 4000.

it is not like the e6600 and e8400 at all... Since you know so much about micro architecture, why is it that you think dx11 will not hurt shader scaling?Obviously with proof please.

instead of pointing to someone and saying you don't know what you are talking about and using a bad analogy to cross reference the term perhaps you should link with all the stuff you said in reference to crossfire.

So E8400 with SSE4 and bigger cache different clocks is somehow different architecture opposed to E6600 with SSE3 with smaller cache and lower clocks?

Although 3870 also share the same architecture as RV770 or 5870 BUT there was huge change with RV770 with ROP and the way AA functions with the card. In case RV770 and 5870 the difference is so subtle mostly marketing gimmick like dx11, eyefinity, SSAA to confuse you I suppose.

 

[hans030390]

Lol.

This reminds me of how I acted on these forums almost ~5 years ago (when I was 14 or 15, specifically regarding SM3.0). To any of you that are still reading (and remember me from then), I do apologize for my behavior...though I was still able to play Bioshock on my 6600gt when the x800 cards could not, which was something I thought might happen (albeit on low/medium settings).

Back on topic, AzN just has some strange hatred towards the 5xxx series. I wouldn't bother arguing with him anymore.

 

[AzN]

I don't have strange hatred towards 5xxx series just laying some hard facts. I just like dissecting hardware and get in to the nutty gritty of things and not some fud going around forums because someone who you think cool said so.

In case of these multi-account trolls it's obvious Schmide is one. if it made him feel better by hiding his original account and telling me I don't know what I'm talking about with personal references all power to him. I can also guess the other guy. So hilarious.

Since Schmide has not posted in this thread before this point, this is considered a member call out. Please refrain from this in the future. If you suspect multiple accounts by other members, PM the moderator account for review. But do not grind your axe here please.

Anandtech Moderator - Keysplayr

 

[MrBorsa]

I don't have strange hatred towards 5xxx series just laying some hard facts. I just like dissecting hardware and get in to the nutty gritty of things and not some fud going around forums because someone who you think cool said so.

In case of these multi-account trolls it's obvious Schmide is one. if it made him feel better by hiding his original account and telling me I don't know what I'm talking about with personal references all power to him. I can also guess the other guy. So hilarious.

I agree with this