Sign in with
Sign up | Sign in
Your question

5830

Last response: in Graphics & Displays
Share
January 26, 2010 8:53:37 PM

wasnt the 5830 supposed to be released yesterday? :??: 

More about : 5830

January 26, 2010 9:02:53 PM

I believe it should be released on the 5th of Feb.
January 26, 2010 9:07:38 PM

any word on specs and pricing
Related resources
January 27, 2010 3:19:12 AM

I'd expect the price to be a bit lower than 4890. However, some sources put the price at $230. As for specs, I can't find anything on the internet to give. But considering the 5800 series have a 256 bit bus width, the 5830 should have it too but with less stream processors than the 5850. This should give it great scaling for overclockers since the 5870 seems to be a bottlenecked by the memory bandwidth (as seen in some reviews, overclocking the core doesn't do as much as overclocking the memory). The 5770 suffers the same memory bandwidth problem when overclocking. Granted the 5770 performs similarly to the 4870, I'd put the performance around the level of the 4890. However, this is speculation as not much information can be found at this time.
January 27, 2010 3:29:40 AM

The $230 seems more likely to me. The 5750, 5770 and 5670 cards have all been priced $20--30 higher than their HD4000 counterparts at launch.
January 27, 2010 3:38:17 AM

jyjjy said:
The $230 seems more likely to me. The 5750, 5770 and 5670 cards have all been priced $20--30 higher than their HD4000 counterparts at launch.


Yeah, it definitely could be considering the premium amd is charging for having dx11 and all the perks that comes with the new gen. I've been waiting to get a new card and the 5000 series just doesn't seem to have the value I'm looking for yet. This is unlikely to change with the release of the 5830 but that doesn't keep me from hoping...
January 27, 2010 3:43:56 AM

chowmanga said:
I've been waiting to get a new card and the 5000 series just doesn't seem to have the value I'm looking for yet.

Hmm. Both the HD5770 and HD5850 look like the best performance for the money in their price range. Not sure what "value" you were expecting if they aren't good enough.
January 27, 2010 3:59:25 AM

jyjjy said:
Hmm. Both the HD5770 and HD5850 look like the best performance for the money in their price range. Not sure what "value" you were expecting if they aren't good enough.


Neither of those cards are a slouch. However the 5850 is beyond my price range and with the 5770 and 4870 similar in price and performance, I don't see the 5770 stand out. Trust me, I know all about the features the 5770 brings over 4870 but the scalability issues in overclocking it is the deal breaker for me. Even though you can overclock the 5770 much further than the 4870, it doesn't mean much if you're not seeing much gain. You're right, at this point the 5850 and the 5770 are the best choices at those price points but I just want more out of my money. Its personal preference.
January 27, 2010 4:01:31 AM

Why would the HD5770 not gain much performance with an OC? I haven't heard anyone else saying that.
January 27, 2010 4:36:58 AM

Look at the overclocking tables tables for each of the three cards in this review released here on THG: http://www.tomshardware.com/reviews/radeon-5770-overclo...

Notice that the gtx275 and especially the 4890 gains much more from core overclocking. The 5770 gains very little from core overclocking and a ton from overclocking the memory. This is the scalability issue I'm talking about - the 5770 is memory starved.

I don't think it says explicitly that the scalability is an issue and while the performance is there, I'd be happier if the card was better balanced so that it provided more equal scalability with core and memory overclocks. This way I'd be held back by the limitations of the hardware, rather than inherent limitations on design. I want to be at the point where temps are whats stopping me from getting more fps, not drops in performance from overclocking the gddr5 too much when the core could be pushed further.
January 27, 2010 7:00:27 AM

chowmanga said:
Look at the overclocking tables tables for each of the three cards in this review released here on THG: http://www.tomshardware.com/reviews/radeon-5770-overclo...

Notice that the gtx275 and especially the 4890 gains much more from core overclocking. The 5770 gains very little from core overclocking and a ton from overclocking the memory. This is the scalability issue I'm talking about - the 5770 is memory starved.

I don't think it says explicitly that the scalability is an issue and while the performance is there, I'd be happier if the card was better balanced so that it provided more equal scalability with core and memory overclocks. This way I'd be held back by the limitations of the hardware, rather than inherent limitations on design. I want to be at the point where temps are whats stopping me from getting more fps, not drops in performance from overclocking the gddr5 too much when the core could be pushed further.

I honestly have no idea what you are talking about. First of all if it had issues with "scalibilty" that would mean that they were able to get a large OC but the performance didn't increase in line with OC amount but that's not what is going on in the article. The only thing that article shows is that they had a hard time getting a good OC in the core. They OCed the core 5% and the memory 19% and saw performance increase 8% overall. What is this "inherent limitation" of design that you are blaming on the memory? What does "drops in performance from OCing the gddr5 too much when the core could be pushed further" mean? They didn't stop OCing the core when it could be pushed further, they OCed it as much as they could. They didn't OC the memory too much, they OCed it as much as they could. The only thing that article shows is that the particular card they tested had a core that didn't OC well, that's it. I don't know where this other stuff you are talking about comes from.
Here is another article about OCing the card and they managed to get a 13% OC on the core and 21% on the memory;
http://www.overclockersclub.com/reviews/sapphire_hd5770...
and another with a similar OC;
http://www.hardocp.com/article/2009/10/12/amd_ati_radeo...
and one more with a good core OC but they couldn't OC the memory as much as others;
http://www.neoseeker.com/Articles/Hardware/Reviews/hd57...
So anyway the card has no issues with overclocking and that the memory OCs even more than the core is a GOOD thing as the memory bandwidth is a lil weak for a GPU of that caliber. I think you were basing way too much on one article and then started coming up with weird theories to explain their results.
January 27, 2010 7:41:42 AM

As with all ocees, your results may vary. Card to card, chip to chip, ram to ram, weve all seen it, so if you get one, and plan to oc, just hope you get a good oceer
January 27, 2010 1:59:02 PM

jyjjy said:
The $230 seems more likely to me. The 5750, 5770 and 5670 cards have all been priced $20--30 higher than their HD4000 counterparts at launch.


All the more reason for you to be pulling for team green! All of the 4000 series parts were launched against some VERY stiff competition from Nvidia and had to be priced accordingly... the 5000 series? Nvidia still hasn't sold a single DX11 GPU and definitely doesn't hold the performance crown... AMD/ATI is just doing what they're supposed to be doing... charging a price premium for a premium product.
January 27, 2010 2:07:11 PM

From what I have read on numerous reviews over the interwebs, is that the 5770s have very GOOD CF scalability, especially considering the immature drivers. This has been somewhat backhandedly attributed to the 128bit bus limiting a single cards performance, but allowing multiple card configurations to run at a high % of scale, due to the cards limitations almost always being on the card, and not the system installed. I am not technically proficient enough to back these concepts, just repeating what I have gathered over time.
January 27, 2010 2:14:01 PM

A 5770 is slower than a cheaper 4870 every day of the week. And if you o/c both the gap widens. Hopefully this new card is more like a 5850 than a 5770. Of course the 5850 was supposed to be 269.99 and everyone in the beginning saw the price of that dropping in ensuing months. But in light the huge performance gap between itself and the 5770, They took the route of raising the pricepoint of the 5850 instead of lowering the price of the 5770.
January 27, 2010 2:24:33 PM

notty22 said:
A 5770 is slower than a cheaper 4870 every day of the week.

If you use the old beta drivers every day of the week then sure, I guess so, but unless you have invented time travel the HD4870 is no longer cheaper.
January 27, 2010 2:25:56 PM

Of course they did, ATi are a business and are out to make money, with no competition they can charge higher prices, yes its annoying that there so expensive, but there not doing anything wrong. Just making money
January 27, 2010 2:28:54 PM

At least they opened with a crazy competitive price, and let the market dictate from there, as opposed to opening at a crazy high price, and letting the market and competition cut it down.

Launch prices were competitive with a non-existant competitor. These are small premiums considering the performance the cards give in direct comparison to current offerings.
January 27, 2010 2:33:13 PM

^^^Guys your are preaching to the wrong person here. Read his other posts and see that your efforts are meaningless.
January 27, 2010 2:36:54 PM

JofaMang said:
At least they opened with a crazy competitive price, and let the market dictate from there, as opposed to opening at a crazy high price, and letting the market and competition cut it down.

Launch prices were competitive with a non-existant competitor. These are small premiums considering the performance the cards give in direct comparison to current offerings.


Very true, ATi's HD5850 beating nVidia's fastest single GPU card for less money and offering better features/tech.......makes it an easy choice for someone looking for great performance and the latest tech.
January 27, 2010 2:38:41 PM

hallowed_dragon said:
^^^Guys your are preaching to the wrong person here. Read his other posts and see that your efforts are meaningless.

This coming from someone who mentions he hates a company in every other post.
laughing it off doesn't hide what that is.
January 27, 2010 2:40:53 PM

notty22 said:
This coming from someone who mentions he hates a company in every other post.
laughing it off doesn't hide what that is.


Nice to mention only part of my specific post. As I said, my hate for the company only affects my purchases. If nVidia has better parts I will recommend them to others as I have previously in the past :kaola: 
January 27, 2010 3:29:21 PM

jyjjy said:
I honestly have no idea what you are talking about. First of all if it had issues with "scalibilty" that would mean that they were able to get a large OC but the performance didn't increase in line with OC amount but that's not what is going on in the article. The only thing that article shows is that they had a hard time getting a good OC in the core. They OCed the core 5% and the memory 19% and saw performance increase 8% overall. What is this "inherent limitation" of design that you are blaming on the memory? What does "drops in performance from OCing the gddr5 too much when the core could be pushed further" mean? They didn't stop OCing the core when it could be pushed further, they OCed it as much as they could. They didn't OC the memory too much, they OCed it as much as they could. The only thing that article shows is that the particular card they tested had a core that didn't OC well, that's it. I don't know where this other stuff you are talking about comes from.
Here is another article about OCing the card and they managed to get a 13% OC on the core and 21% on the memory;
http://www.overclockersclub.com/reviews/sapphire_hd5770...
and another with a similar OC;
http://www.hardocp.com/article/2009/10/12/amd_ati_radeo...
and one more with a good core OC but they couldn't OC the memory as much as others;
http://www.neoseeker.com/Articles/Hardware/Reviews/hd57...
So anyway the card has no issues with overclocking and that the memory OCs even more than the core is a GOOD thing as the memory bandwidth is a lil weak for a GPU of that caliber. I think you were basing way too much on one article and then started coming up with weird theories to explain their results.


GDDR5 memory has built in error checking. AMD implemented it so that if an error is detected, the memory will resend the data. This will lead to a drop in performance if too many errors have to be corrected. Unlike GDDR3 where you'll see artifacting when the memory is pushed too far, overclocking the GDDR5 too much will result in a drop in frames because of the frequent resending of data.

As for design limitations consider this. The architecture for the 5000 and 4000 series are similar as they're evolutions on the r600. Both the 5770 and 4870 have 800 sp's while the 4870 has almost double the memory throughput. As you said yourself, memory bandwidth is weak for that GPU. I'd rather not have to deal with alleviating the bottleneck.

I prefer to look at AIB partner's card rather than samples straight from AMD: http://www.overclockersclub.com/reviews/sapphire_hd5770...

Overclocked, you see a 10fps gain in batman no AA and at most a 9fps gain in l4d. Far Cry gets 6fps and crysis gets 2 and all these numbers from from the lowest resolution. At the resolution I play at, gains are smaller and thats just not enough for me. Call me greedy, but when the 5770 keeps the games I want to play under 60fps now, I'm going to see the card age quickly when newer games come out in the future, even with driver updated. I've been keeping a close eye on the 5770 and have read many reviews of the card, especially one with aftermarket coolers, and I would be the first one to recommend it to someone who needs a card NOW. For myself, I'd want to pay less for that card.
January 27, 2010 4:57:40 PM

chowmanga said:
GDDR5 memory has built in error checking. AMD implemented it so that if an error is detected, the memory will resend the data. This will lead to a drop in performance if too many errors have to be corrected. Unlike GDDR3 where you'll see artifacting when the memory is pushed too far, overclocking the GDDR5 too much will result in a drop in frames because of the frequent resending of data.

I'm aware of this but I don't see how it has anything to do with what you said. Unless you are implying that the people writing these articles are OCing the card wrong? That seems rather presumptuous and still wouldn't limit the OC on the core as you implied. Only one of the articles said they were using an engineering sample and I linked three. If you look around on forums you will see that the OCs in the articles are fairly typical so I still don't follow what you were getting at.
I've seen a lot of gnashing of teeth over the card's 128 bit bus but I really don't get it. Would the card perform better with a wider bus? Sure. This doesn't change the cards ACTUAL performance however which is quite good for its price and that the ram OCs so well makes it even less of an issue.
If you expect a card that can play every game at over 60 fps on the highest settings at high resolutions for $150 well then your expectations are at fault, not the card and certainly not its ability to OC which seems fine by any reasonable standard. I also wouldn't be so sure that many games that push these cards harder than current games are going to be coming around in the near future. The two most intensive games around are still Crysis and Stalker which are both PC exclusive and years old. Almost all notable games are multiplatform these days and until we get the PS4 and/or xbox720 I doubt games are going to become that much more intensive on GPUs.
January 27, 2010 5:16:31 PM

Well, if its the 128 bit bus, then Fermi is doomed, as G200 had a 512.
Its about time people just accepted what GDDR5 brings to the table. To me, the worse thing about GDDR5 is its idle states, or 2D states, where its finicky at lower clocks
January 27, 2010 5:50:34 PM

jyjjy said:
I'm aware of this but I don't see how it has anything to do with what you said. Unless you are implying that the people writing these articles are OCing the card wrong? That seems rather presumptuous and still wouldn't limit the OC on the core as you implied. Only one of the articles said they were using an engineering sample and I linked three. If you look around on forums you will see that the OCs in the articles are fairly typical so I still don't follow what you were getting at.
I've seen a lot of gnashing of teeth over the card's 128 bit bus but I really don't get it. Would the card perform better with a wider bus? Sure. This doesn't change the cards ACTUAL performance however which is quite good for its price and that the ram OCs so well makes it even less of an issue.
If you expect a card that can play every game at over 60 fps on the highest settings at high resolutions for $150 well then your expectations are at fault, not the card and certainly not its ability to OC which seems fine by any reasonable standard. I also wouldn't be so sure that many games that push these cards harder than current games are going to be coming around in the near future. The two most intensive games around are still Crysis and Stalker which are both PC exclusive and years old. Almost all notable games are multiplatform these days and until we get the PS4 and/or xbox720 I doubt games are going to become that much more intensive on GPUs.


Sorry if I wasn't clearer, I'm not talking about OC potential at all. None of these reviewers are doing anything wrong. What I don't want is for the memory to be clocked to the highest possible point and see no gains from further core overclocking. I spend a lot of time overclocking my hardware to get the best balance possible. I'm not going to continue overclocking the core if I see no difference because of the lack of memory bandwidth. When testing my overclocks, I'll only clock as high as I see a gain. We all know how cool the 5770 runs and to me, it would be disappointing if during overclocking, change in core speeds no longer give me additional performance when the temps are good and the fan speed running at 50%. This is what I meant when I said I was worried about bandwidth keeping me from pushing the core further. What I want are gains in performance, not number of clock cycles per second.

My expectations are not at fault. What if I want that card for $130, rather than $150-$200? You even mentioned the premium amd is charging for 5000 series. If you remove the premium that you said ($20 - $30) the card looks much more attractive to me. Again, priced the same as a 4870, the 5770 doesn't stick out to me. I'm not in dire need of a card so I'm willing to wait. If I could overclock the card to much higher performance levels, then I'd be willing to pay the premium, but that is not the case.

And I would count on a game coming out demanding more hardware. There will always be ports from consoles that require more hardware than they should (mass effect, gta4) and with the declining number of PC exclusives, I'd imagine this to happen.
January 27, 2010 6:22:28 PM

JAYDEEJOHN said:
Well, if its the 128 bit bus, then Fermi is doomed, as G200 had a 512.
Its about time people just accepted what GDDR5 brings to the table. To me, the worse thing about GDDR5 is its idle states, or 2D states, where its finicky at lower clocks

........................" Fermi is doomed "........................
Time to get some new material.

January 27, 2010 6:51:21 PM

I was refering to it only having 384 vs last gen. It really has nothing to do with Fermi, and everything to do with bus' and GDDR5, so......
January 27, 2010 8:22:02 PM

chowmanga said:
Sorry if I wasn't clearer, I'm not talking about OC potential at all. None of these reviewers are doing anything wrong. What I don't want is for the memory to be clocked to the highest possible point and see no gains from further core overclocking. I spend a lot of time overclocking my hardware to get the best balance possible. I'm not going to continue overclocking the core if I see no difference because of the lack of memory bandwidth.

I still don't get it. The memory OCs more than the core so how is the memory going to limit your return on a core OC? The card also scales very well in crossfire which offers no increase in memory bandwidth so obviously the memory isn't as much of a limit on performance as you seem to want to believe.
January 27, 2010 9:34:30 PM

sorry
January 27, 2010 10:12:48 PM

jyjjy said:
I still don't get it. The memory OCs more than the core so how is the memory going to limit your return on a core OC? The card also scales very well in crossfire which offers no increase in memory bandwidth so obviously the memory isn't as much of a limit on performance as you seem to want to believe.


Good scaling with crossfire doesn't tell me anything more than the fact that the 1gb frame buffer is sufficient for two 5770 processors. Regardless of rendering method (tiling, alternate frame, half screen), the same data is being loaded into both cards so if a crossfire setup is scaling well, then it's not limited by the amount of memory. Good crossfire scaling is not good enough to make a conclusion on memory bandwidth.

Also, memory overclocking potential and core overclocking potential don't have any correlation. If memory speed is a bottleneck, then overclocking the core isn't going to have much effect. Even though the memory can overclock higher, are you positive it will remove the bottleneck from the memory and put it on the core?

Its not that I want to believe the 5770 lacks bandwidth. The lack of bandwidth is most apparent in the initial reviews of 5770 where it was performing much below the 4870. At the start, the efficiency of the sp's in the 5770 and 4870 were very comparable because they're based on the same architecture. However, with the new generation AMD added new instructions to speed up the pipelining of some specific operations. In implementing these instructions with driver updates, we see the 5770 performing at the level of the 4870. You don't find it curious at all that the 5770 and the 4870 have the same number of execution units with the 4870 has twice the bandwidth? I think my hypothesis is valid.

Don't get me wrong, the 5770 is a great card. Performance is there. Its just not the worthwhile upgrade I'm looking for at that price. That's all I've been trying to convey this whole time. If amd were to increase the bandwidth, you could have a potentially much more expensive card and I can respect that completely. Maybe I'm being too anal with my purchase decisions but $150 doesn't come easy for me and if I'm going to throw it down on something I'm going to play games with, I want to justify it well.
January 28, 2010 4:08:28 AM

ok guys.... why are you all talking about the 5770 and 4870 when the thread is about the 5830?
January 28, 2010 4:51:06 AM

chowmanga said:
Good scaling with crossfire doesn't tell me anything more than the fact that the 1gb frame buffer is sufficient for two 5770 processors. Regardless of rendering method (tiling, alternate frame, half screen), the same data is being loaded into both cards so if a crossfire setup is scaling well, then it's not limited by the amount of memory. Good crossfire scaling is not good enough to make a conclusion on memory bandwidth.

Also, memory overclocking potential and core overclocking potential don't have any correlation. If memory speed is a bottleneck, then overclocking the core isn't going to have much effect. Even though the memory can overclock higher, are you positive it will remove the bottleneck from the memory and put it on the core?


That the potential OC on the memory and core have no correlation was what I've been saying all along. You've been saying how the memory is going to limit the value you get out of OCing the core, despite that the memory is OCed at a significantly larger percentage. Let's take the first article you linked as an example. I guess you are saying that the memory is a bottleneck so it's the OC on the memory that is causing the 8% increase in performance while the 5% core increase may be essentially meaningless because the memory was and still is bottlenecking the core from achieving the performance it could. This is why I brought up crossfire scaling. There is no increase in memory bandwidth from crossfiring. It adds more computing power with the second core. If the memory was such a bottleneck that even an OC on one core would be limited by it then adding a whole second core should see almost no benefit also. This is not the case. The extra computing power offered by crossfire scales exceptionally well on this card which indicates that despite the somewhat limited memory subsystem it is still very sufficient for performance on a much higher level than a single HD5770 core. An OC is a relative mild increase in computational power in comparison so of course it isn't going to be limited either without even taking into account that you are actually increasing the memory bandwidth at a higher % than the core through an OC on this card.
January 28, 2010 4:08:18 PM

jyjjy said:
That the potential OC on the memory and core have no correlation was what I've been saying all along. You've been saying how the memory is going to limit the value you get out of OCing the core, despite that the memory is OCed at a significantly larger percentage. Let's take the first article you linked as an example. I guess you are saying that the memory is a bottleneck so it's the OC on the memory that is causing the 8% increase in performance while the 5% core increase may be essentially meaningless because the memory was and still is bottlenecking the core from achieving the performance it could. This is why I brought up crossfire scaling. There is no increase in memory bandwidth from crossfiring. It adds more computing power with the second core. If the memory was such a bottleneck that even an OC on one core would be limited by it then adding a whole second core should see almost no benefit also. This is not the case. The extra computing power offered by crossfire scales exceptionally well on this card which indicates that despite the somewhat limited memory subsystem it is still very sufficient for performance on a much higher level than a single HD5770 core. An OC is a relative mild increase in computational power in comparison so of course it isn't going to be limited either without even taking into account that you are actually increasing the memory bandwidth at a higher % than the core through an OC on this card.


Actually that article doesn't do much for my point because I misread the table. I'm not saying that the 8% is only due to the memory overclock because I don't know that. However I've yet to see an in depth article on the overclocking characteristics of the 5770 so I'm left to hypothesize.

Crossfire doesn't double the computational power of a single core, processing is done by two cores using the same data loaded into each of the card's ram. The cores execute at the same rate as they would alone, they just do different computations and combine the work done. When we reach the limit of the frame buffer, crossfire scaling gets worse because the cards will then need to make accesses to system ram which is much slower to do. If you run into the limit of the frame buffer, you can fix it by either adding more memory or making the memory faster. There is no reason that a card should require more bandwidth unless they reach the limit of the frame buffer.

http://www.legionhardware.com/document.php?id=862&p=6
Notice that scaling for far cry at 2560 is at 77% and for crysis at that resolution is 69%. Those games would require more (or faster) memory to scale better. Have you ever seen those reviews where the gtx295 bombs at that resolution? Its because of the frame buffer limitations. So again, no conclusions can be drawn about bandwidth with good crossfire scaling.
January 28, 2010 4:38:00 PM

The difference is tho, those who buy a 295 is more apt to own a 25x16 res than those who CF 5770s, and the 295 tanked alot in higher res, with AA usage.
Also, Id point out, ATIs memory management is better than nVidias, so you actually need more on a nVidia card at the same res to keep it from tanking than a ATI of similar perf
January 28, 2010 5:01:53 PM

chowmanga said:
There is no reason that a card should require more bandwidth unless they reach the limit of the frame buffer.

Everything you said I already knew and agree with other than that line and if it is true then what are you saying is the problem with the HD5770? Being limited by the size of the frame buffer and being limited by memory bandwidth are two different things. Higher performance makes a card need more memory bandwidth to keep the processor(s) supplied with the data it needs to function at a high level. If the card does not have enough memory bandwidth to keep up with a simple OC it definitely does not have enough to keep up with the extra computational power provided by crossfiring. Crossfiring does not increase the memory bandwidth as each card needs the same information in their respective memory to work in conjunction on calculations. Any fears you might have about a core OC of this card being limited by its memory bandwidth should be entirely dismissed based on its excellent scaling in crossfire.
That said 2 HD5770s in crossfire actually ARE limited somewhat by the memory bandwidth as is discussed in this article;
http://www.xbitlabs.com/articles/video/display/radeon-h...
and the result is that in certain games the minimum frame rate suffers as the GPUs aren't being supplied with enough data at certain points in the action to keep up with what the 2 gpus can process. So while the average frame rates remain quite high there are points where the limited bandwidth really is a factor. That said a simple 10-15% OC of a single HD5770 should not be something that causes this especially when the memory is also being OCed and to a larger degree.
January 28, 2010 5:03:05 PM

JAYDEEJOHN said:
The difference is tho, those who buy a 295 is more apt to own a 25x16 res than those who CF 5770s, and the 295 tanked alot in higher res, with AA usage.
Also, Id point out, ATIs memory management is better than nVidias, so you actually need more on a nVidia card at the same res to keep it from tanking than a ATI of similar perf


You're probably right - those setups at that resolution is far from the most common. I just chose those examples to try to explain that good crossfire scaling isn't very telling about a single card's memory bandwidth.
January 28, 2010 5:51:01 PM

jyjjy said:
Everything you said I already knew and agree with other than that line and if it is true then what are you saying is the problem with the HD5770? Being limited by the size of the frame buffer and being limited by memory bandwidth are two different things. Higher performance makes a card need more memory bandwidth to keep the processor(s) supplied with the data it needs to function at a high level. If the card does not have enough memory bandwidth to keep up with a simple OC it definitely does not have enough to keep up with the extra computational power provided by crossfiring. Crossfiring does not increase the memory bandwidth as each card needs the same information in their respective memory to work in conjunction on calculations. Any fears you might have about a core OC of this card being limited by its memory bandwidth should be entirely dismissed based on its excellent scaling in crossfire.
That said 2 HD5770s in crossfire actually ARE limited somewhat by the memory bandwidth as is discussed in this article;
http://www.xbitlabs.com/articles/video/display/radeon-h...
and the result is that in certain games the minimum frame rate suffers as the GPUs aren't being supplied with enough data at certain points in the action to keep up with what the 2 gpus can process. So while the average frame rates remain quite high there are points where the limited bandwidth really is a factor. That said a simple 10-15% OC of a single HD5770 should not be something that causes this especially when the memory is also being OCed and to a larger degree.


You're reading that line quite out of context. That entire paragraph is talking about crossfire.

It should have been read like this: "There is no reason that a [single] card should require more bandwidth unless [in crossfire] they reach the limit of the frame buffer [with respect to crossfire scaling]" I wasn't trying to say more memory bandwidth won't help the 5770. If you agree with the rest of that post, then you agree that if you reach the limit of the frame buffer, one can fix it by either adding memory or increasing memory bandwidth. That line is saying the same thing, just in a different way.

The extra computational power isn't correlated to utilized bandwidth in crossfire. Both cards have their own ram that works at the same speed. The core on each card works on the same data set, just doing different computations. This does not mean two 5770 in CF are using only 76.8gb/s bandwidth together. Each card is using their own ram, running at that speed. With your explanation, you're saying that the 5770 would perform the same with half the bandwidth and I sincerely doubt thats the case.

You probably already know that most games don't require more than 512mb vram with exceptions for games like crysis, far cry2, gta4. If both cards are using the same 1gb data, then each are using more or less than 512 mb (save for any memory management optimizations done). The good crossfire scaling tells me that the 5770 doesn't need 1gb when working by itself because two use it just fine except for extreme situations.

January 28, 2010 6:22:09 PM

we may have to wait even longer, 16th feb been mentioned now
January 28, 2010 7:21:57 PM

chowmanga said:
This does not mean two 5770 in CF are using only 76.8gb/s bandwidth together.

Actually that is exactly what it means. Crossfire does not increase the effective bandwidth of the setup. Yes the cards both have their own memory but the system is redundant. They both need to be fed the same data at the same rate to work effectively in conjunction. If this were not the case the issues with minimum frame rates in that article I linked would not arise as the crossfired HD5770s would have doubled memory bandwidth, equal to the HD5870, which as you can see does not have these issues.
January 28, 2010 8:26:55 PM

jyjjy said:
Actually that is exactly what it means. Crossfire does not increase the effective bandwidth of the setup. Yes the cards both have their own memory but the system is redundant. They both need to be fed the same data at the same rate to work effectively in conjunction. If this were not the case the issues with minimum frame rates in that article I linked would not arise as the crossfired HD5770s would have doubled memory bandwidth, equal to the HD5870, which as you can see does not have these issues.

January 29, 2010 12:45:22 AM

jyjjy said:
Actually that is exactly what it means. Crossfire does not increase the effective bandwidth of the setup. Yes the cards both have their own memory but the system is redundant. They both need to be fed the same data at the same rate to work effectively in conjunction. If this were not the case the issues with minimum frame rates in that article I linked would not arise as the crossfired HD5770s would have doubled memory bandwidth, equal to the HD5870, which as you can see does not have these issues.


So you're telling me that any card that scales in crossfire will see no benefit with additional bandwidth? I can't think of a single dual card setup that would provide no benefit over the single card. This implies no card has memory bandwidth issues and overclocking of memory won't provide any more performance. I'm pretty sure more is going on with crossfire than we can keep track of in order to isolate limitations to one specific component.
January 29, 2010 12:50:41 AM



Sorry, jaydeejohn didn't seem to mind so I thought it was okay. Maybe he thought the conversation was worthwhile? Is there anything you can do so that we can keep the topic going? I think we're almost at the bottom of it.
January 29, 2010 3:07:19 AM

chowmanga said:
So you're telling me that any card that scales in crossfire will see no benefit with additional bandwidth? I can't think of a single dual card setup that would provide no benefit over the single card. This implies no card has memory bandwidth issues and overclocking of memory won't provide any more performance. I'm pretty sure more is going on with crossfire than we can keep track of in order to isolate limitations to one specific component.


The overclocking of memory will help alleviate a memory bandwidth neck, JYJJY is saying CrossFire does not double the memory bandwidth, which is true as both 5770's store the required info in their own respective VRAM.
Dual GPU cards will not show any boost (and sometimes have a negative effect) if the game does not support CF/SLI.
January 29, 2010 9:08:54 AM

chowmanga said:
So you're telling me that any card that scales in crossfire will see no benefit with additional bandwidth?

No, what I'm saying is that a card that scales well in crossfire obviously isn't so limited by its memory bandwidth that a core OC is of questionable value, as you were stating about the HD5770.
January 29, 2010 9:48:38 AM

The biggest news by ATI to IT community is not the new catalyst v10.1 drivers, but a second postponement of 5830 by ATI. The unofficial reason for a 5830 that was initially slated for 25 January is PCB issues :-D.
That is so lame excuse :-D, the 5850 and 5830 are same card, same PCB, cooling, 256bit BUS, DDR5, same GPU just lover clocks of GPU core and memory, the biggest difference is the sticker :-D = 5830 is binned 5850. The 5830 cost AMD same as 5850 so they decided to postpone 5830 as long as possible at this stage for 3-4 weeks to make more money for ATI- AMD and its partners, 5850 earns higher margins for whole ATI distribution channel. 5830 delayed from 25 January to 5th
of February and at this stage around 16th February This is a result of lack of competition, NVIDIA hurry up with FERMI ;-D.

P.S 5830 specs:
Same PCB, cooling and connectivity as 5850, 625 Mhz GPU clock, 800 Mhz DDR5 memory, 1280 shaders, that are the slated specs for 5830 by ATI, when it finally becomes available it will be the best cost - value - performance single video card and crossfire solution ;-D. The difference ratio is the same as between 5870 and 5850 ;-D
!