werezHD 5870 1 GHz sounds pretty tempting . The card's price however will be a little to high for my taste , and it`s good to know that those clocks are achievable with some minor voltage bump . However if i had the money i would rather look at the HD 5870 Lightning Edition from MSI , looks a little bit better . But than again Gigabyte's version will end up cheaper . OH decisions decisions ... :)
Way to go Gigabyte !

And it has a shorter custom PCB.

: ]

btarunrThe fact that there's 1 GHz HD 5870s shows that there won't be an HD 5890.

Most probably, there won't be a RV870-based HD5890.
But there will be an updated, higher-end, HD5000 GPU somewhere in 2010 (probably after Fermi's launch).
That's been pretty much confirmed by Richard Huddy in his last bittech interview.

Nonetheless, the RV870 doesn't need higher GPU clocks, it desperately needs higher memory bandwidth - be it with a redesigned memory bus or substantially faster GDDR5 memory.

ToTTenTranzMost probably, there won't be a RV870-based HD5890.
But there will be an updated, higher-end, HD5000 GPU somewhere in 2010 (probably after Fermi's launch).
That's been pretty much confirmed by Richard Huddy in his last bittech interview.

Nonetheless, the RV870 doesn't need higher GPU clocks, it desperately needs higher memory bandwidth - be it with a redesigned memory bus or substantially faster GDDR5 memory.

Can you share some links about the memory affecting performance?

thus far I've not seen memory over clocks make much of a difference so I#ve not seen these bottles neck yet.

pantherx12Can you share some links about the memory affecting performance?

thus far I've not seen memory over clocks make much of a difference so I#ve not seen these bottles neck yet.

ixbtlabs.com/articles3/video/cypress-p12.html

Digit-LifeMoreover, the 48xx had very high memory bandwidth for the cores, and its insignificant increase in the 58xx only balances GPU and local video memory. As we can see, the core has grown by almost two times, while memory bandwidth has grown only by 25%. However, we don't see twofold performance gains versus the 4890. Perhaps, performance is limited by something in high resolutions. Probably by memory bandwidth.

Memory bandwidth is the only thing that didn't double from HD4890, where the performance difference is between 25 and 75% (average 40%).

im the only one who noticed that the introduction card has some bad grammar:laugh:

ToTTenTranzThat's been pretty much confirmed by Richard Huddy in his last bittech interview.

Can you link that article, I don't recognise the name Richard Huddy and I've been reading Bit-Tech for years, d'you mean Richard Swinburne?

TVmanim the only one who noticed that the introduction card has some bad grammar:laugh:

I don't think Andrey Vorobiev's first language is English tbh. Or d'you mean the 5870 has bad grammar, I'm confused.

ToTTenTranzNonetheless, the RV870 doesn't need higher GPU clocks, it desperately needs higher memory bandwidth - be it with a redesigned memory bus or substantially faster GDDR5 memory.

Why is it then, when you increase the memory bandwith on a 5870 there is minimal performance gain and when you overclock the GPU there is good performance gain? Surely if there was a memory bandwith bottleneck overclocking the GPU would result in no gain.

I want this card :)

I dont know if they have any sinks on the ram or the vrm area but it looks like maybe the cooling solution covers that.

InnocentCriminalCan you link that article, I don't recognise the name Richard Huddy and I've been reading Bit-Tech for years, d'you mean Richard Swinburne?

Richard Huddy isn't the interviewer, he's AMD’s Worldwide Developer Relations manager.

www.bit-tech.net/bits/interviews/2010/01/06/interview-amd-on-game-development-and-dx11/1


It's a great read, the best hardware interview I've seen since this one. The guy is completely honest and he even takes all the blame about what happened with the Saboteur. It's not a "let me evade your questions and do PR crap" session (a la Jen Hsu Huang), it's an actual interview.


About the other part:

Richard Huddy]Let me say that I suspect that through 2010 it's possible there will be moments where we don't have the fastest graphics cardWhy is it then, when you increase the memory bandwith on a 5870 there is minimal performance gain and when you overclock the GPU there is good performance gain? Surely if there was a memory bandwith bottleneck overclocking the GPU would result in no gain.

Maybe because those tests were made in low resolutions or detail?

Memory bottleneck is the only logical explanation for the HD5870 to be slower than a HD4870X2.

Meh.

[H]@RD5TUFFMeh.

Thanks for your brilliant contribution to this thread. :shadedshu

@ ToTTenTranz: I agree with you that memory bottleneck is one of the main reasons that the 5870 isn't a double in performance over the 4890. However, I think it also has to do with the fact that games might not fully utilize the number of SPUs on that single core. This could be because of the drivers, or just the fact that games developers need to play catchup with the tech market.

This 5870 OC Edition will be nice, but I still do expect a 5890, which will most likely solve the RAM bottleneck issue with a 384 bit bus (which, incidentally, will mean another 16 ROPs (total of 48 :D)), and higher clocked RAM. This would require a new GPU core, which might come with a die shrink, optimized power systems, or both, allowing for higher clocks. This is what I'm holding out for. ;)

I would love to have two of these water cooled:)

ToTTenTranzRichard Huddy isn't the interviewer, he's AMD’s Worldwide Developer Relations manager.

www.bit-tech.net/bits/interviews/2010/01/06/interview-amd-on-game-development-and-dx11/1


It's a great read, the best hardware interview I've seen since this one. The guy is completely honest and he even takes all the blame about what happened with the Saboteur. It's not a "let me evade your questions and do PR crap" session (a la Jen Hsu Huang), it's an actual interview..

That's right, I remember trying to read that but my girlfriend kept talking to me and nagging me to let her on to Facebook so I didn't read it all. I'll have to give it a proper read when I'm on my lunch today.

Meanwhile ATi video cards price skyrocket :shadedshu , where I live the 5850's current price matches 5870's price at launch day and the 5870 itself is averaging around 525 USD.

Error 404Thanks for your brilliant contribution to this thread. :shadedshu

@ ToTTenTranz:... However, I think it also has to do with the fact that games might not fully utilize the number of SPUs on that single core. This could be because of the drivers, or just the fact that games developers need to play catchup with the tech market...

A big +1. It's all about coding if you ask me on both sides. Obviously Ati has to invest more into drivers and probably educating game devs on how to utilize the full potential of their GPUs. Or maybe they can continue doing what they've been doing so far - leaning on Nvidia's back hoping they can utilize all the magic that is done in TWINTBP at a later point.

ToTTenTranzRichard Huddy isn't the interviewer, he's AMD’s Worldwide Developer Relations manager.

www.bit-tech.net/bits/interviews/2010/01/06/interview-amd-on-game-development-and-dx11/1


It's a great read, the best hardware interview I've seen since this one. The guy is completely honest and he even takes all the blame about what happened with the Saboteur. It's not a "let me evade your questions and do PR crap" session (a la Jen Hsu Huang), it's an actual interview.


About the other part:


That's a confirmation that AMD has a Fermi counter in its roadmap, and it will launch at least before mid-2010 - which means it's most likely a 5000 series card.




Maybe because those tests were made in low resolutions or detail?

Memory bottleneck is the only logical explanation for the HD5870 to be slower than a HD4870X2.

No, that doesn't confirm anything. One can make such statements to be evasive too. AMD doesn't make two GPU architectures in an year, especially not with a mere 6 months' gap. RV7xx's were around for over an year, so were the RV6xx's, and so on. With the amount of investment it takes to make that "something better than Cypress", you can expect it only next year.

btarunrAMD doesn't make two GPU architectures in an year, especially not with a mere 6 months' gap. RV7xx's were around for over an year, so were the RV6xx's, and so on. With the amount of investment it takes to make that "something better than Cypress", you can expect it only next year.

RV790 came out less than a year after RV770.

RV890 wouldn't need to be a dramatically new core, it could just be tweaker in order to support higher clocks and be bundled with (a lot) faster GDDR5.

Of course, a 384-bit bus would be welcome, but that would require a new PCB, so it's unlikely to happen.

Theres no memory bottleneck

Why?

overclock the mem you see basically no improvement

if you oc the core you get massive boosts.

In theory that means no memory bottleneck anyways....

ToTTenTranzRV790 came out less than a year after RV770.

RV890 wouldn't need to be a dramatically new core, it could just be tweaker in order to support higher clocks and be bundled with (a lot) faster GDDR5.

Of course, a 384-bit bus would be welcome, but that would require a new PCB, so it's unlikely to happen.

And the only reason RV790 came out was because of RV770's clock speed and electrical limitations (with which AMD couldn't even make OC models to speak of), as I argued earlier. RV790 can't be used as an example, aside from the fact that it couldn't quite serve its purpose against GTX 285 and NVIDIA's response (GTX 275).

Don't equate RV790's creation to something similar with Cypress. It doesn't face the issues RV770 faced. To address the memory bandwidth "deficiency" issue, the GPU should be given a complete redesign due to the way the different components of the GPU are arranged. That's a substantially higher R&D than what went into making RV790. The bottomline is AMD can't afford to make performance GPUs every half year. It makes a performance GPU every >1 year. It took 15 months to get from RV770 to Cypress.

So AMD won't roll out two performance GPUs within a span of an year.

For the last time, either get a 5870, and actually test memory bandwidth for yourself, read and comprehend reviews, or just die. The 5870 setup engine is starving the core, plus the redesign of the core for using the extra shaders to process the information generated by the tesselation unit doesn't require mega bandwidth.



After so many charts were made, tests by users who actualy owned a 5870, reviews by W1zz, and other sites if you think that it needs more bandwidth than you need to end yourself.

kylzerTheres no memory bottleneck

Why?

overclock the mem you see basically no improvement

if you oc the core you get massive boosts.

In theory that means no memory bottleneck anyways....

SteevoFor the last time, either get a 5870, and actually test memory bandwidth for yourself, read and comprehend reviews, or just die. The 5870 setup engine is starving the core, plus the redesign of the core for using the extra shaders to process the information generated by the tesselation unit doesn't require mega bandwidth.


After so many charts were made, tests by users who actualy owned a 5870, reviews by W1zz, and other sites if you think that it needs more bandwidth than you need to end yourself.

Memory overclocking in those tests isn't high enough to make those tests relevant, IMO.

And, @ Steevo, there's really no reason to tell others to "just die" and "you need to end yourself" when you disagree with them. That's childish.

btarunrAnd the only reason RV790 came out was because of RV770's clock speed and electrical limitations (with which AMD couldn't even make OC models to speak of), as I argued earlier. RV790 can't be used as an example, aside from the fact that it couldn't quite serve its purpose against GTX 285 and NVIDIA's response (GTX 275).
(...)
Don't equate RV790's creation to something similar with Cypress. It doesn't face the issues RV770 faced. The bottomline is AMD can't afford to make performance GPUs every half year. It makes a performance GPU every >1 year. It took 15 months to get from RV770 to Cypress.

So AMD won't roll out two performance GPUs within a span of an year.

The thing is, you (or me) don't know if there are enough improvements to be made to RV870 to justify a new chip. There's always something to improve, and these improvements wouldn't be the exact same thing as what happened in the RV770->RV790 transition (that would be stupid).

Notice that I'm not insisting that there will be a new chip, I'm just saying that nothing is certain at this point.
Richard Huddy did say they will have an unanounced product prepared to make sure that Fermi won't be the fastest graphics card throughout most of 2010 (like what happened with the GTX285 in 2009). Since Fermi is coming in late Q1, they're sure to launch this product before Q4.

btarunrTo address the memory bandwidth "deficiency" issue, the GPU should be given a complete redesign due to the way the different components of the GPU are arranged. That's a substantially higher R&D than what went into making RV790.

They don't, not at all.
All AMD has to do is to pair a Cypress board with the new 7Gb/s memory chips being produced by Samsung and Hynix. It would bump the bandwidth to 224GB/s (45% up from the current 154GB/s that we see in the HD5870).


A Cypress card with 1GHz core and 224GB/s bandwidth could perfectly be the future HD5890 (and the card to compete with a Fermi flagship).

Tests were completed with UNDERCLOCKED memory, and at every point of memory clock the relationship it had as a percentage of gain or loos was significantly less related than the gain or loss created by the changing of core clock.

Is that better? Here is a benchmark run from a actual user, back to back changing only the memory speed.

I've tested in Unigine Heaven, SF4 benchmark and DMC4 benchmark, core always at 950mhz, and memory speed varying from 1000-1300.

Heaven 1680x1050 - 4xAA - 16xAF MAX settings

950/1000 = 31.7fps - score 799
950/1100 = 32.6fps - score 822
950/1200 = 33.6fps - score 845
950/1300 = 34.4fps - score 865

1920x1200 4xAA - 16xAF, MAX - Posterization

950/1000 - 134.64
950/1100 - 139.37
950/1200 - 142.01
950/1300 - 145.44

DMC4 I had to do average of 3 runs across the 4 scenes and experienced some issues, so i wont clog up my post with those results, suffice to say it followed these results quite linearly.

my testing shows;

30% difference in memory bandwidth across the 256 bit bus results in a performance difference of ~9%

I dare make the assumption that even if ATi paired this card with 6.4gbs memory instead of 4.8, we'd see performance of around 10%, given my testing between 4gbs to 5.2gbs

I'd love to speculate how the card would perform with a 512 bit bus, but I honestly don't think I could do it justice. But I really think the choice for 4.8gbs memory was based on how cheap and abundant the memory chips are compared to faster clocked stuff, and the fact that performance on this GPU seems to have little to gain from the speed alone.


forums.techpowerup.com/attachment.php?attachmentid=30510&d=1257887705

A simple chart to show the relatinoship of memory speed to gain.

Blue is the memory speed increase, red is the actual performance increase, yellow is the net loss of effect that people are proposing. So for evey 10% increase your 3% of win is overbalanced by 7% of fail, netting your over 200% more fail per clock.

ToTTenTranzMemory overclocking in those tests isn't high enough to make those tests relevant, IMO.

What tests?

I'm talking from my personal experiences with the HD58 and 57 series.

ToTTenTranzThe thing is, you (or me) don't know if there are enough improvements to be made to RV870 to justify a new chip. There's always something to improve, and these improvements wouldn't be the exact same thing as what happened in the RV770->RV790 transition (that would be stupid).

Right, and therefore neither of us can confirm anything based on vague statements someone made to the press.

My contention is that AMD won't release a new performance GPU this soon (within 12 months of Cypress' release), simply because it can't afford to. It can't do to Cypress what it did to RV770 (a change it could afford and desperately needed), because Cypress already has the resources to run at high clock speeds.

ToTTenTranzThey don't, not at all.
All AMD has to do is to pair a Cypress board with the new 7Gb/s memory chips being produced by Samsung and Hynix. It would bump the bandwidth to 224GB/s (45% up from the current 154GB/s that we see in the HD5870).


A Cypress card with 1GHz core and 224GB/s bandwidth could perfectly be the future HD5890 (and the card to compete with a Fermi flagship).

So you've answered yourself. AMD won't have to come up with a new GPU.

The fact that there's 1 GHz HD 5870s shows that there won't be an HD 5890.

I don't agree with this statement. Look at the 4870 and the 4890. Companies like sapphire came out with the 4870 Toxic and a 4890 Toxic.

If you ask me this doesn't say that. What they could do is go in a rework the 5870 give it a 1000MHZ clock speed and a 384Bit Bus Speed and call that a 5890. That would easily out class a 5870.

Plus if they went in a reworked the GPU you could actually see a 5890 with a 384 bit bus sure but maybe a core clock of 1200MHZ. Not saying this would happen but look how much more power they got by reworking the 4870 into a 4890 about 20% to 25% more power. I think they could still easily do that.

Although it probably really depends on what Femi does and how fast that is.