ATi's counter: overclocked 5870, and "5990"

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

thilanliyan

Lifer
Jun 21, 2005
11,871
2,076
126
Well yeah, but then realize Fermi has a much bigger die (which means heat distribution at x watts will be better than what would happen with a smaller die).

So 300 watts overclocked for 529mm2 Fermi is going to result in lower chip temperatures than 300 watts for Cypress on a 333 m2 die (assuming heat sinks are roughly equivalent). Correct me if I am wrong.

Even if the chip temperatures are lower, you still have to get rid of that 300w heat otherwise it will build up and have runaway temperatures so having a bigger die doesn't really mean much. As long as it is under a threshold temperature (100-120C for video cards these days), the ability of the heatsink and your case to move the heat away from the GPU is what matters (in terms of cooling).
 
Feb 19, 2009
10,457
10
76
Well yeah, but then realize Fermi has a much bigger die (which means heat distribution at x watts will be better than what would happen with a smaller die).

So 300 watts overclocked for 529mm2 Fermi is going to result in lower chip temperatures than 300 watts for Cypress on a 333 m2 die (assuming heat sinks are roughly equivalent). Correct me if I am wrong.

The GPUs themselves dissipate thermals fine, even small ones. Conductivity is quite high and use of copper/good thermal compound it isnt a problem. What is however, is the overall dissipation your heatsink/fan has to do. More heat = require bulkier or more expensive heatsink/fans.

Also consider power loads. The 470 is rated at 225W, with 2x6 pins, thats already at max. Very little room for OC before you run into instability.

Obviously ATI does not feel threatened by Fermi at all. So i bet they know a thing or two about its performance. Releasing OC models isn't a counter at all, its what AIB makers have been doing for ages to compete with each other. I mean there's hardly a need, because any reference 5850 or 5870 is going to OC like crazy. Getting to 950mhz core and 5000 mhz ram speed is do-able with a slight voltage mod, for both cards and they still run quiet.

We will all know soon enough really, only a few more days to go.
 

toyota

Lifer
Apr 15, 2001
12,957
1
0
The GPUs themselves dissipate thermals fine, even small ones. Conductivity is quite high and use of copper/good thermal compound it isnt a problem. What is however, is the overall dissipation your heatsink/fan has to do. More heat = require bulkier or more expensive heatsink/fans.

Also consider power loads. The 470 is rated at 225W, with 2x6 pins, thats already at max. Very little room for OC before you run into instability.

Obviously ATI does not feel threatened by Fermi at all. So i bet they know a thing or two about its performance. Releasing OC models isn't a counter at all, its what AIB makers have been doing for ages to compete with each other. I mean there's hardly a need, because any reference 5850 or 5870 is going to OC like crazy. Getting to 950mhz core and 5000 mhz ram speed is do-able with a slight voltage mod, for both cards and they still run quiet.

We will all know soon enough really, only a few more days to go.
that 225 watts takes into account some overclocking and is not close to the actual power usage for a stock card. if it has a TDP of 225watts then it probably wont go much over 150 during gaming. even something like furmark still would not likely push it over 200watts even if it was overclocked a little. not to mention we dont officially know if the gtx470 is rated for 225watts TDP.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
that 225 watts takes into account some overclocking and is not close to the actual power usage for a stock card. if it has a TDP of 225watts then it probably wont go much over 150 during gaming. even something like furmark still would not likely push it over 200watts even if it was overclocked a little. not to mention we dont officially know if the gtx470 is rated for 225watts TDP.


I hope the 470 GTX only needs 2x6 pin.
 
Last edited:

cbn

Lifer
Mar 27, 2009
12,968
221
106
Even if the chip temperatures are lower, you still have to get rid of that 300w heat otherwise it will build up and have runaway temperatures so having a bigger die doesn't really mean much.

Thank you for correcting me.

If the fan and heatsink become the bottleneck....then we could expect chip temperatures to rise.
 

Red Storm

Lifer
Oct 2, 2005
14,233
234
106
ATI responding with just a small overclock on the 5870 definitely seems to support the rumor that Fermi may only be 5-10% faster. When performance is that close, it all comes down to price and power figures.
 

Plimogz

Senior member
Oct 3, 2009
678
0
71
ATI responding with just a small overclock on the 5870 definitely seems to support the rumor that Fermi may only be 5-10% faster. When performance is that close, it all comes down to price and power figures.

And features.
 

legcramp

Golden Member
May 31, 2005
1,671
113
116
Bring back the X800 XT crap. Hell they could call it the Radeon HD X5870 XT Turbo plantinum edition deluxe. That name alone would gurantee a few thousand sales to fanboys right there.

You mean like the 6800Ultra? bahaha I should know, I had one on launch date.
 

Daedalus685

Golden Member
Nov 12, 2009
1,386
1
0
In the reviews of the 5850 toxic a few reviewers mentioned that sapphire had been told form on high that no over clocked cards can be released outside of the very tight speeds capped in the stock BIOS. For the 5850 this castrated the ability to overclock as a vendor substantially, for a 5870 I think it was something shy of 900mhz max.

All this could mean is that they will allow AIBs to release with a newly designed bios at over 900mhz. There may be no need to release anew product number at all if that is the case, we will just start seeing the ROG edition ASUS card, and a toxic sapphire.


On a related note, semi-accurate forums have some rumours floating around that ATI had taped something out on TSMC's 32nm before it was cancelled. There are rumblings that that was going to be the 5800 refresh but will not have to wait/not see the ligth of day. The same rumour is reporting the delay could mean complications for NI and the 6000 series (too far off to be sure, I have not logically expected it until TSMC's 28nm is ready anyway).
 

dzoner

Banned
Feb 21, 2010
114
0
0
1. Since these are A3 chips, they are still silicon layer challenged and their overclocking power/performance ratio will be considerably worse than the 58xx ships. That whole AMD did their 40nm homework, Nvidia ate theirs thing ... like in Anands article. The negative repercussions to Nvidia of this haven't gone anywhere.

2. And still homework related, Nvidia's yields are far lower than AMD's and that won't change until B silicon comes on line ... if Nvidia doesn't just move on to Fermi II. There's certainly a possibility if yields are still very low, Nvidia doesn't order up any more A3 wafers and just markets what they have instead of trying the 'selling at a loss but making up for it in volume' strategy.

AMD knows Nvidia has squat and even the 'official' overclock is almost window dressing.
 
Last edited:

cbn

Lifer
Mar 27, 2009
12,968
221
106
1. Since these are A3 chips, they are still silicon layer challenged and their overclocking power/performance ratio will be considerably worse than the 58xx ships.

Can you give us a link or explanation on how A,B and C steppings factor into how chips work or overclock? This sounds really interesting
 

Blue Shift

Senior member
Feb 13, 2010
272
0
76
Does that mean my XFX XXX 5870s are already uh "5990"s?

No, it means that if you smash your pair of 5870s together with all the force you can muster, then there's a 1x10^-500 chance that a "5990" could be born.
 

SlowSpyder

Lifer
Jan 12, 2005
17,305
1,001
126
Do they not realize that you can probably overclock the 480s as well?

The link says that they feel 875-900MHz is all that's needed to keep pace, not that 875-900MHz is all they'd come out with. They could have been cherry picking cores for months now, or even pulling a 4890 where they have a physically different chip in the works. It's pretty early yet to even attempt an educated guess.
 

SolMiester

Diamond Member
Dec 19, 2004
5,331
17
76
Well yeah, but then realize Fermi has a much bigger die (which means heat distribution at x watts will be better than what would happen with a smaller die).

So 300 watts overclocked for 529mm2 Fermi is going to result in lower chip temperatures than 300 watts for Cypress on a 333 m2 die (assuming heat sinks are roughly equivalent). Correct me if I am wrong.

However, Fermi doesnt need to run as fast as Cypress for performance and therefore wont need such a high head room for clocking......a refresh will be even better!
 

at80eighty

Senior member
Jun 28, 2004
458
3
81
I want to know where the 5970's are at right now.

Im moving to singapore soon & word from a few is there are places that sell the Sapphire kind at the equivalent of USD655, give or take.

at that price I'm really tempted though its a wee bit out of my budget presently. wonder if I can CF it with any of the later 6xxx cards down the line - if so i'd surely grab one
 

SlowSpyder

Lifer
Jan 12, 2005
17,305
1,001
126
However, Fermi doesnt need to run as fast as Cypress for performance and therefore wont need such a high head room for clocking......a refresh will be even better!

Right, but if Fermi can overclock 10% while the 5870 can overclock 25% then AMD probably has the faster part. But, right now Fermi overclocking is completely unknown. Will it struggle to gain an additional 15MHz or will 800MHz be the norm?
 

yh125d

Diamond Member
Dec 23, 2006
6,907
0
76
I still hold hope for a 5890. 950/1250+ clocks, burly power circuitry, etc...


Article only said that this isn't a 5890, not that there won't be one ;)