ATi Radeon 4890 Discussion

[error8]

4890

So it seems to be an overclocked 4870, with 850 mhz on the gpu and 975 mhz on the GDDR5. It has 2X6pin power connectors and a slightly different cooler.

EDIT:corrected

 

[TC91]

meh

 

[sawtx03]

Originally posted by: error8
4890

So it seems to be an overclocked 4870, with 850 mhz on the gpu and 975 mhz on the GDDR5. It has one 6pin+ one 8pin power connectors and a slightly different cooler. So, the gpu voltage has been increased and we're again approaching 2900XT power consumption.

That card has 2 actual 6-pin connectors (the link even says dual 6-pins), not 1 6-pin and 1 8-pin. The board looks like it can switch out one of the 6-pin connectors for an 8-pin but it looks like that is probably will be for a higher speed part.

 

[error8]

Originally posted by: sawtx03

Originally posted by: error8
4890

So it seems to be an overclocked 4870, with 850 mhz on the gpu and 975 mhz on the GDDR5. It has one 6pin+ one 8pin power connectors and a slightly different cooler. So, the gpu voltage has been increased and we're again approaching 2900XT power consumption.

That card has 2 actual 6-pin connectors (the link even says dual 6-pins), not 1 6-pin and 1 8-pin. The board looks like it can switch out one of the 6-pin connectors for an 8-pin but it looks like that is probably will be for a higher speed part.

Yes, you're right. I was in a hurry reading that. There are only 2X6pin connectors and for "future" there is place to install an 8 pin. So maybe power consumption is not that over the edge after all.

 

[evolucion8]

Originally posted by: Azn

4850 is not far off 4870. 4870 is 20% faster than 4850 but than again the core is clocked much higher. 625 vs 750.

As for 4830 and 4850. It's about 15% difference. Then again clock difference is merely 50mhz. Performance discrepancy comes from shader and tmu units but I've heard RV770 is really 32TMU card and later added 8TMU. Something about those 8TMU not really doing anything.

RV770 with only 16ROP and 40TMU it's pretty much efficient with 256bit bus. Of course the extra bandwidth does play some role in ROP and AA performance with 4870.

Overclock a 4850 to 4870 core clocks and you would easily find all the bandwidth is really doing. I suspect not much.

Usually the HD 4850 leads the HD 4830 between 6fps and 12fps, but the performance gap between the HD 4850 and HD 4870 is larger than that, it can be as much as 30fps, specially when anti aliasing is used. So I don't know where did you pull that info about the TMU's, they scale linearly with the stream processors amount, and they are all available for those cards except with the HD 4830. And like I said before, the performance lead of the HD 4870 over the HD 4850 is much wider than the performance lead of the HD 4850 over the HD 4830 and I really find it doubtfull that is relative only to the core speed alone.

http://www.legitreviews.com/article/811/4/

http://www.techpowerup.com/reviews/ATI/HD_4830/25.html

http://www.hardwaresecrets.com/article/641/5

http://www.overclockersclub.com/reviews/pc4830/7.htm

 

[Qbah]

Well, that's a disappointment, if it's true At least before, the X850XT and X1950XTX were top cards of their generation (while the X800 and X1800 were competitive). So that "refresh" and improved process gave ATi something. nVidia will still have the fastest single GPU and multi GPU card. I really don't see the point of this card, especially if it will be at 299$ and the GTX275$ at 249$. The HD4870 and HD4850 are more than competitive at their price points. This card won't challenge the GTX285 with those clocks.

Seriously, what's the point?

 

[Jacen]

I am pretty sure that won't be the price point. Have some patience.

 

[error8]

Originally posted by: Qbah
This card won't challenge the GTX285 with those clocks.

Seriously, what's the point?

That is true. I also thought that ATi will make a challenging card for GTX 285. But it's just slightly faster then 4870. I don't see the benefits of having this card on the market.

 

[thilanliyan]

I also don't really see much point in this card...unless their yields are so good that they can run ALL 4870s at those speeds. I seriously hope it's a new card with added shaders...would be boring if it isn't.

 

[mhouck]

Now FUD says R790 has more efficient shaders

So the rumors continue

 

[Hauk]

Originally posted by: mhouck
Now FUD says R790 has more efficient shaders

So the rumors continue

Saw that article today. Here's what was most intriguing:

"We are not sure what ATI did to its shaders, but this is the reason why RV790 chips are actually differ from the RV770, and that the pin count is not the same."

 

[Cookie Monster]

Well the pin count could be because it could be using a different substrate, or maybe due to the PCB changes etc. The 950MHz version sounds more interesting, mainly because it has the potential to compete with the GTX285 while being cheaper.

 

[AzN]

Originally posted by: evolucion8

Originally posted by: Azn

4850 is not far off 4870. 4870 is 20% faster than 4850 but than again the core is clocked much higher. 625 vs 750.

As for 4830 and 4850. It's about 15% difference. Then again clock difference is merely 50mhz. Performance discrepancy comes from shader and tmu units but I've heard RV770 is really 32TMU card and later added 8TMU. Something about those 8TMU not really doing anything.

RV770 with only 16ROP and 40TMU it's pretty much efficient with 256bit bus. Of course the extra bandwidth does play some role in ROP and AA performance with 4870.

Overclock a 4850 to 4870 core clocks and you would easily find all the bandwidth is really doing. I suspect not much.

Usually the HD 4850 leads the HD 4830 between 6fps and 12fps, but the performance gap between the HD 4850 and HD 4870 is larger than that, it can be as much as 30fps, specially when anti aliasing is used. So I don't know where did you pull that info about the TMU's, they scale linearly with the stream processors amount, and they are all available for those cards except with the HD 4830. And like I said before, the performance lead of the HD 4870 over the HD 4850 is much wider than the performance lead of the HD 4850 over the HD 4830 and I really find it doubtfull that is relative only to the core speed alone.

http://www.legitreviews.com/article/811/4/

http://www.techpowerup.com/reviews/ATI/HD_4830/25.html

http://www.hardwaresecrets.com/article/641/5

http://www.overclockersclub.com/reviews/pc4830/7.htm

I don't know where you get 4870 beating 4850 by 30fps from but 4870 is roughly 20% faster than the 4850 and 4850 is 10-15% faster than 4830. That's with 2x the bandwidth and 125mhz to the core. Overclock the 4850 to 750 and you've got less than 10% performance difference.

As for RV770 later adding 8 more TMU and SP was included in some RV770 launch articles.

I think sschevy2001 linked here showing the differences between a 4850 and 4870 when you overclock the 4850 to 4870 core clocks.

http://www.xbitlabs.com/articl...024mb-gs_11.html#sect1

So basically all that bandwidth on 4870 is sitting mostly idle. It helps some sure but not enough to say RV770 is bandwidth limited with gddr3.

 

[MyLeftNut]

Originally posted by: Azn

Originally posted by: evolucion8

Originally posted by: Azn

4850 is not far off 4870. 4870 is 20% faster than 4850 but than again the core is clocked much higher. 625 vs 750.

As for 4830 and 4850. It's about 15% difference. Then again clock difference is merely 50mhz. Performance discrepancy comes from shader and tmu units but I've heard RV770 is really 32TMU card and later added 8TMU. Something about those 8TMU not really doing anything.

RV770 with only 16ROP and 40TMU it's pretty much efficient with 256bit bus. Of course the extra bandwidth does play some role in ROP and AA performance with 4870.

Overclock a 4850 to 4870 core clocks and you would easily find all the bandwidth is really doing. I suspect not much.

Usually the HD 4850 leads the HD 4830 between 6fps and 12fps, but the performance gap between the HD 4850 and HD 4870 is larger than that, it can be as much as 30fps, specially when anti aliasing is used. So I don't know where did you pull that info about the TMU's, they scale linearly with the stream processors amount, and they are all available for those cards except with the HD 4830. And like I said before, the performance lead of the HD 4870 over the HD 4850 is much wider than the performance lead of the HD 4850 over the HD 4830 and I really find it doubtfull that is relative only to the core speed alone.

http://www.legitreviews.com/article/811/4/

http://www.techpowerup.com/reviews/ATI/HD_4830/25.html

http://www.hardwaresecrets.com/article/641/5

http://www.overclockersclub.com/reviews/pc4830/7.htm

I don't know where you get 4870 beating 4850 by 30fps from but 4870 is roughly 20% faster than the 4850 and 4850 is 10-15% faster than 4830. That's with 2x the bandwidth and 125mhz to the core. Overclock the 4850 to 750 and you've got less than 10% performance difference.

As for RV770 later adding 8 more TMU and SP was included in some RV770 launch articles.

I think sschevy2001 linked here showing the differences between a 4850 and 4870 when you overclock the 4850 to 4870 core clocks.

http://www.xbitlabs.com/articl...024mb-gs_11.html#sect1

So basically all that bandwidth on 4870 is sitting mostly idle. It helps some sure but not enough to say RV770 is bandwidth limited with gddr3.

+1
Someone over at XS did a test between 4850 and 4870 clock for clock and performance differences was 10% or less most of the time.

 

[evolucion8]

you guys don't understand my point. My point is that is quite strange that there's so much little performance difference between the HD 4830 and the HD 4850 considering than the former has less stream processors and a slighly lower core, and the other point is, if the HD 4870 and HD 4850 has the same stream processors and besides the core speed difference, why the performance difference between them is more pronounced than the HD 4830 vs GHD 4850? What it comes to my mind is that if the stream processors are disabled, and the performace drop is very low, is because there's another bottleneck somewhere, it could be the TMU's? it could be anything. But definitively there's a small advantage of the GDDR5 usage on the HD 4870, but probably the same performance level could be available using high speed GDDR3 running over 2.4GHz.

 

[AzN]

The link provides there isn't a pronounced difference from 4850 and 4870 when you overclock the core on 4850. 10% mostly.

when you consider AA benchmarks it's mostly dictated by bandwidth and ROP no wonder that's where you see the performance difference between 4830 4850 and 4870 when you consider memory controller is tied down to ROP.

I've seen benchmarks where 9800 gtx beats 4870 without AA but with AA 4870 beats the living crap out it.

 

[thilanliyan]

Some more info:
http://www.dailytech.com/article.aspx?newsid=14616

 

[josh6079]

Originally posted by: thilan29
Some more info:
http://www.dailytech.com/article.aspx?newsid=14616

Interesting:

DailyTech
The chip is not just an overclocked RV770, as there are a few tweaks and modifications that we are still investigating.

 

[Comdrpopnfresh]

so if someone has a 4870 1gb, and can overclock it to 850mhz, is that equivalent?

 

[error8]

Originally posted by: Comdrpopnfresh
so if someone has a 4870 1gb, and can overclock it to 850mhz, is that equivalent?

Probably. He would also have to raise ddr5 to 3900mhz. But 850 mhz is hard to obtain on 4870, even with a voltmod.

Let's see if 4890 has something different in its gpu, although I doubt it.

 

[Mech0z]

No, as stated its not just an oced card.

 

[Comdrpopnfresh]

Originally posted by: error8

Originally posted by: Comdrpopnfresh
so if someone has a 4870 1gb, and can overclock it to 850mhz, is that equivalent?

Probably. He would also have to raise ddr5 to 3900mhz. But 850 mhz is hard to obtain on 4870, even with a voltmod.

Let's see if 4890 has something different in its gpu, although I doubt it.

I have the asus eah4870 1gb dk. Stock core of 750, and 1gb of gddr5 @ 3600. I can easily set the core to 850mhz, and I don't touch the memory.
Agreed it is not just OC'd... but it seems the largest change is making 1gb of gddr5 the base reference model

 

[Paratus]

Just me, but I'd say the biggest change is the fact that it's a different chip.............

 

[HurleyBird]

In chronological order:

Originally posted by: josh6079

Originally posted by: thilan29
Some more info:
http://www.dailytech.com/article.aspx?newsid=14616

Interesting:

DailyTech
The chip is not just an overclocked RV770, as there are a few tweaks and modifications that we are still investigating.

Originally posted by: error8
Let's see if 4890 has something different in its gpu, although I doubt it.

Originally posted by: Mech0z
No, as stated its not just an oced card.

FAIL

Originally posted by: Paratus
Just me, but I'd say the biggest change is the fact that it's a different chip.............

WIN

 

[error8]

Originally posted by: HurleyBird
In chronological order:

Originally posted by: josh6079

Originally posted by: thilan29
Some more info:
http://www.dailytech.com/article.aspx?newsid=14616

Interesting:

DailyTech
The chip is not just an overclocked RV770, as there are a few tweaks and modifications that we are still investigating.

Originally posted by: error8
Let's see if 4890 has something different in its gpu, although I doubt it.

Originally posted by: Mech0z
No, as stated its not just an oced card.

FAIL

Originally posted by: Paratus
Just me, but I'd say the biggest change is the fact that it's a different chip.............

WIN

So simple and yet such a comprehensive post in the same time. Bleah..

Let's see what the big gpu improvements this card will have. It's impossible to be a new chip, if the card is still bellow gtx 285 at performance.