• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

ATi 4870 X2 (R700) thread

Page 12 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Originally posted by: lopri
GDDR5 is what makes RV770 really fly, from what it looks like. I mean, the only difference between HD 4870 and HD 4850 is the memory (and yes, some core overclocking).
Click to expand...

Here's an article (in French, but the graphs need no translation) which examines the performance effect of different memory speeds in the 4800 series cards:
http://www.hardware.fr/article...-4800-gddr5-utile.html

A few conclusions can be drawn upon examining the results:

* The average performance delta between the 4850 and 4870 increases with AA enabled
* At 1920x1200 the 4870 is about 25% faster than the 4850
* With both gpu's clocked at 725MHz the gddr5 version is 8-12% faster
* With 725 core and 993 mem clocks for both cards the gddr5 version is actually 10-12% slower (which doesn't really say much because I'm sure the 4870 gpu and the driver are tuned for 1800MHz memory clocks, so it's about as relevant as to putting high octane gas into a low tech engine and claiming it provides no benefit)
 
Originally posted by: munky
Originally posted by: lopri
GDDR5 is what makes RV770 really fly, from what it looks like. I mean, the only difference between HD 4870 and HD 4850 is the memory (and yes, some core overclocking).
Click to expand...

Here's an article (in French, but the graphs need no translation) which examines the performance effect of different memory speeds in the 4800 series cards:
http://www.hardware.fr/article...-4800-gddr5-utile.html

A few conclusions can be drawn upon examining the results:

* The average performance delta between the 4850 and 4870 increases with AA enabled
* At 1920x1200 the 4870 is about 25% faster than the 4850
* With both gpu's clocked at 725MHz the gddr5 version is 8-12% faster
* With 725 core and 993 mem clocks for both cards the gddr5 version is actually 10-12% slower (which doesn't really say much because I'm sure the 4870 gpu and the driver are tuned for 1800MHz memory clocks, so it's about as relevant as to putting high octane gas into a low tech engine and claiming it provides no benefit)
Click to expand...

GDDR5 is a higher latency memory than GDDR3. Hence the stellar clocks and quad rates. Makes perfect sense when both GDDR3 and GDDR5 are clocked the same, the GDDR3 would be faster.

 
keys - I was fairly certain that GDDR5 and GDDR3 had similar latencies... could you dig up your source that shows the disparity?
 
GDDR5 doubles the base speed 3 times (like DDR3), GDDR3 and GDDR4 double it twice (like DDR2).
Like DDR2 to DDR3, the latencies double as well.

It is inherent of the DDR method of increasing speed.

So when you downclock a GDDR5 to GDDR3 speeds, you are lowering the mhz without tightening the latencies, resulting in slower overall speed.
 
I don't even game on the PC anymore (360 now) but for some reason I'm still excited about the 4870x2 mainly because I could never get the "wow" IQ at 1920x1200 with the 8800GTX. It'll be interesting to see how this card takes on Crysis and it's upcoming expansion. Major credit to Ati for getting back in the fight with a knockdown! Impressive stuff.

Question. I have an Asus P5K Deluxe which I believe is Pci-e 1.1 complaint and not 2.0 Will there be a performance hit by not having a 2.0 Pci-e slot?
 
Originally posted by: lavaheadache

I don't think my minimums got lower by any noticeable difference, but the image quality greatly improves.
Click to expand...
If you're running vsync without triple buffering that's exactly what they'll do.

So on a 60 Hz vsync?d display if your framerate is 30-59 FPS it'll drop down to 30 FPS so yes, even your minimum can be impacted.

The whole point of triple buffering is to stop this and what is under question is the ability for triple buffering to function correctly with AFR style rendering.
 
but triple buffering introduces input lag...

One huge benefit of vsync that is not often discussed... if you are getting over 60fps, it caps it to 60. Which means that you are at less then 100% utilization, and produce less heat and take less electricity
 
Originally posted by: taltamir
GDDR5 doubles the base speed 3 times (like DDR3), GDDR3 and GDDR4 double it twice (like DDR2).
Like DDR2 to DDR3, the latencies double as well.

It is inherent of the DDR method of increasing speed.

So when you downclock a GDDR5 to GDDR3 speeds, you are lowering the mhz without tightening the latencies, resulting in slower overall speed.
Click to expand...

Are you talking of latency in terms of clock cycles or seconds?
 
Originally posted by: allies
keys - I was fairly certain that GDDR5 and GDDR3 had similar latencies... could you dig up your source that shows the disparity?
Click to expand...

Thats what i thought too. Its why GDDR4 never really took off because of its minimal speed increase compared to its higher latencies (in comparison with GDDR3).
 
Originally posted by: taltamir
but triple buffering introduces input lag...

One huge benefit of vsync that is not often discussed... if you are getting over 60fps, it caps it to 60. Which means that you are at less then 100% utilization, and produce less heat and take less electricity
Click to expand...
Triple buffering lowers input lag when used in conjunction with V-sync.
 
Originally posted by: Aberforth
R700 is 80% faster than GTX 280 and much cheaper

ZOMG
Click to expand...

"Up 80% faster." 50% faster on average. Still impressive.

The part that really has me intrigued is the 4850x2. I'm guessing that's going to come in at $350 MSRP and be about 20-25% faster than the 280. I may just have to pick me up one of those.

- woolfe
 
which is more like 20-30% faster when you consider that microstutter means that the TRUE FPS is lower then what is shown (not as slow as a single card, but still slower then the measured amount).

But the bottom line is... it is still FASTER.
 
07/10/2008 12:50 PM
Originally posted by: Aberforth
R700 is 80% faster than GTX 280 and much cheaper

ZOMG
Click to expand...

07/08/2008 06:29 PM
Originally posted by: jaredpace
I hear july 14th performance NDA, july 15th release, end-july wide availability, 15% or greater performance over 2x HD4870 Gddr5 in crossfire.
That makes it like 1.8 times as fast as a Geforce GTX 280 in certain scenarios. Also, 2GB versions will be available later on, perhaps august...
Click to expand...


🙂
 
Originally posted by: jaredpace
07/10/2008 12:50 PM
Originally posted by: Aberforth
R700 is 80% faster than GTX 280 and much cheaper

ZOMG
Click to expand...

07/08/2008 06:29 PM
Originally posted by: jaredpace
I hear july 14th performance NDA, july 15th release, end-july wide availability, 15% or greater performance over 2x HD4870 Gddr5 in crossfire.
That makes it like 1.8 times as fast as a Geforce GTX 280 in certain scenarios. Also, 2GB versions will be available later on, perhaps august...
Click to expand...


🙂
Click to expand...

right..sue me 😀
 
Originally posted by: jaredpace
07/10/2008 12:50 PM
Originally posted by: Aberforth
R700 is 80% faster than GTX 280 and much cheaper

ZOMG
Click to expand...

07/08/2008 06:29 PM
Originally posted by: jaredpace
I hear july 14th performance NDA, july 15th release, end-july wide availability, 15% or greater performance over 2x HD4870 Gddr5 in crossfire.
That makes it like 1.8 times as fast as a Geforce GTX 280 in certain scenarios. Also, 2GB versions will be available later on, perhaps august...
Click to expand...


🙂
Click to expand...

So, who is my future girlfriend and where can I find her? 😀

Seriously, thats very impressive, but not expected, considering the performance of a standard 4870

Also the 2GB should remove the advantage the GTX currently has @ 2560x1536, so that should be interesting
 
Now, poor me hoped that my gorgeus orange/black coloured corsair vx550 pew pew psu would hold up for a video card upgrade (8800gt currently), but looking at load consumption of a single hd 4870 tells me that a vx550 will nothing but short when it comes to delivering enough juice for a 4870x2. Soo, would a corsair tx650 suffice for a e4300 @ 3ghz, 2 harddisks, 3 fans and a 4870x2?
 
50%-80% faster is just freakin awesome. Can't wait. I'd really like to end this obsession and actually keep a card for more than six months. Looks like this might do it..
 
Originally posted by: taltamir
which is more like 20-30% faster when you consider that microstutter means that the TRUE FPS is lower then what is shown (not as slow as a single card, but still slower then the measured amount).

But the bottom line is... it is still FASTER.
Click to expand...

The 20-30% micro-stutter penalty seems arbitrary since we don't know how bad the effect would be. It could be better or it could be worse.

BTW increased latency is not inherent to 'DDR's method of increasing speed'. There is an inherent fixed latency (in ns) which is due to the underlying semiconductor physics. When latency is specified in terms of clock cycles, its value increases as clock speed increases. Like you rightly said, decrease in clock speed should be followed by tightening of latencies. The 'problem' is not with DDR but with specifying latency in terms of clock cycles (which definitely has its advantages)

For example GDDR3 and GDDR5 both have latency of 12 -16ns. There are cases like GDDR4 where the increased latency was due to trade offs in objectives (power savings for latency) with the address being sent in 2 cycles instead of 1. Most DDR speed improvements come from increased prefetch depth, I/O, termination, complex signaling, multiple clocks, etc.
 
4850x2 vs 4870x2 - PCI-e connectors?

We've seen 4850 - 1x6pin PCI-e (PCI-slot 75W + 6pin cable 75W=150W)
4870 = 8pin cable 150w + 6pin cable 75w + pci-e slot 75W

4870X2 is rumored to have same 1x6 + 1x8pin + PCI slot 75w, what about 4850X2? 2x6pin(or 1x8pin = 225W in total) - 4870 seems overdimensioned, 4870x2 maybe normal? , but will the 4850x2 be highly oc'bla with so much less watts ?

5-10% of gamers have the 8 and 6pin in PSU, 4850x2 a winner already?
 
ppl are too busy comparing to new 260 and 280 GTX..they perform bad!

9800GX2 beats 4870 all the way, 4850@Crossfire are little higher, but cost more.

With new prices, nvidia is at perf/price again (not incl G200 series)

also check the "regret getting HD4850 thread"...
 
4850 crossfire is $290 right now AR. how much less than that is 9800gx2? also, how much better will 4850 CF (or gt 200 for that matter) be in a month than 9800gx2? 6 months? I'd take the new gen from either camp any day over an old tech g92 part.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top