[Autopsy] ATI Radeon HD 4870 Update: Review

Quote:

---

Originally Posted by Sr7

What do you mean being "driven at a hardware level as opposed to software"?

Makes no sense.

BTW If you deduced all those specs I'd love to see your scratch work :rolleyes:

---

think off chip memory controller or some other form of IO coordinater which makes crossfire done at a hardware level instead of a software level. again predicting the actual R700 to be around 70% faster than GT200 should this come true.


in my head... basically involved checking the number of shaders rumored and also assuming clock speed increases... net result, was wrong on the number of shaders(RV770 side) and clocks(was expecting 10-20% more clocks 50% more shaders and assumed around a 10% boost from the faster mem) but end numbers were right due to architectural improvements.

similar nv side expected lower clocks and 240/128 = 1.875 and 10% lower clocks and it's right around 1.7

all in all lots of speculation.

GT280 is 13% faster than a 4870 512mb. When we see the 1gb 4870 that percentage will reduce even more when playing at high resolutions.

Can't wait to have one :yepp:

Quote:

---

Originally Posted by xlink

think off chip memory controller or some other form of IO coordinater which makes crossfire done at a hardware level instead of a software level. again predicting the actual R700 to be around 70% faster than GT200 should this come true.


in my head... basically involved checking the number of shaders rumored and also assuming clock speed increases... net result, was wrong on the number of shaders(RV770 side) and clocks(was expecting 10-20% more clocks 50% more shaders and assumed around a 10% boost from the faster mem) but end numbers were right due to architectural improvements.

similar nv side expected lower clocks and 240/128 = 1.875 and 10% lower clocks and it's right around 1.7

all in all lots of speculation.

---

Even if they pulled off that incredibly complex feat (which they didn't) with the unified memory, there's no reason that that would imply greater performance over regular crossfire.. it just improves compatibility a great deal. Also, you're still going to have things like microstutter.

Quote:

---

Originally Posted by Sr7

Even if they pulled off that incredibly complex feat (which they didn't) with the unified memory, there's no reason that that would imply greater performance over regular crossfire.. it just improves compatibility a great deal. Also, you're still going to have things like microstutter.

---

the whole microstuttering issue will probably be imperceiveable though.

going to an on-card bridge chip takes A LOT less time than going to the system's northbridge or south bridge. remember CF has traditionally been implemented via software, so there is A LOT more overhead.

CF and SLI usually boost performance by, what 50-70% right? I'd think that if done right the boost here would be more like 90% or so.

Thats just insane! almost as fast as the GTX280 for half the price :eek2:




ATi--> :cord: <--Nvidia



Its almost ironic, seeing as Nv is following in the footsteps 3DFX in the final days. You know, using pretty much the same chip and charging alot for it just because it had a hit at one time. Just because the people flocked to the 8800GTX at $700 dont mean they will to the GTX280 with competition like this!

Not to say that the company is in any danger of going out of business, tho between Nehelam, larabee(maybe) and this... its gonna get rocky

Quote:

---

Originally Posted by xlink

the whole microstuttering issue will probably be imperceiveable though.

going to an on-card bridge chip takes A LOT less time than going to the system's northbridge or south bridge. remember CF has traditionally been implemented via software, so there is A LOT more overhead.

CF and SLI usually boost performance by, what 50-70&#37; right? I'd think that if done right the boost here would be more like 90% or so.

---

The boosts that multi-gpu delivers are due to increased processing power.

Do you know the bottlenecks of multi-gpu? I don't think you do. Because time/latency, the northbridge, and the southbridge certainly aren't the bottlenecks for SLI or CrossFire.

The cards do batched processing and present their frames interleaved. The things that hurt scaling are when someone makes 1 frame they render depend on the previous one somehow, which slows things down dramatically because you have to transfer the contents of one frame to the other card. While this situation would be improved with unified memory, it would not increase performance over any game that is optimally performing with multi-gpu today. The best case scenario is you get more applications which can be naive about how they code their games and still get scaling. You get compatibility and cost savings (need half the memory that you do today), not absolute speed increases.

Trust me, you're making results up based off of bottlenecks that you don't understand.

Quote:

---

Originally Posted by cookerjc

OK, the Chinese review done, i am translating it, put it to the en. site.

http://translate.google.com/translat...sl=zh-CN&tl=en
if you dont want to wait, just follow the link but it will be really slow for oversea readers

---

Although I had several server error messages, (I bet your hit count was out of sight!) I finally and slowly got to see most of your review.

Oh, by the way that was a good review. Let's hope that AMD/ATI decide they need to hang some clouds over the GTX 260 release tomorrow.

I don't know where I'd be with out the Google translator.

On the last page at the bottom is this...

Quote:

---

Do you think Radeon HD 4800 series will be successful? » Be on your comments!

---

Has anybody said no!

4870 looks very tasty, might switch out my 8800GTX for one if more numbers follow same pattern ;)

Quote:

---

Originally Posted by Sr7

The boosts that multi-gpu delivers are due to increased processing power.

Do you know the bottlenecks of multi-gpu? I don't think you do. Because time/latency, the northbridge, and the southbridge certainly aren't the bottlenecks for SLI or CrossFire.

The cards do batched processing and present their frames interleaved. The things that hurt scaling are when someone makes 1 frame they render depend on the previous one somehow, which slows things down dramatically because you have to transfer the contents of one frame to the other card. While this situation would be improved with unified memory, it would not increase performance over any game that is optimally performing with multi-gpu today. The best case scenario is you get more applications which can be naive about how they code their games and still get scaling. You get compatibility and cost savings (need half the memory that you do today), not absolute speed increases.

Trust me, you're making results up based off of bottlenecks that you don't understand.

---

uhh well...

traditionally memory latency has been considered negligible.

if you have two GPUs and they share a central memory controller off die, then that will add some latency, but in theory you could get pretty close scaling.

the bottlenecks in software mediated multi GPU solutions are are obvious... the software. each game has to be coded for in the drivers to enable a multi GPU configuration and I don't know if you've noticed but as of late programmers suck. when you add more and more human interaction into the mix the product will be less and less efficient if you haven't noticed.

and I don't know if you realized it or not, but "While this situation would be improved with unified memory, it would not increase performance over any game that is optimally performing with multi-gpu today" and as it is most games are not optimally performing... how many instances are there where a multi GPU solution is slower than a single GPU solution?

and as for the 90&#37; figure, I already said it was pure speculation.

anyone noticed that power consumption of the 4870?

6W less then a 280GTX :shock:

...nelson time
good to see ati on a comeback.
http://img.photobucket.com/albums/v5...lson-muntz.gif

http://xoqolatl.ovh.org/files/cram1.jpg
is that the exact same sink mounted 2 diffrent ways? if so, it's ..conveniently convenient :|
didnt know u could mount them like that.

http://translate.google.com/translat...sl=zh-CN&tl=en

That's a margin that really shows a new gen. Grats AMD.

ouch for NVIDIA, well i hope that there is still alot room for improvements in GT200 if we talking about driver optimisation, otherwise nvidia got gangbanged :)

wow! Beats the GTX260 in a lot of games except Lost Planet and is roughly 25&#37; faster than the 4850.

Lets hope Rivatuner / ATITool gets updated soon so we can see how a 4850 @ 750MHz core stacks up against the 4870. It'll be interesting to see how much of its gains are from memory bandwidth alone rather than both the core speed increase

Too bad current drivers don't support power play yet

we might see entirely different results :yepp:

Quote:

---

Originally Posted by adamsleath

.
is that the exact same sink mounted 2 diffrent ways? if so, it's ..conveniently convenient :|
didnt know u could mount them like that.

---

lol

The reverse mount isn't as efficient as the normal mount but we are talking a few degrees here.

I just purchased 10,000 AMD shares....CLIIIIIIMB! :D

Next level will be how good can this baby overclock!! Am still waiting for more reviews but i already know what my next card will be

I rememebr reading this was $299 still true?

So did nvidia get to open that can of whoop-ass or what? :down:

No, they didn't. NVidia is getting owned on pretty much every level at the moment:

Both at $200, a 4850 is considerably faster than a 9800GTX. Maybe the 9800GTX+ can match a 4850 but then it's about $240, which is significantly more expensive. At $240 most users would go with a HIS IceQ3 overclocked 4850 instead of a 9800GTX+, because it'd be faster.

At $300 nVidia has nothing. But ATI has its 4870's, which is again considerably faster than nVidia's $400 GTX 260's. Which makes the $240 GTX+ even less significant because most people would spend $60 more and get a 4870 for a superb leap in performance instead of spending $240 on a G92(b).

At $400 nVidia has GTX 260 which has been pretty much rendered worthless with the above figures of 4870's performance.

At $600 nVidia has the GTX 280 which is slower than nVidia's own EOL'ed $500 9800GX2. Yeah, SLI problems, microstuttering etc., but most users don't know about that and look only at performance numbers.

Then will come the R700 which will be around $500-600 and will kick 9800GX2's and GTX 280's ass by a cool margin. And with ATI's rumored new multi-GPU-on-a-single-card technology it might not even have the previous mGPU solutions' problems.

AAAnd, there's the 4850 Crossfire option for $400 which kicks solid ass also.

Actually I can't justify buying a nVidia product right now, especially in the $200-350 range; which happens to be the deciding factor when it comes to sales and profit.

Quote:

---

Originally Posted by annihilat0r

No, they didn't. NVidia is getting owned on pretty much every level at the moment:

Both at $200, a 4850 is considerably faster than a 9800GTX. Maybe the 9800GTX+ can match a 4850 but then it's about $240, which is significantly more expensive. At $240 most users would go with a HIS IceQ3 overclocked 4850 instead of a 9800GTX+, because it'd be faster.

At $300 nVidia has nothing. But ATI has its 4870's, which is again considerably faster than nVidia's $400 GTX 260's. Which makes the $240 GTX+ even less significant because most people would spend $60 more and get a 4870 for a superb leap in performance instead of spending $240 on a G92(b).

At $400 nVidia has GTX 260 which has been pretty much rendered worthless with the above figures of 4870's performance.

At $600 nVidia has the GTX 280 which is slower than nVidia's own EOL'ed $500 9800GX2. Yeah, SLI problems, microstuttering etc., but most users don't know about that and look only at performance numbers.

Then will come the R700 which will be around $500-600 and will kick 9800GX2's and GTX 280's ass by a cool margin. And with ATI's rumored new multi-GPU-on-a-single-card technology it might not even have the previous mGPU solutions' problems.

AAAnd, there's the 4850 Crossfire option for $400 which kicks solid ass also.

Actually I can't justify buying a nVidia product right now, especially in the $200-350 range; which happens to be the deciding factor when it comes to sales and profit.

---

Or you can buy one because that's "The Way Games Are Supposed To Be Played", by paying thru your arse. :ROTF:

lol... :D :ROTF:

The translation is done apparently so here it is in english.

http://en.expreview.com/2008/06/24/f...0-and-hd-4850/

Quote:

---

Originally Posted by AriciU

The translation is done apparently so here it is in english.

http://en.expreview.com/2008/06/24/f...0-and-hd-4850/

---

Is it just me, when I click on it I come to a blog page with all articles listed and nothing happens when I try to click on that 4870 and 4850 review.

EDIT: NVM, worked now. LOL 25 pages, should be a good read.