Crown back to Team Red, 3870X2

http://img3.pconline.com.cn/pconli [...] _thumb.jpg

3DMark06 1280*1024 scores 18467 with a 3.6GHz Intel quad-core

12578@2560*1600

Really though power, huh?


I wonder what toy our green team's gonna show us.

A heavily OCed 8800GTS 512MB gets around 17200+ in 3Dmark06 with the 3.6Ghz quad cpu.

I'm not sure about this until I see some real benchmarks.

I like the black PCB, now if only they'd change that red shroud!

Yea, I am interested to see how this turns out as well. Even though I just got the card in my sig about a month ago. lol

Best,

3Ball

The word is it's faster than the Ultra. I don't know if that's in optimized setups or what... I'm really not sure how the Crossfire scaling works on it or if it is better than 2 HD3870 in scaling.

Eitherway... the initial reports of the HD3870X2 have me really excited for the summertime with the competition heating up with the G100 and RV770.

Yep. The proof is the person right above you. ^^

EDIT: Maybe I should of said 17000+

Readily bragging as always

And you fell for it

heheh

Sorry I don't live on the West Coast. It's 11:20 here and I'm tired. CROSSFIRE. YOU'LL GET CAUGHT UP IN THE ... CROSSFIRE.. CROSSFFFFFFFIIIIIIREE.

Just makes the game that much more interesting with balls flying everywhere.

You know as well as I do I have too many other expensive hobbies already.

Anyways.. HD3870X2 vs 2 HD3870s would be a fun test. As long as all clocks were equal... to see if the Dual GPU design has any interconnect advantages.

I'd also like to see a 9800 GX2 vs 2 8800 GTS G92s all at the same clocks to see the similar kind of thing from the nVidia camp.

Look, I calculated that the 9800 GX2 that is coming out will be about 11% faster then the 3870 X2 with no AA on.

But, 2 3870 X2 in crossfire, will be about 24% faster then 2 9800 GX2 in SLI with no AA on.

I came out with these numbers after hours of meditation and about 10mins in calculations.

So the Crown will go both ways, depending how you look at it!

It does beat the 8800 Ultra

http://en.expreview.com/?p=219#more-219

And im going to buy 2 of these and run them in X-Fire

Spoonboy found this interesting read concerning sli vs xfire.

http://www.legionhardware.com/docu [...] id=716&p=0

To me it isn't clear that the 3870x2 will be the better of the two.
We will just have to wait and see.

Put your money where your mouth is, I believe that all the ATI cards have a benefit because of 3Dmark. But if you look at the Game performance... The Ultra still beats the 3870x2 So it seem's that it shouldn't be an improvement.

Let them battle, I'll wait for benchmarks, and buy a new card in 2 years. My GTS 512 plays crysis at high... everything at high. without AA at 30-40 fps @ 1680x1050.

Sicking tired of people putting up benchmarks with 3Dmark, oh no look at my 3D mark its so high... Few moments later... hey how come you have 30 fps and I 20 while playing crysis.... My 3Dmark showed different. This has been going for a long time now.

Just hold your horses and wait for Game benchmarks that show the 9800 GX2 and 3870x2 ...

That Chinese site seems to have done more game benchmarks on the 3870x2. vr-zone forums has a good summary:
http://forums.vr-zone.com/showthread.php?t=228355

Here are all of the benchmarks together on one chart:

http://www.pconline.com.cn/diy/gra [...] 34_31.html

Although the review is in Chinese the 3870 X2 beat the 8800 Ultra in 7 out of 9 GAMING benchmarks
excluding 3D Mark.

Im sick and tired of people who think they know it all and the and talking about the 9800 GX2
(sorry but you deserved it)

6 out of 9 NFS graph is wrong but lets wait until 9800GX2!

A better indicator of what gaming performance? Current only or future?

The thing about 3Dmark is that it's NEVER been a great game predictor it's been a better feature and hardware tester within it's own products.

Optimizations and Floptimizations negate any real comparison. Perhaps the HD series is that much better, however because games are optimized and sold the way they are it's irrelevant, because they play differently than the benchmark runs.

Unlike what some people like Radium seem to suggest, it's not built to favour either company, if anything in the GF6&7 generation 3Dmk favoured nV because of their ability to use DST (which rarely was a game feature), now both companies offer the support (plus Fetch 4) and so really it's just the see-saw back and forth between designs. The main thing is the standard for 3Dmk involves massive workloads without much regard to AA performance, which reflects poorly on most other tests/games, but it may reflect the future performance of shader based games, so neither is a cut and dry route.

As always the best thing 3Dmark shows is..... whcih is best at 3Dmark.

That being said it sounds like 3Dmark Vanatage will give an even more narrow picture which may break more people's insistance of using it as a game predictor.

Don't get me wrong 3Dmark and it's bungholio scores have a very good place in people's toolbox, but too often they're used in ways they were never intended to be used and to predict/say things they aren't capable of.

Actually it's a single card versus single card. The GX2 was dual card single slot so it's not the same. As long as the software supports it the same physical limits to the Ultra are available to the X2, whereas the GX2 is limited in both cooling options and bridge scaling options. So as long as price/performance is worth it it's a relevant player, although mainly for the e-wang stuff not for sensible solutions.

It's not a impressive as a single chip solution, but considering the future both companies and intel are headed with their more modular designs, this is far from laughable, and really more of a preview of things to come.

The thing that wouldbe laughable is if it becomes a dead-end similar to the GX2 where alot of their advancement/work gets lost.

Not my cup of tea by any means, but this is likely the future of high end graphics, so anyone interested in such things better get used to it.

I'm just getting the Gigabyte GIGABYTE GV-RX387512H Radeon HD 3870 512MB to hold me over until midrange R700 by the end of the year.

The 3870x2 card looks good though. It has two GPU's on one PCB. I"m sure it will do better than the 9800x2, but what we all really need from ATI and Nvidia are dual, triple and quad core GPU's on one die. Unless of course there's something about GPU architecture that makes that unlikely.

When is the Evil Netburst Empire getting into the discrete GPU game? They've routed the rebels with their C2D onslaught, so their next target is the neutral planet ruled by Princess Envidia, who's main rival recently joined the Rebel Alliance. Not since the days Princess Envidia was stood up by that loutish FX has she been in such dire danger.

checked the thg review already. the 3870 is sweet, i just hope they come under $500.

the thing that intrigues me the most is that its over a year already and ati still cant outperform a year old architecture by a large margin. my theory is that even nvidia can outdo their old architecture themselves... its been so stagnant argh!

You might be correct it is the future:

http://www.techpowerup.com/

"According to a news story by the Chinese newspaper Commercial Times, ATI's RV770 GPU is due to be commercially released in late Q2 2008. Two of such chips are projected to power AMD’s new high-performance ATI R700 graphics card."

so another 3870x2 sort of card coming out this summer.

On the 3dmark point, I meant that taking test 3 results would be a better indicator of gaming performance than the aggregate score, as ati comes in slightly behind comparable nvidia cards in this test, which reflects true life as they are behind on average in gaming.

Depends on how you look at it.
Alot of people are confusing multi-chip with multi VPUs, and what has always be mentioned as the goal for the R7xx generation is modularity on a more simplified setup (dual GPU massive board is not simplified for the AIBs).

This probably gives the best description of the options and some of their benefits/drawbacks from a processor fab perspective (try to ignore CPU vs VPU for a second);
http://www.tomshardware.com/2005/1 [...] page2.html



Now yipsl is talking about multi-core on a single die which would essentially be similar to what we have for most current multi-core CPUs and would be detailed by the Athlon on that page; however this isn't as effective for the GPU which is already multifaceted and has a ton of parrallel processors where the best would be of course the giant itanium core with little duplication, but this increases costs significantly.

Now the other two options easily seen are either the Smithfield or Presler, however from the early talki it sounds like the RV770 is using a Smithfield type dual-core die with a double core Presler packaging idea (see above left) for the R700.

So what you end up with is 4 cores easily modularized for 3 different SKU lines;

HD4800 = 2 RV770 die for a total of 4 core on a single R700 'presler style' package/VPU (still single socket).
HD4600 = 1 RV770 die with both functioning core for a total of 2 cores.
HD4200 = 1 bin'ed RV770 with one core disabled giving you 1 core.

This should allow very god scaling because all of the communication is done on die and not through long wire traces on the PCB and there is no need to duplicate memory etc.

However should you find yourself behind the nV solution again then you could take a page from the HDxxxxxX2 and put 2 R700 packages/VPUs together and have 8 cores on one card to make the HD4900 or whatever.

To me that lets AMD satisfy almost all of their production needs with just 1 chip run which should be far more cost saving than having 3 seperate chips. Of course you need to learn to effectively manage the cores. However if succesful you could probably satisfy all of your customers with that chip and then a low-power, high functionality IGP style chip that would be good for HTPCs, OEMs, and IGP solutions.

Gluing single cores together is also doable but the scaling and interface surfaces offer different limitations, whereas a dual core being stuck to another dual core is far easier than sticking 4 cores together.

Anywhoo, we'll have to wait and see, but to me that's the best way to approach the future and keep cost low and yields high (compared to one giant 1-2 Billion transistor chip yield).

As always, just my two frames' worth from the cheap seats.


More pics to get a feel for what 2 dies on one package looks like;

http://www.tomshardware.pl/busines [...] armill.jpg
http://images.tomshardware.com/200 [...] cessor.jpg
http://ismailfaruqi.files.wordpres [...] resler.jpg
http://img158.imageshack.us/img158/6462/2ja4.jpg


Consider each die dual-core and you get what I mean.