ATi 4870X2(R700) PCB Debut

[spursindonesia]

A single R700 on a P35 board, i think it will still kick some serious a$$es.

[clonez]

arent theyre rumors, that crossfire will become compatible with nvidia chipsets?

would increase the amount of potential customer (concerning multi-gpu solutions)
but i cant see a good reason to buy nvidia chipsets, anyway (except triple-sli o.c)

[Shintai]

arent theyre rumors, that crossfire will become compatible with nvidia chipsets?

would increase the amount of potential customer (concerning multi-gpu solutions)
but i cant see a good reason to buy nvidia chipsets, anyway (except triple-sli o.c)

What chipsets?

nVidia is dead in the Nehalem chipset business, and they are soon dead in the majority of AMDs when fusion comes. So unless you want CF on a VIA CPU

[jcool]

nVidia is dead in the Nehalem chipset business...

Yeah, payback's a
They should've thought about this when they so snootily refused Intel SLI support for their chipsets.. serves them right, I guess.

Note to all NV fanboys: I own lots of NV cards, so keep the flames.

[Helmore]

I thought the change from PCIe 1.0 to 2.0 only has to do with the power provided through the slot, and not really about bandwidth?

No, PCIe gen2 brings you double the bandwidth of PCIe gen1, so no you have 8,0 GB/s (PCIe 16x) in each direction with gen2 while you 'only' had 4,0GB/s (PCIe 16x) in each direction with gen1 . Gen1 is still more than sufficient when using a single card and this will still be the case with R700 (or 9800GX2 for that matter). I don't think gen1 will even be of any hindrance when doing CF with two 4870s for example, if you have 2 full 16x slots that is.

Why does everyone keep saying that NVIDIA does not get the right for QPI, I just won't believe it until it is finally so late that I just can't deny it anymore . No NVIDIA fanboy here though, I for some reason liked ATI more than NVIDIA although I can't really think if a good reason why, maybe I just like red or something.

[Logos]

just a bit more infos here:
http://forums.vr-zone.com/showthread.php?t=296463

Radeon HD 4870X2 to be available by end Jul, custom designs by August
We have seen how the 4850 can take on the 9800GTX, 4870 taking on the 260GTX. According to news we received, the Radeon HD 4870 X2 which integrates two RV770XT GPUs (R700) model 4870X2 ATi made cards will go on sale by end of July.

The ref model PCB is named Spartan, model B509. Clk speeds are uncertain at press time. It will come with 16X32M x32 GDDR5 memory, 512 bits wide, 2GB(256Bit、1GBx2) and will sell for 499 USD.

Source: OCW

[ken]

......

Radeon HD 4870X2 to be available by end Jul, custom designs by August
We have seen how the 4850 can take on the 9800GTX, 4870 taking on the 260GTX. According to news we received, the Radeon HD 4870 X2 which integrates two RV770XT GPUs (R700) model 4870X2 ATi made cards will go on sale by end of July.

The ref model PCB is named Spartan, model B509. Clk speeds are uncertain at press time. It will come with 16X32M x32 GDDR5 memory, 512 bits wide, 2GB(256Bit、1GBx2) and will sell for 499 USD.

Source: OCW

this is maddness

[JamesBong420]

2GB(256Bit、1GBx2)

do my eyes see this correctly 2 gb¿? omgoooses!!!

[flopper]

this is maddness

its my new card.

[jcool]

Mine too

[adamsleath]

2GB
if true
my card

[Unbornchild]

Speculation posted by someone on that forum, source without link.

[adamsleath]

yes, that thought did occur to me, but hopefuly someone can squeeeeeze 2GB onto the card
maybe they have triple sided gddr5 now?

[F.stop]

time will tell

[Chickenfeed]

A 1.1 16x slot will probaley be fine even on the 4870x2 as most of the gpu to gpu traffic is handled on the pcb and not over the slot, just like the 3870x2 and GX2.

The only reason a PCIe 2.0 board would be faster ( at this time ) is due to chipset differences and not the slot itself. ( eg 780i vs 790i ; 790i tends to be about 5-10% faster with SLI )

And madness? This isn't madness... this is AMD! ( sorry had too :p ) If all reference cards ship with 1GB of GDDR5 on black PCBs, I'll die of happiness.

EDIT: Heres the apparent OCW source - http://my.ocworkbench.com/bbs/showth...335#post433335

[grimREEFER]

"Radeon HD 4870X2 to be available by end Jul, custom designs by August"
i wouldnt hold out for the custom designed ones, apparently making a decent non-reference version of these dual die cards is very difficult.
note the sparseness of non-reference versions of every dual die card ever made by anyone.

[jVIDIA]

SPARTAN ... hummmm where did I see that before ?!

AHHHHH here:

GTX280 Review @ Guru3D

[ToTTenTranz]

I guess the fact that there is a 2GB model pretty much proves that there's no shared memory between the two GPUs.

[Aberration]

Or that the sharing is not ideal for most applications and games.

[G0ldBr1ck]

yes, that thought did occur to me, but hopefuly someone can squeeeeeze 2GB onto the card
maybe hey have triple sided gddr5 now?

qimonda has 1Gbit GDDR5 chips listed on there site as available july 8th, so that would be 8 chips per GPU to make 2 Gigabytes. Picture shows that many chips.

[Helmore]

There has never been any real proof for shared memory, the proof that is represented lately only suggest a better cooperation between the 2 GPU cores. This could result in lower memory usage but not shared memory, not for this generation at least as future generations can improve on this.

[annihilat0r]

Obviously, if ATI's really going for the multi-GPU way from now on, they'll have to improve the GPU cooperation, starting from now

[Aberration]

They need an MCM type package.

[Helmore]

I don't think they desperately need an MCM type of package, the connection between the 2 GPUs can be of low enough latency without the GPUs residing that close to each other.

[LordEC911]

Well, the 1Gb on GPU-Z can be taken two different ways...
#1- Shared memory
#2- 2x1Gb, like speculated.

Everyone was surprised at 40TMUs, everyone was surprised again with 800SPs, most were surprised at ~950tran and ~260mm2.
Why can't AMD/ATi surprise again with the R700?