Sunday, July 6th 2008

R700 to Come in 2048MB Flavour?

Although unreliable sources have been pointing out to the possibility that The Radeon HD 4870 X2 (R700) could come in a 2GB variant for a while now, we now get reports from even the Taiwanese industry observer, DigiTimes that the R700 in fact could come in a 2048 MB GDDR5 flavour. Similar reports have also been sourced from the likes of HKEPC and ChipHell. Current products such as the Radeon HD 4870 come equipped with "IDGV51-05A1F1C-40X" 512 Mbit GDDR5 chips made by Qimonda, the cards feature 8 such chips. It's already known that the R700 comes with a total of 16 memory chips, 8 on each side of the PCB, Qimonda readies its 1 Gbit memory chips slated for a July, 2008 release, Hyundai/Hynix already have their parts in the making.

Expect this product to launch by late August.With inputs from DigiTimes
Add your own comment

41 Comments on R700 to Come in 2048MB Flavour?

#1
tkpenalty
GDDR5 = 2GB
GDDR3 = 1GB.

Thats what I've heard.
Posted on Reply
#2
btarunr
Editor & Senior Moderator
The other way round looks more realistic to me. We do have 1 Gbit GDDR3 chips, although DT says GDDR5 2 GB.
Posted on Reply
#3
calvary1980
2gig GDDR5 would cost alot.

- Christine
Posted on Reply
#4
Hayder_Master
by: calvary1980
2gig GDDR5 would cost alot.

- Christine
and more power + more heat
Posted on Reply
#5
lemonadesoda
It will cost a fortune, because it will me the FireGL version!
Posted on Reply
#6
tigger
I'm the only one
Omg the fireGL version of the R700 will cost two arms and legs.

Does the card really need 2gb?
Posted on Reply
#7
alwayssts
Yes, there is gigabit GDDR5. It is made by Hynix and is rated at 5gbps, the same as the the 512mb version from Qimonda. Hynix made the announcement back in November of last year, around the same time that the other companies made their 512mb anouncements (with rated speeds up to 6gbps as opposed to the 1gb 5gbps). Since Qimonda now has their chips available, it's not a stretch Hynix does/will as well.

edit: Qimonda has 1gb versions available as well, in both 4gbps and 5gbps. Qimonda has a shipping date of July 2008 on their site....Coincidence? I think not. :P

http://www.qimonda.com/graphics-ram/gddr5/index.html

look at the bottom of the page.


I don't think it's a stretch that AMD will use the 512mb and 1gb 5gbps on their 4890 models (rumored to be clocked in the 800+ core region), should they decide to release them. The only thing holding them up would be the 512mb 5gbps parts that have a availability date of Q308, which could mean now and that could mean next month or the month after. Don't be surprised if that does come to pass though.

Wouldn't it be interesting if they had a line-up like this? :

4850 - 625/1000-1200
4870 - 750/1800 (4gbps GDDR5)
4890 - 875/2400 (5gbps GDDR5)

Of course, with each part having their own respective X2 parts of the 1 and 2GB variety. The possibility is there, and the core and memory have been shown to both be capable.
Posted on Reply
#8
Ben Clarke
by: tigger69
Omg the fireGL version of the R700 will cost two arms and legs.

Does the card really need 2gb?
Of course it does, how else will we get acceptable frame rates on medium settings in Crysis?
Posted on Reply
#9
tigger
I'm the only one
Yes,but crysis is one game,and its getting tired now.I suppose the rich have to have something to spend their money on :p
Posted on Reply
#10
Mussels
Moderprator
well i'm not really surprised. 2GB was bound to happen very soon with the way video cards are progressing, and a dual GPU card makes sense.
Posted on Reply
#11
Jelle Mees
1024MB is more then enough for at least another year.
Posted on Reply
#12
Mussels
Moderprator
by: Jelle Mees
1024MB is more then enough for at least another year.
yeah... except that ram isnt shared in crossfire or SLI, so this is more or less a dual GPU 1GB card.
Posted on Reply
#13
KieranD
2gb might be needed because monitors are getting higher resolution and programs are getting more intensive. The Fire GL cards are for workstations and servers, you can use them in doing CAD and design.

2gb one for each gpu.

still the heat will be a lot but plenty faster tho i guess.
Posted on Reply
#14
btarunr
Editor & Senior Moderator
by: alwayssts
Yes, there is gigabit GDDR5.
Thanks for the information.
Posted on Reply
#15
jyoung75
I really hope this is 2GB of SHARED memory rather than 1GB per card. According to the latest buzz on ray tracing with Radeon cards, it's memory intensive http://www.tgdaily.com/content/view/38145/135/. I want my next card to be a ray tracing monster, since that's where the industry is headed. I hope the 4870x2 is that monster :rockout:
Posted on Reply
#18
eidairaman1
by: jyoung75
I really hope this is 2GB of SHARED memory rather than 1GB per card. According to the latest buzz on ray tracing with Radeon cards, it's memory intensive http://www.tgdaily.com/content/view/38145/135/. I want my next card to be a ray tracing monster, since that's where the industry is headed. I hope the 4870x2 is that monster :rockout:
we honestly dont know if it will head that way, but hope for the best and prepare for the worst, meaning cards that support both incase of a sudden dropoff of one technology over the other, but with games being created in raster/vector, ray tracing technique- cards will need to support both because im unsure ray tracing can emulate raster/vector efficient enough.
Posted on Reply
#19
Megasty
The way this thing is lining up, I may only have to buy one of them & just CF it with the 4870. However, the thought of having 4 gb of video memory across 4 GPUs is just sick.
Posted on Reply
#20
Mussels
Moderprator
by: Megasty
The way this thing is lining up, I may only have to buy one of them & just CF it with the 4870. However, the thought of having 4 gb of video memory across 4 GPUs is just sick.
except that like i said before, the ram isnt addivite. you'd only have 1GB of ram usable for the most part, as it duplicates the data across the video cards. (although they have gotten AA and AF to use the ram on the other cards to some extent, it seems)
Posted on Reply
#21
Megasty
by: Mussels
except that like i said before, the ram isnt addivite. you'd only have 1GB of ram usable for the most part, as it duplicates the data across the video cards. (although they have gotten AA and AF to use the ram on the other cards to some extent, it seems)
I did say across 4 GPUs, meaning 1 gb per GPU. I do know how the 3870x2 works & it appears that they're using the same arch with the PLX chip. I can just hope this PLX allows for more bandwidth than the prior version.
Posted on Reply
#22
WarEagleAU
Bird of Prey
They could turn tale and make it additive. That would be something. Of course, their is a slight premium on the chips (kind of hard to see with a kick ass card at 299 tops, 309 on the egg) but the heat and power are reduced on GDDR5 compared to 3/4 flavors.
Posted on Reply
#23
imperialreign
I wonder, though, even with each GPU having it's own 1GB of mem, and even at the bandwidth involved - just how long would it take to load everything up? I mean, comparatively, that was the biggest issue with the 3870x2 versus crossfired 3870s . . .

but, also, if each GPU has 1GB, how often would the card have to swap out files while in game? WOuld both GPUs have to have the same files loaded up?

But - if each GPU could access all DRAM modules, that would severelly reduce the need to swap out textures and other files as often
Posted on Reply
#24
largon
Shared memory is practically impossible in any sense that would effectively increase the total amount usable for both GPUs. Just forget it.
Posted on Reply
#25
Rash-Un-Al
Marchitecture

2GB is likely to be a simple and pure function of marketing.

The 48xx series cards have proven to not be memory buffer- or bandwidth-limited. This is especially so – at nearly any conceivable resolution, with or without filtration – with the 4870, with it's monstrous GDDR5 bandwidth capabilities. Add to this the fact – with 512 MB or 1024 MB of memory – the interface will remain 256-bit. Thus, whether it has 512 MB or 4096 MB it can access, write to, and read from the memory only to the extent to which the (256-bit) interface allows. This is why 1GB variants of cards of the same model and technology do not perform any faster. On the contrary, in some instances, they perform even more slowly because 3rd-party, larger-memory renditions of a card often bring about cheaper and less capable memory, which run at similar frequencies to the original memory, but that also introduce higher latencies to achieve said frequency.

For your (GDDR5) reading enjoyment:
http://forums.techpowerup.com/showthread.php?p=866224#post866224
Posted on Reply
Add your own comment