Wednesday, November 11th 2009

AMD Radeon HD 5970 Specs Surface

In a few days from now, AMD will unveil its new flagship graphics accelerator, the ATI Radeon HD 5970, which will intends to cement the brand's performance leadership over every product from rival NVIDIA. The HD 5970, codenamed "Hemlock", is a dual-GPU accelerator, with two codenamed "Cypress" GPUs in an internal CrossfireX configuration.

Built on the 40 nm process, these GPUs will feature 1600 stream processors each, and will each have a 256-bit wide GDDR5 memory interface to connect to 2 GB of memory (4 GB total on card). The clock speeds are where the specifications of these GPUs differ from their single-GPU avatar, the Radeon HD 5870. The core is clocked at 725 MHz, while the memory runs at 1000 MHz (4000 MHz effective).

The accelerator will not have a rear panel identical to those of other Radeon HD 5000 series accelerators. It has the usual broad air vent occupying one slot, while the other has two DVI-D and one mini DisplayPort (DP) connector. The mini DP connector can give out DVI output using a dongle, and in this way, support for ATI Eyefinity technology remains intact. The NDA covering this accelerator is said to expire on the 19th of November, not very far away.Source: TechConnect Magazine
Add your own comment

147 Comments on AMD Radeon HD 5970 Specs Surface

#1
inferKNOX
W1zzard said:
i just got an answer regarding "4 GB card requires 64-bit OS".

This is not the case. Any such cards will work fine in a 32-bit OS. Once the texture is created in system memory (which you have to do in any case) the GPU is tasked with copying it to video memory without any CPU interaction (DMA transfer). In other words, the GPU is told: "copy 16 MB of texture from main memory address X into GPU memory". Done.
And how about how much system memory is needed to use the card? Is it alright to have less system RAM than GFX card RAM and still have the system run okay? (Coz I could have sworn I heard it here on TPU from someone I consider intelligent like yourself, that it can't.:confused:
Lol, sorry ToTTenTranz:ohwell:)
Posted on Reply
#2
Polarman
I'm just trying to imagine how much heat and noise this thing will generate. Ouch!
Posted on Reply
#3
Zubasa
Polarman said:
I'm just trying to imagine how much heat and noise this thing will generate. Ouch!
Hardly any noise, ATi usuaully let these cards run @90C so its dead quiet. :laugh:
Posted on Reply
#4
OnBoard
W1zzard said:
i just got an answer regarding "4 GB card requires 64-bit OS".

This is not the case. Any such cards will work fine in a 32-bit OS. Once the texture is created in system memory (which you have to do in any case) the GPU is tasked with copying it to video memory without any CPU interaction (DMA transfer). In other words, the GPU is told: "copy 16 MB of texture from main memory address X into GPU memory". Done.
Microsoft
Various devices in a typical computer require memory-mapped access. This is known as memory-mapped I/O (MMIO). For the MMIO space to be available to 32-bit operating systems, the MMIO space must reside within the first 4 GB of address space.

For example, if you have a video card that has 256 MB of onboard memory, that memory must be mapped within the first 4 GB of address space. If 4 GB of system memory is already installed, part of that address space must be reserved by the graphics memory mapping. Graphics memory mapping overwrites a part of the system memory. These conditions reduce the total amount of system memory that is available to the operating system.
http://support.microsoft.com/kb/929605/en-us

Doesn't that mean that with HD5970 you can only have max 2GB of RAM no matter how much is installed on your 32bit OS, hence the "you need 64bit OS" for this (and similar) card(s)?
Posted on Reply
#5
eidairaman1
The Exiled Airman
Maximum 32Bit OS supports for Main system memory is 3.25GB, if you want more than that you need 64bit OS.
Posted on Reply
#6
W1zzard
OnBoard said:
http://support.microsoft.com/kb/929605/en-us

Doesn't that mean that with HD5970 you can only have max 2GB of RAM no matter how much is installed on your 32bit OS, hence the "you need 64bit OS" for this (and similar) card(s)?
no. it maps maximum of 256 mb for the graphics card
Posted on Reply
#7
Animalpak
Zubasa said:
Hardly any noise, ATi usuaully let these cards run @90C so its dead quiet. :laugh:
a friend of mine had a 4870 X2 would say that warmed so much behind the case that dvi usb cables etc. ... Were soft from the heat.
Posted on Reply
#8
1Kurgan1
The Knife in your Back
Soylent Joe said:
Wow, this thing is going to be badass, and probably close to a grand.
A grand? Not a chance, this is 2 underclocked 5870 GPU's, so expect it to be under the cost of 2x 5870's. Anyone whos after this card will most likely want to get 2x 5870's instead, so they have to have the price under that. The real question is, are these binned down chips? Or are they just underclocked because of heat? If they are underclocked because of heat, anyone who puts a waterblock on this will be getting 2 full fledges 5870's for cheap.

AsRock said:
But if CF is not supported by the game it's going run worse than a single 5870 ?.
At stock clocks, yes, the 5970 would be slower in a game that doesnt support CF.

Animalpak said:
a friend of mine had a 4870 X2 would say that warmed so much behind the case that dvi usb cables etc. ... Were soft from the heat.
You could cook breakfast on the backplate of a 4870x2. And I'm not even kidding you, if I was to actually hold my finger on my backplate of my 4870x2 it would literally burn me. The exhausting air is pretty hot too, but I think your friend was exaggerating a bit there.
Posted on Reply
#9
newfellow
Looking to the specs this will cost small damn fortune. 4GB? my god that's gonna be something real. Wonder how will the little sister be of 5850x2. 2 x 1600SPs hell they can't even utilize low end 800SPs how the heck they think 3200SPs can be utilized and not to even consider what the heck will use such power? Considering any G92 even is way too much for todays demand.

I'm so gonna see when the power runs out even on, atm, equipment before even considering this kind of hardware plus the drivers most improve like 500% style and fast.
Posted on Reply
#11
eidairaman1
The Exiled Airman
all i hear is griping, if you want to gripe, send an Email to AMD.
Posted on Reply
#12
mtosev
newconroer said:
Maybe, but isn't the architecture still the same? The cards don't technically have access to 4gb in real time, it's 2gb/2gb instead?
there are two GPUs , GPU0 has 2GB available and GPU1 has 2GB of ram. if a game supports CF/SLi then all 4GB will be used. if i remember correctly part of the image is rendered by first gpu and the other part of the picture is rendered by the second gpu.
Posted on Reply
#13
Mussels
Moderprator
ToTTenTranz said:
That makes no sense at all! The memory subsystem is handled by the GPU itself, not the CPU.

Only if the PC used an UMA architecture like the consoles (or systems with IGPs), that constraint could exist.
it is a valid question, DX9 apps have to be loaded into system ram, then moved to the video card.

if it was larger than 2GB, a 32 bit OS/app would only be able to use the first 2GB.

then again, 32 bit apps can only use 2GB of address space total, so they'd be optimised to use less than that anyway


W1zzard said:
i just got an answer regarding "4 GB card requires 64-bit OS".

This is not the case. Any such cards will work fine in a 32-bit OS. Once the texture is created in system memory (which you have to do in any case) the GPU is tasked with copying it to video memory without any CPU interaction (DMA transfer). In other words, the GPU is told: "copy 16 MB of texture from main memory address X into GPU memory". Done.
i missed w1zzards post - but he pretty much said the same thing.

you have a 2GB limit between video ram and system ram for a 32 bit app, and they get to balance it out themselves so it doesnt crash by going over that limit.

sure a 2GB card wont reach its max potential in a 32 bit OS in DX9 - but so few games use 1GB of Vram, it doesnt matter just yet (and due to DX10's saving of system ram, it would also delay the problems for DX10/11 users)

mtosev said:
there are two GPUs , GPU0 has 2GB available and GPU1 has 2GB of ram. if a game supports CF/SLi then all 4GB will be used. if i remember correctly part of the image is rendered by first gpu and the other part of the picture is rendered by the second gpu.
This is not true. ram is not additive - each GPU in crossfire or SLI contains a copy of the ram in its entirety, its not divided up in any way
Posted on Reply
#14
erocker
eidairaman1 said:
all i hear is griping, if you want to gripe, send an Email to AMD.
Bah, and no one should be using a 32bit O/S nowdays anyways.
Posted on Reply
#15
pr0n Inspector
I will put this simply: to make memories accessible to software, they must have unique addresses. There is only 4G of these addresses in a plain 32-bit OS. The end.

If you want to see it with your own eyes, open Device Manger, look under the Resources tab of your video card. See those funny hex numbers? Those are the addresses assigned to your card's vRAM.
Posted on Reply
#16
mtosev
Mussels said:

This is not true. ram is not additive - each GPU in crossfire or SLI contains a copy of the ram in its entirety, its not divided up in any way
I ment that each GPU has it's own 2GB of ram and it addresses that amount of ram. and that the other gpu has it's own 2GB of ram.

if i'm wrong please correct me.
Posted on Reply
#17
Mussels
Moderprator
pr0n Inspector said:
I will put this simply: to make memories accessible to software, they must have unique addresses. There is only 4G of these addresses in a plain 32-bit OS. The end.

If you want to see it with your own eyes, open Device Manger, look under the Resources tab of your video card. See those funny hex numbers? Those are the addresses assigned to your card's vRAM.
4GB to the OS, 2GB address space to any one application.

we all know this - i have a nice educational link in my sig where we all hammered it out and finally got some accurate info (even w1z participated, thanks w1zzy)
Posted on Reply
#18
pr0n Inspector
Mussels said:
4GB to the OS, 2GB address space to any one application.

we all know this - i have a nice educational link in my sig where we all hammered it out and finally got some accurate info (even w1z participated, thanks w1zzy)
No. you are referring to the 50/50 memory splitting of Windows.

I am talking about the number of addresses available in a 32-bit OS. There are only 4G of addresses, and video card memory takes priority over system memory, thus bigger vRAM = less addresses for system RAM = system RAM "disappeared". What's more, video card RAM isn't the only thing that needs addresses, other devices need them too, so there's even less left for system RAM.
Posted on Reply
#19
Monkeywoman
i want one soo bad, going to wait till tax return time before i drop 600 on a vid card
Posted on Reply
#20
ToTTenTranz
W1zzard said:
i just got an answer regarding "4 GB card requires 64-bit OS".

This is not the case. Any such cards will work fine in a 32-bit OS. Once the texture is created in system memory (which you have to do in any case) the GPU is tasked with copying it to video memory without any CPU interaction (DMA transfer). In other words, the GPU is told: "copy 16 MB of texture from main memory address X into GPU memory". Done.
Of course. The system's CPU doesn't keep track of what's in the graphics card, which means it doesn't have to spend its registers to address the information in the graphics card.

Unless it's a UMA system, but that's a whole other story.
Posted on Reply
#21
eidairaman1
The Exiled Airman
erocker said:
Bah, and no one should be using a 32bit O/S nowdays anyways.
Trust me, My new machine will have Win 7 X64 and Win XP SP3 in dual boot from separate HDs.

Also FYI i was talking about a topic posted earlier

http://tinyurl.com/ykg9avt
Posted on Reply
#22
Mussels
Moderprator
ToTTenTranz said:
Of course. The system's CPU doesn't keep track of what's in the graphics card, which means it doesn't have to spend its registers to address the information in the graphics card.

Unless it's a UMA system, but that's a whole other story.
it doesnt keep track of whats in video memory, but it does have to keep a copy in local memory.

there are links about this in the x64 thread in my sig.

DX9 = system ram copied to video ram. System ram gets modified, then copied to video ram before the video card getsi t

DX10 (and up) = direct access to video ram. No system copy = less ram/address space used
Posted on Reply
#23
EarlZ
Just waiting for the GT300 to stomp this card and an X2 version to burry it :toast:
Posted on Reply
#24
Jizzler
^ 5970 X2? Yes, I'm liking that idea!

Then Crossfire a couple of them!
Posted on Reply
#25
facepunch
Jizzler said:
^ 5970 X2? Yes, I'm liking that idea!

Then Crossfire a couple of them!
agreed then buy a water block and over clock the hell out of them:rockout:
Posted on Reply
Add your own comment