ATI Radeon HD 4000 Series discussion

[darkskypoet]

So then with this PLX chip: http://www.plxtech.com/products/expresslane/pex8648.asp
48 Lanes can be divied up into 16 for link to main board, and 16 to each GPU... from the spec page, it seems it can allow for a flow through (non-blocking) passing of data in both host centric, and peer to peer... Allowing I guess for system to pass same data to both cards at once without having to share the 16 lanes between two gpus... further it then allows full 16x PCIE 2.0 comm between GPUs... So from PCI SIG FAQ, that's about 16GB/s... Add to this whatever bandwidth is available through Crossfire Side Port... and that's not bad... Certainly greater then whats available on the 3870x2.

Is it enough to then to start using the other gpu's memory / cache to make 2 gpus and 2 sets of memory better then the sum of its parts?

[Shintai]

So then with this PLX chip: http://www.plxtech.com/products/expresslane/pex8648.asp
48 Lanes can be divied up into 16 for link to main board, and 16 to each GPU... from the spec page, it seems it can allow for a flow through (non-blocking) passing of data in both host centric, and peer to peer... Allowing I guess for system to pass same data to both cards at once without having to share the 16 lanes between two gpus... further it then allows full 16x PCIE 2.0 comm between GPUs... So from PCI SIG FAQ, that's about 16GB/s... Add to this whatever bandwidth is available through Crossfire Side Port... and that's not bad... Certainly greater then whats available on the 3870x2.

Is it enough to then to start using the other gpu's memory / cache to make 2 gpus and 2 sets of memory better then the sum of its parts?

PCIe 2.0 is 8GB in each direction, not 16. Plus that bandwidth will also be shared with the CPU for half of it.

[darkskypoet]

8 gb in either direction.. equals 16gb/s total. Yes. each GPU has 16 lanes to bridge chip, so 16GB/s total bandwidth between each gpu and bridge. Yes. and 16GB/s total bandwidth between bridge and system PCIE Switch. Yes.

No, there is no half of it is shared with CPU... Whatever is required to be passed through to / from cpu, memory, etc goes through bridge to one or both GPUs. It's not hard set.. Period. Add to that the bandwidth available over CF side port. So You're right cpu uses some bandwidth, gpus use some bandwidth, but this half stuff isn't true. As well, the GPUs can use their full 16 lanes to talk to each other as needed, and only need to cede lanes to incoming / outgoing data as necessary. One of the advantages of having a fully programmable 48 lane PCIE 2.0 switch on the card, is that you get the flexibility of adapting to data as it flows, no having to hard set anything like this. As well, the switch features some built in QoS features which would assist in the shaping of traffic between gpus, and between system and gpus.

So what were you trying to say Shintai?

In fact... might as well check this out... Here is a white paper on a chip like this in a graphics setup... Kind of neat if you care to read it: http://www.plxtech.com/pdf/apps/ExpA...cs_07Mar07.pdf I think it's actuall the one used on the 3870 x2. If I have my model numbers right, but it also supported dual cast, and low latency ops among other things.

[Helmore]

You have too high hopes for that CrossFire(X) sideport, this sideport has always been there since the X1950 Pro and it is nothing special so far. What else do you think those 2 golden fingers at the top of the card are for? That is simply the CFX sideport. They have been improving it gradually, but I don't expect miracles this time.

@zerazax: There won't ever be >2x scaling, not even when building a monolithic die with twice the processing units.

[Hornet331]

You have too high hopes for that CrossFire(X) sideport, this sideport has always been there since the X1950 Pro and it is nothing special so far. What else do you think those 2 golden fingers at the top of the card are for? That is simply the CFX sideport. They have been improving it gradually, but I don't expect miracles this time.

@zerazax: There won't ever be >2x scaling, not even when building a monolithic die with twice the processing units.

i think your mixing things up. The new CF Sideport is on the chip, its not the 2 normal CF connector to allow CF.

Rumors say its a new way to connect 2 cores directly, and allow better scalling.

[Jowy Atreides]

If the 4870x2 is on a black PCB, I'll be so stoked. I'd much sooner have 1 full coverage block for a dual gpu card then 2 separate blocks so I doubt I'll touch 2 4870s now. I might just order a 4850 to play around with until the 4870x2 is out though Wouldn't mind trying to get one past 750hz somehow hehe.

it is a black pcb

[Helmore]

i think your mixing things up. The new CF Sideport is on the chip, its not the 2 normal CF connector to allow CF.

Rumors say its a new way to connect 2 cores directly, and allow better scalling.

Nope I'm not mixing things up here, that sideport has been on the chip since the X1950Pro and yes it does connect 2 cores directly. It has done so since X1950Pro.
It might not be the exact same sideport anymore, but I don't expect miracles that's all.

[Helmore]

I have been told by a reliable source just an hour ago that R700 has a PCIe 2.0 PLX chip. He has been reliable in the past and I have begged him for this info the second he had it. apparently the assemblie lines have fired up...

unfortunaly he has no tech. info other then what can be visably seen.

So this has basically been confirmed:
http://www.xtremesystems.org/forums/...53&postcount=5 and
http://www.xtremesystems.org/forums/...&postcount=131

Seems I guessed right , although I got my info here:
http://www.hardware.info/nl-NL/news/...700_opgedoken/
It's Dutch so not everyone can read it, or just grab a translator, but they have said something like this about a week ago.

[zerazax]

You have too high hopes for that CrossFire(X) sideport, this sideport has always been there since the X1950 Pro and it is nothing special so far. What else do you think those 2 golden fingers at the top of the card are for? That is simply the CFX sideport. They have been improving it gradually, but I don't expect miracles this time.

@zerazax: There won't ever be >2x scaling, not even when building a monolithic die with twice the processing units.

Then go tell that to Eric Demers, where I got that quote from. He by the way worked on R600 and RV770 probably as well, so I'd say he's quite a bit qualified to know what the heck the architecture is capable of.

And you've got it completely wrong. The CrossFireX Sideport is a PART of the chip. Look at this:

http://www.rage3d.com/reviews/video/...70-diagram.jpg
It's connected to the hub.

Also, take a look at the hub:

http://www.rage3d.com/reviews/video/...s/Hartog/8.jpg

Again, its part of the chip on the hub. The CrossFireX sideport was not part of R600's architecture layout nor R520/R580 when those gold fingers started appearing. Furthermore, the slide even states CrossFireX interconnect which is similar to that slide of R700 where it states "new GPU interconnect" or whatever.

Oh and just for effect, from anandtech's review (link)

The CrossFire Sideport

Although AMD isn't talking about it now, the CrossFire Sideport is a new feature of the RV770 architecture that isn't in use on the RV770 at all. In future, single-card, multi-GPU solutions (*cough* R700) this interface will be used to communicate between adjacent GPUs - in theory allowing for better scaling with CrossFire. We'll be able to test this shortly as AMD is quickly readying its dual-GPU RV770 card under the R700 codename.

One thing is for sure, anything AMD can do to assist in providing more reliable consistent scaling with CrossFire will go a long way to help them move past some of the road blocks they currently have with respect to competing in the high end space. We're excited to see if this really makes a difference, as currently CrossFire is performed the same way it always has been: by combining the output of the rendered framebuffer of two cards. Adding some sort of real GPU-to-GPU communication might help sort out some of their issues.

Or, how about extremetech (link)

This may be our first hint at what is new with the dual-chip "R700" product coming in a few months—these hubs may communicate between two RV770 chips in a fashion that is more efficient than in past multi-GPU boards.

[Helmore]

@Helmore:

Actually, If you think of it.. with both Intel and AMD having IMCs... The next move would be onto CPU module. Never mind just the Marchitecture speak, but think right now on an AMD platform, you have Memory ->CPU -> Chipset (PCI Express arbiter) -> VC... With CSI, same thing. The move to the CPU will be made to drop the extra steps out of System ram to video ram. Now honestly, I couldn't quantify the additional penalties for having to route memory requests to cpu IMC and then passing data back through CPU... and In reality it is probably less so then utilizing say PCI Express slots off the SB (think NB <-> SB config), but i think we'll be seeing these GPGPU cores on CPU sooner rather then later. I mean think of the benefits in terms of CPU to GPGPU comm, and memory to both... Now don't get me wrong, it won't happen over night. But it will happen, and will cause a massive shift in the industry. And then what are we looking at? An SGI Shared memory infrastructure? Or tiered system memory? Embedded DRAM a la Bit Boys (shared CPU/GPU L3 Cache) and very fast DDR3 or DDR 5 system memory modules for access by both? I mean in a nutshell this is Fusion. Think of the tremendous redesign in such a chip... You could probably shave the CPU component down some, and leverage the strengths of GPGPU transistors to take on jobs that the CPU's FPU currently handles. For me this is the real game at hand, and a lot of the tech(s) being developed now are stepping stones to something more like a CPU / GPGPU design. Once you start to play with that sort of idea; the magnitude of design shift and paradigm rethink is tremendous... yet the benefits could be equally great. Now chips like the Cell start to become somewhat more interesting if you consider that in 5 years time the same sorts of ideas will be played out in PC land.

(It also puts AMD/ATI's Vector SP design into a different light, AMD is banking on merging the two in the near future. So does it make more sense for them to have gone with Nvidias very scalar method of processing? Or AMD's more vector oriented SPs? I honestly don't know, but I do wonder if they are thinking in terms of what will make a faster unified CPU/GPU. To be able to harness half the power of rv770 as an extended FPU would be incredible. I think part of AMDs teething problems, are that they can't afford to just R&D for its own sake. They can't afford not to make a product off of their stepping stones. So they learn, and drop a product to us; moving towards a goal in the 5 year time frame (maybe longer who knows). Perhaps their design decisions on discrete 'now' are being informed by what they see fusion as being later. Again, Multi GPU and multi die communication then becomes more important because it may be a requirement for their fusion plans. As well, the same might be said for vector simd SP approach vs scalar sp approach. Hard to say. But exciting none the less.

Intel's Nehalem has 3 x DDR 3 channels? Interesting, as that surely is laying the groundwork for adding Larabee later on... consider the die size comparisons, and compare GT280 to Nehalem... Lots of space left to add video, non? Further... intel's 45nm node is far smaller then that, and far better 'tweaked'... If Nvidia using 3rd party libraries can lay 1.4 billion transistors at 65nm, intel can do far more on in house 45nm. However, i bet they would do it MCM, so that the Frequency of Nehalem is not held back by the transistor rich graphics die. Also, when you have 14+ fabs, why not break production down into multiple dies? Allows for Discrete and embedded video. (side note: MCM allows for completely separate power planes w/o trying to do it on 1 die, I believe this hurt Phenom quite a bit)

AMD uses a Superscalar design on since R600, RV770 for example has 160 Superscalar shader units and one superscalar unit consists of 5 scalar units (4 basic and 1 'super size') or ALUs in other words. So this makes for 800 ALUs in RV770, while GT200 has 240 ALUs although their ALUs are a little flexible in certain form of code (mainly with dependencies). All scalar ALUs, so no vector design from either of them, although it could be said that ATI has in some sort of way 160 vector processing units as 1 unit can process 5 scalars at the same time, but then you are just discussing definitions.

Larabee is designed to run at 3,2GHz. so there pretty much is no clock deficit between Larabee and Nehalem and this will make integrating both on 1 die a lot easier. Although projects like Fusion will be a major shift in how things are done right now, they won't ever make discrete graphics cards obsolete, not in the next 15 years at least. I also don't think that they will shave some parts of the CPU down because that GPUGPU could also do it and that is because a CPU will always be faster in certain workloads and that's simply workloads that can't be mulitthreaded.

[Helmore]

Then go tell that to Eric Demers, where I got that quote from. He by the way worked on R600 and RV770 probably as well, so I'd say he's quite a bit qualified to know what the heck the architecture is capable of.

And you've got it completely wrong. The CrossFireX Sideport is a PART of the chip. Look at this:

http://www.rage3d.com/reviews/video/...70-diagram.jpg
It's connected to the hub.

Also, take a look at the hub:

http://www.rage3d.com/reviews/video/...s/Hartog/8.jpg

Again, its part of the chip on the hub. The CrossFireX sideport was not part of R600's architecture layout nor R520/R580 when those gold fingers started appearing. Furthermore, the slide even states CrossFireX interconnect which is similar to that slide of R700 where it states "new GPU interconnect" or whatever.

Oh and just for effect, from anandtech's review (link)



Or, how about extremetech (link)

I currently have that Rage3d.com link open in another tab atm . I never said that that sideport was not part of the chip, just that there has been this sideport on the chip since the X1950Pro. This CrossFireX sideport is connected to the hub this time around and it apparently is a 'relatively low bandwidth interconnect'. Oh BTW, I hope I'm completely wrong, it's just that I don't like to be disappointed so I just lower my hopes to decrease the chance of disappointment .
Only the AnandTech article states that this sideport is meant specifically for communication between adjacent GPUs, this has so far not been stated anywhere else. But if this is the case then they drastically changed this sideport compared to previous generations and with good results hopefully as well.

[ToTTenTranz]

Although projects like Fusion will be a major shift in how things are done right now, they won't ever make discrete graphics cards obsolete, not in the next 15 years at least.

15?! I'll give it half of that. 7,5 years, or 10 years maximum!

[darkskypoet]

Yeah I was considering the 5 wide structure in a SIMD vector sense... as you can't stuff it in other ways. This leads to the lower minimum throughput on the AMD architecture vs the higher minimum throughput of the nvidia architecture. However, if you can keep them full (have more multi op shader heavy code) then all of a sudden it becomes a much more powerful beast. Perhaps it is all in the definitions, like most things are. lol.

As to the shaving of the CPU, they can so do it; as you put some ops don't work in GPGPUs I agree, however there is over lap between the two, and having a set of GPGPU cores local allows for some of this overlap to be removed. There is always something better to allocate transistors for, and transistors are never free, thus if at all possible to make a more efficient design i am sure it will happen, in this case its more a case of adding to the cpu this functionality, and dropping out some of the redundant bits.

Not sure where you get your numbers for Larrabee; but I found this:

"Larabee is a bit different from today's modern GPUs. While they are also highly parallelized structures, they are not as CPU-like as Larabee is. Larabee will feature something in the range of 16-24 cores operating at 1.75-2.5GHz, with 32KB L1 cache per core and a shared 4-6MB L2 cache. This is all very preliminary, and the final specifications are far from set.

The cores will be connected in a ring bus not entirely unlike the one used by the Cell Broadband Engine inside of the PlayStation 3."

http://www.nordichardware.com/news,7459.html

Maybe you could elaborate your sources a bit better, In a quick search I couldn't find much more recent. Always prefer to be corrected with accurate info. (or even slightly less spurious stuff)

Last... I am sorry, but even after being up all night working on a Rousseau paper, I wouldn't dare predict 15 years from now. Lets see... hmm 15 years ago...

"IBM introduces the PowerStation 355 workstation. Features include 42 MHz processor, 16 MB RAM, IBM 6091 color display, graphics adapter. Price is US$15,995. [928.22] "
"Sun Microsystems introduces and ships the SparcStation 10 Model 512MP, with 50 MHz SuperSparc+ CPUs. Speed ratings are 65-68 SpecInt92 and 80-85 SpecFP92. Price is US$33745. [2046.29]"
"IBM announces the RS/6000 Model 990 computer. It features 71.5 MHz Power2 processor, CD-ROM drive, 15 Micro Channel slots. Prices start at US$124,500. Performance is rated at 126 SPECint92, 260 SPECfp92. [2049.2] [1812] "

http://www.islandnet.com/~KPOLSSON/w...t/work1993.htm

"Intel introduces details of the Pentium processor. It uses 32-bit registers, with a 64-bit (60 MHz) data bus, giving it an address space of 4 GB. It incorporates 3.1 million transistors, using 0.8-micron BiCMOS technology. Speeds are 60 MHz (100 MIPS) and 66 MHz (112 MIPS). Prices are US$878 (60 MHz) and US$964 (66 MHz)."

"Motorola begins shipping its PowerPC 601 processors, in 50 MHz and 66 MHz speeds. The processors feature 32 kB cache, and incorporate 2.8 million transistors in a 0.65-micron CMOS process. Pricing is about US$350 and US$470, respectively, in 1000 unit quantities. [57]"

"Cyrix announces the Cx486S2/40 and Cx486S2/50 clock-doubled processors. Prices are US$179 (40 MHz), and US$199 (50 MHz) in 1000 unit quantities. [985]"

"Cyrix announces the Cx486SV/25 and Cx486SV/33 3.3 volt processors. Prices are US$139 (25 MHz), and US$159 (33 MHz) in 1000 unit quantities. [985]"

http://www.islandnet.com/~KPOLSSON/c...t/comp1993.htm

Yes, and 15 years from now... I doubt a 'video card' will be anything more then a chunk of a greater system die. Will there be some firm somewhere with a dedicated discrete video solution for some sort of uberpowerful system? Maybe... but will it be a gaming card? no. High end workstation 'card'? doubt it. Look at what has become integrated into system hubs, or 'chipsets' in 15 years. Nah, I doubt video cards will be recognizable as such in that time frame. Rather the CPU will morph into something able to run massively parallel code itself... You'll just get a faster one of those, when you throw away / recycle your 'deck'... LOL 2.8million transistors to 1.4 billion transistors... Roughly a 500x increase in transistors on a die in 15 years... Even allowing for only an increase of 10x in the next 15 years... Do you have any notion of what you could do with 14 billion transistors? Maybe say 100x more... 140 Billion transistors... Count all the transistors in your system right now... I am sure you'll see what I am driving at.

Like I said tho, who the hell knows what the state of the art will look like in 15 years. I doubt that video cards as we know them today will be a part of that.

[Helmore]

Hmm, I guess I was wrong on the clocks for Larabee...
About that 15-years from now scenario, I don't think any of us has any idea what it will be like. One thing is probably for sure though, one of us is wrong and I hope it's not me just because that would hamper upgrading as we know it now. Now you can buy a decent system, then when a better video card is released you buy that to replace your current and this way you can keep upgrading your system piece for piece. That will be a lot tougher when having the GPU integrated on the CPU.
The first fusion processors will be here at the end of 2009 and they will be aimed at laptops. It will probably take another 5 years before every PC and laptop has a GPU integrated along with their CPU and then we will see games that use those cores for physics and AI. What happens afterwards is hard do predict though, but discrete graphics cards won't be gone for at least 8 years from now and trying to look further than that only thickens the 'fog'.

[Hornet331]

Hmm, I guess I was wrong on the clocks for Larabee...
About that 15-years from now scenario, I don't think any of us has any idea what it will be like. One thing is probably for sure though, one of us is wrong and I hope it's not me just because that would hamper upgrading as we know it now. Now you can buy a decent system, then when a better video card is released you buy that to replace your current and this way you can keep upgrading your system piece for piece. That will be a lot tougher when having the GPU integrated on the CPU.
The first fusion processors will be here at the end of 2009 and they will be aimed at laptops. It will probably take another 5 years before every PC and laptop has a GPU integrated along with their CPU and then we will see games that use those cores for physics and AI. What happens afterwards is hard do predict though, but discrete graphics cards won't be gone for at least 8 years from now and trying to look further than that only thickens the 'fog'.

havendale/auburndale comes a bit earlyer then fusion.

[darkskypoet]

It's not a case of 'upgrading' as we know it... It'll be closer to a case of buying drop in components onto something resembling a barebones deck (cyberpunk reference). Look at the do it yourself lappies; cept think smaller. And think of the commoditization of smaller hardware. drop optical drives for memory cards, hard drives for ssd implementations of greater and greater densities... essentially it'll be a much smaller form factor with no moving parts except perhaps for cooling... Even then.. I am not so sure about that. Solid state everything, and as transistor densities grow, everything becomes smaller. You'll have a line of AMD / Intel / nvidia system ships to drop in, perhaps a few extra terrabytes of memory ram / storage.. who cares at that point with the future speed of storage, probably really only differentiate between cache and system memory. Sure prefab, and DIY decks / tablets / 'smart phones' (yuck). Perhaps you buy the extra chip to drop into your DIY deck that acts as an accelerator coprocessor for GPGPU tasks, perhaps you dropped one in that did something else... Point being it'll be chips and teeny form factors. The majority of which will be commoditized like PCs are already starting to be / are today. As for 90&#37; of systems sold, the off the shelf OEM type is fine... There will always be people like us, but the paradigm is changing... Displays will get bigger, systems smaller, more powerful, more solid state. Hell it wouldn't surprise me if most of what we have today could be printed into flexible composite circuit sheets that you pick your options, print and slip into a casing you picked out from a selection in a showcase. 15 years is a long time, and all we are doing is pushing current.

[darkskypoet]

"I currently have that Rage3d.com link open in another tab atm . I never said that that sideport was not part of the chip, just that there has been this sideport on the chip since the X1950Pro. This CrossFireX sideport is connected to the hub this time around and it apparently is a 'relatively low bandwidth interconnect'."

Do you have a link for the rv670 block diagram with crossfire side port shown on it? i can't find one. Please share your source diagram... Ditto for x1950xt... no Crossfire side port connected to the memory bus / hub seen there either.

http://www.google.ca/imgres?imgurl=h...&ct=image&cd=1


The physical connectors on the card, are not what we mean when we say Crossfire X side port... In this case it seems to be something worthy enough of note that it is now included on the block diagram for RV770, and does not appear on the Ringbus diagram for r600, nor any block diagram I can find for rv670... Notice in particular the Ring Bus diagram for R600 - this shows the ring stop for PCIe, which is now located attached to the new hub on rv770... The Crossfire X side port is also connected to this hub. So how does that equate to Cross Fire X side port being low bandwidth? (can see relatively low bandwidth as in less then GDDR5 throughput, even PCIE 2.0 x16 is relatively lower bandwidth then that, granted) As well, if it was present on R600, why is it absent from all the diagrams?

http://www.ixbt.com/video3/images/r600/diag_ringbus.png

http://hothardware.com/Articles/ATI_...rrived/?page=2

[mascaras]

Asetek Intros First Liquid Cooling For ATI HD 4000 Series


Asetek’s industry-leading, LCLC liquid cooling solution is now available for AMD’s latest ATI Radeon HD 4870 family of graphics cards. The unique, single-slot liquid cooling approach provides OEMs with an extremely high overclocking capability enabling them to design and produce outstanding gaming and entertainment products with virtually no noise.

To meet the demands of advanced gaming or graphics-intensive applications, Asetek’s highly integrated solution lowers the GPU temperatures found on the 4870 by as much as 26 degrees, enabling it to run significantly cooler in extreme performance modes of operation. And, since the Asetek solution is totally liquid cooled, the graphics card can run completely silent, except for the system heat exchanger fan that typically runs at a quiet 30 dB(A).




Asetek


http://www.techpowerup.com/63990/Ase...00_Series.html

[Taoof]

Hi guys,
could someone tell if is there any new about relase date?

[purecain]

if anyone knows where i can get the new bios for 4870 with powerplay fix (fan fix) please pm me or post here- http://www.xtremesystems.org/forums/...d.php?t=192589
thanks... if i find it first i'll add it to this thread for everyone else....

[emoners]

if anyone knows where i can get the new bios for 4870 with powerplay fix (fan fix) please pm me or post here- http://www.xtremesystems.org/forums/...d.php?t=192589
thanks... if i find it first i'll add it to this thread for everyone else....

i'd like to know too...& how to get it installed

[spursindonesia]

@mascaras, bro, regarding the pic in your post about Asetek watercooling block for HD 4800 series, isn't that an integrated watercooling solution for Intel LGA775 socket CPUs ?

[MpG]

Must be a mis-pic, since Asetek's website claims that it's a single-slot solution. Even assuming that's not a socket 775 block shown, that thing would probably occupy 2 or three extra slots beneath the graphic card.

And only 26 degrees lower? Already hearing better results from the HR-3.

[bigslappy]

here a link.... BB has the VisionTek 4870 @ $ 299 free shipping !



http://www.bestbuy.com/site/olspage....=1212192380239

[DarkDisciple]

here a link.... BB has the VisionTek 4870 @ $ 299 free shipping !



http://www.bestbuy.com/site/olspage....=1212192380239

Don't forget, if you live in a state where best buy has stores, they will charge you tax. 305 + 8 in shipping is $313. 299 + 7% sales tax is $319. So if you live in a state without best buy (are there any?) you win.