• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

384-bit Memory Interface for HD5000-series?

Status
Not open for further replies.
Looks like a monster.

If priced right this card will be an epic success for ATI.
 
Well after the 2900 I think they may be reluctant to go down the 512 route again although I thought the half increments were nvidia's way.
 
do video cards really need 512bit gddr5? that is a ton of bandwidth right there. ~300gb/s(i think)
 
do video cards really need 512bit gddr5? that is a ton of bandwidth right there. ~300gb/s(i think)

I suspect not. GDDR5 is kind of designed to make wide buses unnecessary (in my opinion, I mean). This is probably a sensible reaction to Nvidia announcing a 512-bit bus for its GT300 cards.
 
do video cards really need 512bit gddr5? that is a ton of bandwidth right there. ~300gb/s(i think)


The 40nm GPU in this card is going to be amazing. Every single bit of memory bandwidth will be a great asset to this card.

I'm predicting 30,000+ scores on 3dMark06 with a single card. This card is going to be HOT!!! :cool:
 
Last edited:
The 40nm GPU in this card is going to be amazing. Every single bit of memory bandwidth will be a great asset to this card.
Yes, because you can really know that before the card is released. If this card is anything like the current generation, the added memory bandwidth probably won't make much difference.
 
What I think would make a real difference would be if they decoupled the shader clock from the core clock so that they can seriously speed up the shaders. That's the one area where the Radeons are lagging behind Nvidia's cards.
 
What I think would make a real difference would be if they decoupled the shader clock from the core clock so that they can seriously speed up the shaders. That's the one area where the Radeons are lagging behind Nvidia's cards.
Well, that might not be possible with their architecture. I think they would have their shaders run at a much faster speed if they could do it, since it would obviously give some pretty huge increases in shader performance.
 
Seriously I think a 256 bit on GDDR5 let alone GDDR3 isn't enough these days. I do hope ATI will use a 384 bit or whatever instead of a 256 bit.
 
We will probably be surprised with 512bit ddr5. Going to see what happens :)
 
Well the thing with 384 bit is you're going to have weird memory sizes again, like the current nvidia 2xxx series 896mb and 1792 mb respectively. The nvidia 3xxx series if 512 bit is going to go back to more normal 1gb, 2gb etc. sizes. I guess it doesn't really matter, I just don't particularly like seeing the odd memory sizes.
 
Well the thing with 384 bit is you're going to have weird memory sizes again, like the current nvidia 2xxx series 896mb and 1792 mb respectively.
Actually, the GTX260/GTX295 have 448-bit memory buses. The 8800GTX and 8800Ultra had 384-bit memory buses.
 
Actually, the GTX260/GTX295 have 448-bit memory buses. The 8800GTX and 8800Ultra had 384-bit memory buses.

Really? Hmm so we won't have to see weird memory sizes with 384 bit? It's the 448 bit that requires that or was it a cost cutting measure by Nvidia?
 
Really? Hmm so we won't have to see weird memory sizes with 384 bit? It's the 448 bit that requires that or was it a cost cutting measure by Nvidia?
The 8800GTX and 8800Ultra had 768MB of RAM each, so the memory sizes were still a bit "weird". I don't understand your problem with odd memory sizes though. It's not like it really makes a difference.
 
Really? Hmm so we won't have to see weird memory sizes with 384 bit? It's the 448 bit that requires that or was it a cost cutting measure by Nvidia?
if ATI uses a 384 bit bus then we will certainly see a 5870 with 1536mb of ram and a 5850 with both 1536mb and 768mb of ram.
 
I think an 1152MB card would be more likely than a 1536MB card.
1152? where would you derive that from?1536mb would be the logical number based on the way memory chips are arranged. it would be twice as much as 768mb which makes sense. all the other memory capacities are multiples of their bus widths.

256 bit= 256mb, 512mb, 1024mb and so on
384 bit= 384mb, 768mb, 1536mb and so on
 
Last edited:
Yuck.
I guess it saves em money. 768MB of memory is cheaper than 1GB. And 1.536 is cheaper than 2GB. But it looks.... weird.
 
Yuck.
I guess it saves em money. 768MB of memory is cheaper than 1GB. And 1.536 is cheaper than 2GB. But it looks.... weird.
that memory capacity is not about saving money and its not weird. its just a multiple of the bus width.
 
I'm aware it has to do with the bus width. I'm just saying that if they pack 768 megs of memory on the card that will be cheaper to produce than 1GB and 1536mb should be cheaper for them on the production side than 2GB. It's just math.
 
I'm aware it has to do with the bus width. I'm just saying that if they pack 768 megs of memory on the card that will be cheaper to produce than 1GB and 1536mb should be cheaper for them on the production side than 2GB. It's just math.
but thats NOT why it will have 768mb and 1536mb though. thats like saying it was about money on the gtx260 because they used 896mb instead of 1024mb. it will have it because thats just how it works out based on the architecture. theres no need to make this complicated.
 
1152? where would you derive that from?1536mb would be the logical number based on the way memory chips are arranged. it would be twice as much as 768mb which makes sense. all the other memory capacities are multiples of their bus widths.

256 bit= 256mb, 512mb, 1024mb and so on
384 bit= 384mb, 768mb, 1536mb and so on
1152 is 384*3. 384MB = 384*1, 768MB = 384*2, 1152MB = 384*3, 1536MB = 384*4. It just seems more logical to me for them to go with a memory capacity closer to 1GB, since 1536MB would probably be overkill and would increase the cost of each card for little benefit.
 
1152 is 384*3. 384MB = 384*1, 768MB = 384*2, 1152MB = 384*3, 1536MB = 384*4. It just seems more logical to me for them to go with a memory capacity closer to 1GB, since 1536MB would probably be overkill and would increase the cost of each card for little benefit.
have you ever seen a 128bit card with 384mb ram or a 512bit card with 1536mb? you have been here for years and you should know it doesnt work that way.
 
have you ever seen a 128bit card with 384mb ram? you have been here for years and you should know it doesnt work that way.
Why doesn't it work that way? I don't see anything that would prevent them from using 3 times the bus bandwidth rather than one, two, or four. Just because it hasn't been done before doesn't mean it isn't possible. Hell, before the 8800 series, cards with memory buses that weren't powers of two were also unheard of.
 
Last edited:
Why doesn't it work that way? I don't see anything that would prevent them from using 3 times the bus bandwidth rather than one, two, or four. Just because it hasn't been done before doesn't mean it isn't possible. Hell, before the 8800 series, cards with memory buses that weren't powers of two were also unheard of.
its an even multiple of the memory bus width like I showed you. I am sure someone with more technical finesse can explain it to you better.
 
That is incorrect. Many cards use a memory size equivalent to the bit width of their memory buses, which is not an even multiple.
perhaps I am not using the correct terminology then. obviously showing you the numbers like I did earlier didnt work for you either. you can think whatever you want but you are wrong. no 384bit card will have ever have 1152mb of ram.
 
If you can give me a technical reason for that then I'll believe you. But saying that it's impossible just because it hasn't been done before is not proof of anything.
 
If you can give me a technical reason for that then I'll believe you. But saying that it's impossible just because it hasn't been done before is not proof of anything.
obviously I dont know the exact technical reason and probably only a couple of people on here do. I can guarantee you that it will not be done the way you are thinking.
 
I wasn't joking about the binary bit you know.

From what I understand it is not that nvidia currently uses a 1.5 bus in the 384(256+128) it is that they use a 2+1 sort of system.

The number are as cannondale said not pulled out of nowhere.

They are multiples of 2. Everything in a computer is.

How much ram they use is up to them, the more ram, the higher the bus width or the ram speed has to be to ensure a fluid flow of data because being a parralel system there is only so much paths for data to go through.

That is why 2GB is not always better because you may have all that ram and even if you can fill it you can't out it to use as it is still going through the same bus as 1GB and unless there is room for the data it will sit there waiting.

If I am wrong someone correct me.
 
obviously I dont know the exact technical reason
So then you should stop talking, since you're obviously not qualified to rule anything out. I don't know the technical details either, which is why I'm not saying that it's definitely possible. But I won't say that it's impossible either, unless I'm presented with some evidence to that effect.
 
Wouldn't it be dependant on the video card's memory controller rather than someone's opinion?

For example the x58 motherboards and the i7 support triple channel memory. Thus 3 memory banks of 1 GB chips work on it.

However 3 GB of memory would not work in a dual channel p45 motherboard with a core 2 chip, because those pieces of hardware only support dual channel memory operation.

So unless you're privy to the video memory controller on the new 5000 series, you can't CONCLUSIVELY prove that memory has to be a multiple of 2 even with a 384 bit bus. For all we know the 5000 series COULD support triple channel addressing of the video ram. In which case 1152 would work because, 384x3 = 1152.
 
Hi, 3GB of ram does work in dual channel mode.

You just need 1.5GB per channel. As long as there is equal amounts in both channels then the computer can see it as one and have an effective 128bus width.
 
Hi, 3GB of ram does work in dual channel mode.

You just need 1.5GB per channel. As long as there is equal amounts in both channels then the computer can see it as one and have an effective 128bus width.

Show me some 1.5 GB ram sticks and I'll order em straight away! :rolleyes:
 
So then you should stop talking, since you're obviously not qualified to rule anything out. I don't know the technical details either, which is why I'm not saying that it's definitely possible. But I won't say that it's impossible either, unless I'm presented with some evidence to that effect.
well then you should stop talking also. :p

I can at least show you how it has worked up until now where all you have done is pull a theory out of your lower region. nothing supports what you have theorized at all where as I can look at the history of every card made in the last 5-7 years and see a pattern or even formula if you will. ;)
 
1GB plus a 512 stick.

Yeah that'll work awesome in DUAL channel! :rolleyes:

Besides that is irrelevant.

If the 5000 series can address triple channel video memory, then an 1152 MB memory buffer is logical and possible. 3x384 = 1152.
 
Last edited:
Yes, it does.

Tons of people have done it.

There are 4 ram slots to a usual mobo, with 2 slots per channel. As long as you have 1.5 per channel it works fine.

I remember being a bit confused about it myself when someone explained how they did it a few years ago but if you think about it, it works.
 
Status
Not open for further replies.
Back
Top