Where is the 512 Bit x 8 = 4096 Memory Video Card?
The GTX 280 had a 512 Bit Bus but short on VRam.
What's the issue with a 512 Bit Video Card?
The GTX 280 had a 512 Bit Bus but short on VRam.
What's the issue with a 512 Bit Video Card?
What's the issue with a 512 Bit Video Card?
- Thread starter Z15CAM
- Start date
- Status
The higher the bit-count, the more traces are needed on the PCB to run between the GPU itself and the individual RAM modules. These extra traces cost extra, both in raw manufacturing and in engineering hours to make them work right. As such, the bus width is generally kept lower and they instead use higher frequency/better bandwidth memory standards.
So it's not a Bus issue but the Cost of Memory?
Samsung can produce fantastic memory look at the MV-3V4G3D-US_DDR3 sticks, where are them sticks now. It seems ONLY NSA and Specific Gov't Agencies can obtain it.
Monopoly.
Samsung can produce fantastic memory look at the MV-3V4G3D-US_DDR3 sticks, where are them sticks now. It seems ONLY NSA and Specific Gov't Agencies can obtain it.
Monopoly.
Last edited:
ViRGE
Elite Member, Moderator Emeritus
- Oct 9, 1999
- 31,516
- 167
- 106
It also requires additional RAM chips (16/32 for a 512bit bus), which becomes another element that keeps costs higher in the long run.
It's a cost issue period. The larger memory bus makes several aspects of the card more expensive, which is why it is only done infrequently (typically when there is no other choice).
I believe them cards Exist: but - When will we see them?
In reference to Cost of Ram Issues: I bought 16GB of Samsung MV-3V4G3D-US_DDR3 through Newegg.ca for $96 then it simply DISAPPEARED. Now it's worth 4X's the Price from Private Vendor's.
I'm stating the Cost Factor of FAST Ram is not the Issue, Why because it can be Mfgr'd on a Large Scale To Day.
In reference to Cost of Ram Issues: I bought 16GB of Samsung MV-3V4G3D-US_DDR3 through Newegg.ca for $96 then it simply DISAPPEARED. Now it's worth 4X's the Price from Private Vendor's.
I'm stating the Cost Factor of FAST Ram is not the Issue, Why because it can be Mfgr'd on a Large Scale To Day.
Last edited:
ViRGE
Elite Member, Moderator Emeritus
- Oct 9, 1999
- 31,516
- 167
- 106
You can't compare DDR3 and GDDR5. The latter is much faster, but it's also bigger and more expensive. The cost of RAM is a big issue for video cards, as would be the costs of implementing the wider memory bus.
brybir
Senior member
- Jun 18, 2009
- 241
- 0
- 0
aside from the cost issue, which is, IMO, the primary reason you don't see it often, it does present additional technical hurdles for engineers to overcome (just a few examples):
1. It is difficult to create a 512 bit width bus that maintains the same frequency and clock as a smaller bus. For example, in a 256 bit bus, generally speaking you are seeing four 64 bit being utilized. Each of these discrete 64 bit memory channels has to operate within the parameters needed for the card or the card may be rejected. By jumping to a 512 bit bus, you are talking eight 64-bit being utilized, and each of those 8 memory channels needs to operate at spec or may be rejected.
2. Additional trace connections and other connections in the memory controllers increases the potential for errors due to interference or other issues to cause problems. This relates to issue #1 above, more connections and components, more points of potential failure and potentially higher warranty claims etc.
So, at the end of the day the 512 bit bus presents additional technical complexity which in turn leads to increased costs. The engineers then need to sit down and see if the extra complexity and therefore cost is worth it for what they are trying to do.
With the increased frequency ramps of GDDR5 and other neat technical tricks, AMD and Nvidia have been able to drastically scale the frequency of their memory subsystems up over time to keep up with the demands of ever larger resolutions and more detailed and complex games.
1. It is difficult to create a 512 bit width bus that maintains the same frequency and clock as a smaller bus. For example, in a 256 bit bus, generally speaking you are seeing four 64 bit being utilized. Each of these discrete 64 bit memory channels has to operate within the parameters needed for the card or the card may be rejected. By jumping to a 512 bit bus, you are talking eight 64-bit being utilized, and each of those 8 memory channels needs to operate at spec or may be rejected.
2. Additional trace connections and other connections in the memory controllers increases the potential for errors due to interference or other issues to cause problems. This relates to issue #1 above, more connections and components, more points of potential failure and potentially higher warranty claims etc.
So, at the end of the day the 512 bit bus presents additional technical complexity which in turn leads to increased costs. The engineers then need to sit down and see if the extra complexity and therefore cost is worth it for what they are trying to do.
With the increased frequency ramps of GDDR5 and other neat technical tricks, AMD and Nvidia have been able to drastically scale the frequency of their memory subsystems up over time to keep up with the demands of ever larger resolutions and more detailed and complex games.
What ARE you blathering about?
That RAM has a frequency of 1600mhz, my 2 gen old video card has a frequency of 5500mhz............
I run an 12GB Ram Disk (Preferable 4GB) Ramdisk Image with out affecting Win7 [64} Ultimate leaving just 4GB's of Samsung MV-3V4G3D-US_DDR3 running at 1867 Mhz 1.35v 9-9-9-24-1T 1.35 volts which blows Aida x's [64] Benchmarks with any comparison to any other processor.
My Processor is an I7 2700K topping at 4600 24/7 Mhz's - Doesn't meat I can give you a 5.2 Ghz Bench.
How about a MSI GTX 780 FozerÉ
Is it worth the PriceÉ
What do you want me to TEST.
Video Cards are the least of my concern except when it comes to Video Compilation.
My Processor is an I7 2700K topping at 4600 24/7 Mhz's - Doesn't meat I can give you a 5.2 Ghz Bench.
How about a MSI GTX 780 FozerÉ
Is it worth the PriceÉ
What do you want me to TEST.
Video Cards are the least of my concern except when it comes to Video Compilation.
Last edited:
My Key Board [redacted] at Times but I know every one of us know what we're talking about.
Profanity isn't allowed in the technical forums (and no, I don't know what that post has to do with VRAM, either).
-- stahlhart
Last edited by a moderator:
Oookay... so you are posting in the Video Card forum because?
All but pretty low end video cards have significantly higher frequency and bus width than even the highest end DDR3 system memory since they are using GDDR5. For example a mid-level enthusiast card like a GTX760 is 6 GHz @ 256 bit width = 6 * 256 / 8 = 192 GB per sec, while 2000 MHz DDR3 dual channel is only 128 bit width for 2GHz * 128 / 8 = 32 GB per sec. So a mid level video card can easily be 5x higher throughput than FAST desktop memory. The top of the line GPUs are more like 10x faster. There is absolutely no comparison to system memory. Video card memory is both faster and wider.
A GPU or CPU requires an individual trace for each bit. That's a ball on a BGA, pin or landing pad on a socketed CPU, etc... it's also die space in the form of pads that connect to whatever kind of package is used. So, memory bus width requires low density silicon to make all the electrical connections. Wider = more silicon = more expensive.
The die space required is also on the outer edge of the silicon chip. So a small processor with a wide memory bus might require dead space just to get enough linear edge for the pads to make connection to the packaging... but foundries don't like leaving dead space, so they fill that space with stuff that allows them to charge more for the silicon. So wide memory bus goes hand in hand with more cores / SP / EUs whatever the processing units are called on whatever kind of processor you're talking about.
Last edited:
BrightCandle
Diamond Member
- Mar 15, 2007
- 4,762
- 0
- 76
All good points made so far, its also worth adding that every bit is also a pin on the GPU which not only adds cost but becomes a manufacturing difficulty. There is only so big you can go or so small with the pins before you can't get a reliable chip mount.
Outside of the GPU the trace lines have to near to the same length to ensure they meet their frequency standards and that is going to require more layers on the PCB which costs a lot more money.
When it comes to spending the pin budget there are other important things like power and PCI-E connections and such and memory is given as much as possible of what remains after the critical parts are done. But also you really really don't want more layers on your PCB if your trying to do this on a budget where people will buy it. 512 bit is just plain expensive and problematic.
Outside of the GPU the trace lines have to near to the same length to ensure they meet their frequency standards and that is going to require more layers on the PCB which costs a lot more money.
When it comes to spending the pin budget there are other important things like power and PCI-E connections and such and memory is given as much as possible of what remains after the critical parts are done. But also you really really don't want more layers on your PCB if your trying to do this on a budget where people will buy it. 512 bit is just plain expensive and problematic.
I believe a 512 Bit 4096 VRam GDDR5 Video Card is very much within our reach - Why don't we have it.
Granted a $500 GeForce GTX 280 512 Bit/1024 VRam was rather expensive back in it's time but you can buy it for $30 today.
So the Price of VRam is the Culprit?
They did it 10 years ago.
Granted a $500 GeForce GTX 280 512 Bit/1024 VRam was rather expensive back in it's time but you can buy it for $30 today.
So the Price of VRam is the Culprit?
Last edited:
It's got nothing to do with technical feasibility; it's simply uneconomical.
If you want a card with a 512 bit bus, go find 100,000 other friends that would be willing to put the money in up front, and you'll get one in about 2-3 years.
If you want a card with a 512 bit bus, go find 100,000 other friends that would be willing to put the money in up front, and you'll get one in about 2-3 years.
Sorry, but this sounds like threads asking why a $50 card with 4 GB of cheap slow DDR3 is worse than a $100 card with with 2 GB of fast GDDR5. Isn't 4 GB always better than 2 GB?!
"More bits" is not by itself better than less bits. A 128-bit CPU would not be better for gaming than a 64-bit CPU, even though "it's got double the bits man!!!"
Unless money really is no object and you're happy to pay $1,000+ for a video card (actually 4 of them in quad SLI, but what's $4,000 between friends) you want the best performance for your dollar.
AMD and nvidia design cards to work with the most efficient mix of memory bandwidth, PCB cost, RAM cost and GPU power to fit within the limits of a price point.
If a card design offers 30 fps instead of 60 because AMD spent their budget on 512-bitness for the memory interface instead of a better GPU, you'd buy a 60 fps nvidia card instead, even though it's only 256 or 384 bits.
Part of my argument is WTF happened to Samsung MV-3V4G3D-US DDr3 let alone GDDR 5, really what's the difference between MB DDR3 and GDDR 5 when you can run DDR3 at 2133 Mhz, 9-9-9-24-1T @ 1.5 volts today and can't get DDR5 above 1250 MHz's in a Video Card?
Gotta say I loved the old Hynix SSD 9800 XT's in a Slot 1 running a Taulatin 1.3 Celron, WinSE at 1800 Ghz - but them days are gone.
Now I'm running an IB between 4.6to 5.2 Ghz's - Take it or leave it.
Gotta say I loved the old Hynix SSD 9800 XT's in a Slot 1 running a Taulatin 1.3 Celron, WinSE at 1800 Ghz - but them days are gone.
Now I'm running an IB between 4.6to 5.2 Ghz's - Take it or leave it.
Last edited:
NO - But give me the link - LOL
Alright, we've had enough nonsense for one evening. If you wish to discuss DDR3 memory, take it to the Memory subforum
-ViRGE
Last edited by a moderator:
- Status
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 24K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 19K
-
-
-
Facebook
Twitter