• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

why is video RAM faster than system RAM?

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

Tazon

Member
Joined
Oct 14, 2003
Location
Argentina
today i was reasoning thus: why is it that newest video cards can reach memory speeds of over 1100mhz in DDR at stock speeds, while system memory is peaking around 500mhz or so, i recently saw an OCZ article saying they had attained highest DDR speed at 772mhz or so (obviously NOT STOCK).
moreover, it would seem the video memory is cheaper than system memory. video memory is not sold separately, but i get this from comparing similar cards with different memory setups.

so why is video memory cheaper and faster??
 
I don't know if it is cheaper, but I think the reason that video cards have used DDR2 and most recently DDR3 is because they can.

System memory is dependent on the CPU, chipset, and motherboard. These three components are all designed and manufactured by different companies, (usually). Look at what had to happen before Intel systems started using DDR2. First of all Intel had to come up with a new processor that would support DDR2, then they had to design a chipset, then they had to get motherboard companies to utilize their new chipset, and finally they had to get memory companies to start making the stuff. That would mean for the average consumer to make the leap to a DDR2 platform he/she would need to upgrade CPU, motherboard, and memory. That is a lot of money to make the switch to different memory.

Companies like Nvidia/ATI on the other hand design their board with whatever technology is available, and just buy all the parts and slap them on the board. They don't have to wait for the rest of the industry to catch up with the technology. They are the industry.
 
Also graphics cards can use a higher speed RAM and be stable because there's no DIMM sockets and compatibility issues to worry about. The GFX manufacturers, and through the reference boards, ATi/Nvidia have complete control over the whole host/memory solution, so it can be highly tuned and optimised. Commodity mobos and commodity DIMMs are all designed down to the lowest common denominator. It means pretty much any components should work with each other, but it ain't gonna be the fastest.

Road Warrior
 
Isn't video card memory like 2ns and system memory like 5-6ns? Woudln't that make a differance?
 
The ns just refers to cycle time -- ie. 1/GHz. It's just a different way of expressing the same thing :)

The main reason is that they use advanced forms of DDR, which ramp up to much higher clock speeds than the classic DDR still being used in systems. DDR2 (which just came out for PCs) can sustain higher MHz much easier than DDR1 because of a redesign. DDR3 can go higher still, resulting in the insane memory speeds you see on high end cards today.

The one problem with DDR2/3 is that they increase latency as well as increase clock speed, so the performance increase may be marginal at best for the lowest of speeds. Once you get into uber-high clocks though, the performance should be a bit better than DDR1.

JigPu
 
JigPu - interesting comment. How much of a latency hit is there in the GDDR3 ships that come w/ the newer V-cards? I'm not really familiar with video memory so forgive my ignorance. Can you set the timings w/ video memory? (cas-tras...etc?)

-Adrayic
 
Same as the gddr2, but I believe most run around 4-5-5-16. Might be higher or lower, depends on card, and how incoherent I am at the moment.
 
Another thing to remember is that single channel systems have 64-bit bus width for system ram, while video cards have 64-bit, 128-bit, 256-bit. That also drasticall increases the bandwidth. Duel channel systems are basically 128-bit. System ram is probbaly not cheaper than video card ram, as the little ram chips themselves are doubtfully going to be magically cheaper just because they are going on a video card. 512mb of pc3200 can be had for 50-70 dollars right now. video cards memory controlles are built into the gPU, so i imagine that has helped them attain higher levels of bandwidth. I'm sure with a memory controller built into the cpu like amd is doing they can attain higher levels of bandwidth on the motherboard. If you are going to have anthing like 1gb of ram that price would go way way up. Its not that the technology doesn't exist, its that it becomes uneconomical for PC's.

just to give you an idea though, pc3200 ram is called 3200 because thats about how many mb /s you'll get with it. pc3200 in single channel mode provides approximately 3.2gb/s bandwidth. now take a 256-bit memory bus and 1ghz ram, like what you find on geforce 6800 Gt. Basically you multiply 3200 X 2.5, that means at 1000mhz 64-bit you'd get about 8000 mb/s memory bandwidth. on a geforce 6800 gt at 1ghz ram, since its 256-bit, you'd multiply that by 4 and end up with 32000mb /s. Thats 10X the performance of pc3200 memory on a single channel system LOL.
 
Last edited:
Back