512MB of GDDR5
At what point does 512MB become a hindrance? I know the 4870 can beat the GTX 260 with 896MB, but will performance continue to scale with speed increases absent a corresponding increase in frame buffer?
"ATI also gave each memory controller hub access to its own L2 cache which will further increase the data transfer speeds to and from the memory. This was done in order to take advantage of the massive bandwidth potential that comes with the implementation of GDDR5 onto some R770 cards. So, even though it looks like the HD4850 and HD4870 “only” have a 256-bit interface, because of the bandwidth afforded by this new memory design, it actually acts like a 512-bit interface." --www.hardwarecanucks.com
---
"The major difference is that the HD 4870 uses GDDR5 while the HD 4850 uses GDDR3. Certainly the biggest benefit of GDDR5 is that is offers twice the memory bandwidth than GDDR3 at the same clock."
"Please note that GDDR5 offers twice the bandwidth per pin at the same clock than GDDR3/4. So a 256-bit GDDR5 card has the same bandwidth as a 512-bit GDDR3 card at the same clock." --www.techpowerup.com
---
"That memory bandwidth perfection is due to the Radeon HD 4870 using GDDR5 memory instead of GDDR3, obviously. Though the real clock frequency is 900 MHz, the outcome in effective bandwidth is the sustained data rate x4. See, the memory frequency is double that of double data rate. This will give the 4870 an astounding 115.2GB/s at 3600 MHz, while still being on the 256-bit memory bus. That in fact is more memory bandwidth than the GeForce GTX 260 (111.9 GB/s) which has a much wider 448-bit memory bus, but uses GDDR3."
"There are some distinct advantages to be found for GDDR5 memory. It's has much higher frequency based memory versus tight timings. In the end this gives the Radeon HD 4870 a performance boost as GDDR5 memory will leverage overall peak bandwidth to a theoretical (roughly) 3.6 Gbps. And that's just crazy fast (GDDR3 on 4850 = 2.0 Gbps)." --www.guru3d.com
---
"The 4870's core clock is 750MHz, and even more importantly, it's paired up with 512MB of GDDR5 memory. The base clock on that memory is 900MHz, but it transfers data at a rate of 3600MT/s, which means the 4870's peak memory bandwidth is nearly twice that of the 4850." --www.techreport.com
---
"The move to GDDR5 is an important one for AMD as it helps to provided all 800 shader processors with enough bandwidth in order to compete with the new 512-bit memory controller that NVIDIA used on the GT200. Providing over 115 GB/sec of bandwidth on a 256-bit wide bus is pretty impressive considering the GT200, using a 512-bit bus, can push about 141 GB/sec of bandwidth." --www.pcper.com
---
"...but the use of GDDR5 boosts memory bandwidth by 80% [versus the HD 4850]. The more bandwidth-limited a particular game and resolution setting is, the better the Radeon HD 4870 should do." --www.extremetech.com
---
"GDDR5 introduces features and functions that go beyond previous GDDR standards and enables GDDR5 to operate at data rates up to 6 Gbps, three times the performance of todays high speed GDDR3. The memory ICs feature adaptive power management, error compensation, adaptive interface timings and date eye optimization. The ICs being used are part number IDGV1G-05A1F1C-40X, which are entry level GDDR5 chips and are rated for just 4 Gbps. ATI has them clocked at just 3.6 Gbps, so these should have some overclocking head room left in them." --www.legitreviews.com
---
"ATI's been keen to pack as much power into a small a die space as possible, where each die's size directly reflects upon the cost structure, hence the continued use of a 256-bit memory controller. Doing so inexorably leads to bandwidth implications that are partially offset by using the latest technology, GDDR5." --www.hexus.net
---
"So, what's going to be different with GDDR5, is it just another marketing name for some further voltage and small under the hood changes? In a word: No.
"After playing a big part in GDDR4's development in the ATI days, it's not surprising to see GDDR5 also trumpeted as largely the baby of AMD who led both the development efforts and making sure the standard was adopted by both JEDEC and memory manufacturers. GDDR5 has been in development for around four years now, with around a year of internal development at ATI followed by a three year process of working with JEDEC to bash out the specification into a ratified standard, bringing us to today and the launch of the first graphics board sporting this new memory, the Radeon HD 4870.
"If there are three words you need to take home from this particular section of today's article, they are - Bandwidth per pin. If that's too tricky to remember, then make those three words - Bandwidth, bandwidth, bandwidth. While there are myriad improvements to GDDR5 over its predecessors, the one that will send a shiver down your spine is the memory bandwidth improvements offered by the standard.
"We're all used to taking the clock speed of GDDR memory on a graphics board and doubling it to glean its effective data rate, as data is sent on both the rising and falling edge of each clock cycle. However, with GDDR5 you can calculate its effective data rate by doubling the memory's clock speed... Then doubling it again.
"How has this been achieved? Quite simply, by increasing the amount of bandwidth per pin for GDDR5, thanks largely to a quarter data rate command clock. While GDDR4 topped out at offering around 4GB/second per pin, GDDR5 should be able to scale all the way up to offering 8GB/second per pin, and in essence the use of GDDR5 on a 256-bit memory bus should offer approximately equivalent bandwidth to pairing GDDR3 with a 512-bit wide bus.
"Away from (although partly to help with the enabling of) memory bandwidth, GDDR5 also offers myriad other improvements, with particular highlights including the ability of modules to 'train' themselves adaptively to adjust timings and so on dependant on the routing and the like of the device to which it is connected. GDDR5 modules also feature error detection, and an on-chip thermal monitor to keep an eye on temperatures.
"The aforementioned 'training' capability also removes many of the difficulties regarding routing traces on the PCB between the GPU core and DRAM - As each module trains itself on boot-up and then monitors and adjusts as necessary, it is designed to handle any mismatches or oddities in the routing between devices. In short, this makes PCB design for manufacturers using GDDR5 on their graphics devices much, much easier.
"While voltages are increased from GDDR4, they still average out at around 30% lower than the voltages required for GDDR3 - For example, while a GDDR3 module with 2GB/second per pin of bandwidth requires 2 Volts, an otherwise similar GDDR5 module which offers 4GB/second per pin only requires 1.5 Volts.
"Lastly, GDDR5 supports the creation of memory modules with 16 DRAM banks, allowing for twice as much memory capacity to be squeezed onto each module in theory, which could again potentially simplify board designs.
"Whichever way you look at it, GDDR5 looks set to be a sure-fire winner as time goes on and availability becomes more commonplace, grabbing the attention of the market where GDDR4 failed." --www.elitebastards.com
---
In my estimation the answer is "yes", as:
- Neither the frame buffer nor bandwidth currently serve as a bottleneck (sometimes even at resolutions beyond 1920x1200).
- The GDDR5 frequencies that are being exercised with the introduction of the HD 4870 are conservative. GDDR5 is stated to be able to scale to nearly twice the current frequencies, in the future.
