Computers, small and large, rely on volatile memory (DRAM) for data processing before saving the results to storage. The memory type varies across devices, often based on the size and power requirements:
DDR5/DDR4 memory is commonly used in high-performance laptops, desktops, and workstations.
LPDDR5/LPDDR4 is the go-to choice for smartphones and low-power notebooks.
GDDR6/GDDR7 is used in graphics cards.
Desktop RAM vs. Mobile RAM vs. GPU RAM
Instead of these different standards, why don’t we use a common, uniform solution across all computers? As is usually the case in computer hardware, this diversification boils down to the power and performance requirements of different devices:
Desktop PCs and laptops have higher memory requirements, often with minimal latency as their processors process data sequentially. Power efficiency is less of a concern, so the memory uses higher voltages and power.
Mobility devices like smartphones and convertibles are optimized for power efficiency. Hence, the memory runs at lower voltages for longer battery cycles. Bandwidth is comparable to desktops, but memory latency is relatively higher.
GPUs are massively parallel, relying on their high bandwidths to mask the higher processing latency. Hence, graphics memory has the highest bandwidth and power. The latency is also higher.
DRAM & Memory FAQs
Burst-length: When the CPU or cache requests new data, the address is sent to the memory module and the required row and column are located. The entire column is sent across the memory bus in bursts (words):
DDR4 has a burst length of 8. Each word is 64 bits (per channel), so up to 128 bytes or 16-bits of data is sent per burst.
DDR5 upgrades burst length to 16, but divides the bus into 32-bit subchannels. Consequently, there are two 16-bit bursts per DIMM.
Prefetch is the amount of data fetched internally by the DRAM cores per access (read/write) before it’s explicitly requested by the CPU or the cache. The prefetched data allows for faster memory accesses across the bus when needed. DDR5, GDDR6/GDDR7, and LPDDR5 have a prefetch depth of 16n (16 words) per channel.
DDR4 was the de facto memory standard for PCs for the last decade, but has largely been replaced by DDR5 memory. It supports larger memory capacity, higher bandwidth, and is more efficient:
DDR4
DDR5
Die Density
Up to 16 Gb
Up to 64 Gb
Data Rate/Pin
1600 – 3200 MT/s
4,800 – 8400 MT/s
Bandwidth (64-bit)
Up to 25.6 GB/s
Up to 67.2 GB/s
Voltage (VDD)
1.2v
1.1v
Burst Length
8
16
Prefetch
8n
16n
Channels/DIMM
64-bit x1
32-bit x2
DDR5 is twice as fast as DDR4 and quadruples the per-die memory density. It’s more power efficient, supports on-die ECC, and splits the larger 64-bit channel into two independent 32-bit subchannels.
Mobile RAM: LPDDR4 vs. LPDDR5
LPDDR4 and LPDDR5 feature similar data rates as their PC counterparts, with half the bus width. Consequently, the overall bandwidth is halved. The lower voltage offers substantial efficiency gains.
LPDDR4 | LPDDR4X
LPDDR5 | LPDDR5X
Data Rate/Pin
3200 – 4266 MT/s
5500 – 8533 MT/s
Bandwidth (32-bit)
Up to 17 GB/s
Up to 34.1 GB/s
Voltage
1.1V (Core), 0.6V (I/O)
1.05V (Core), 0.5V (I/O)
Burst Length
16
16
Prefetch
16
16
Channels
16-bit x2
16-bit x2
GPU RAM: GDDR6 vs. GDDR7
By 极客湾Geekerwan, CC BY 3.0, https://commons.wikimedia.org/w/index.php?curid=158558390
Graphics memory is all about the bandwidth. This allows for a massively parallel processor that frequently switches between different task groups to maximize utilization. While DDR and LPDDR memory transmit 2 bits per cycle, GDDR7 uses PAM 3 signaling to transmit 3 bits.
GDDR7
GDDR6X
GDDR6
Die Density
Up to 64 Gb
Up to 32 Gb
Up to 32 Gb
Data Rate/Pin
up to 48 Gbps
up to 24 Gbps
up to 20 Gbps
Bandwidth (384-bit)
1.5 TB/s (28 Gbps)
1.01 TB/s (21 Gbps)
0.96 TB/s (20 Gbps)
Voltage
1.2 V
1.35 V
1.35 V
Signaling
PAM-3
PAM-4
NRZ (PAM-2)
Burst Length
16
16
16
Prefetch
32n
16n
16n
Channels
16-bit x2/10-bit x4
16-bit x2
16-bit x2
GDDR7 delivers data rates of up to 28 Gbps on high-end GPUs, translating to a peak bandwidth of 1.5 TB/s for a 384-bit bus.
GDDR6 uses 16n prefetch and high clock speeds (~20 Gbps) to achieve bandwidths of nearly 1 TB/s.
GDDR6X uses PAM-4 signaling (4 bits per cycle) to achieve a slightly higher bandwidth despite lower internal clocks.
Processors, PC gaming, and the past. I have been writing about computer hardware for over seven years with more than 5000 published articles. Started off during engineering college and haven't stopped since. Find me at HardwareTimes and PC Opset.