[AMBER] RTX 2080 vs RTX 3060 TI

From: Renato Freitas via AMBER <amber.ambermd.org>
Date: Tue, 28 Feb 2023 18:06:00 -0300

I recently bought an RTX 3060 TI, and ran some benchmarks to compare it
with my old RTX2080. The results are below:

#RTX 2080
JAC_NPT_4fs = 721 ns/day
FactorIX_NPT_4fs = 243 ns/day

#RTX 3060 TI
JAC_NPT_4fs = 740 ns/day
FactorIX_NPT_4fs = 242 ns/day

I wonder why they have roughly the same performance even though RTX3060 TI
has 4864 cores and RTX2080 has 2944 cores.

>From previous discussions on the list (
http://archive.ambermd.org/202009/0392.html), I found that memory
bandwidths are an important factor in performance. But both bandwidths are
exactly the same (448 GB/s) (according to https://www.techpowerup.com/):

https://www.techpowerup.com/gpu-specs/geforce-rtx-3060-ti.c3681
https://www.techpowerup.com/gpu-specs/geforce-rtx-2080.c3224

Looking at the output of the "deviceQuery" (see below) I see that the RTX
3060 TI has 38 multiprocessors vs 46 in the RTX 2080. Could this be the
reason for the almost identical performance?

Thank you,

Renato

########################################################
Device 0: "NVIDIA GeForce RTX 2080"
  CUDA Driver Version / Runtime Version 11.7 / 11.7
  CUDA Capability Major/Minor version number: 7.5
  Total amount of global memory: 7981 MBytes (8368816128
bytes)
  (46) Multiprocessors, ( 64) CUDA Cores/MP: 2944 CUDA Cores
  GPU Max Clock rate: 1710 MHz (1.71 GHz)
  Memory Clock rate: 7000 Mhz
  Memory Bus Width: 256-bit
  L2 Cache Size: 4194304 bytes
  Maximum Texture Dimension Size (x,y,z) 1D=(131072), 2D=(131072,
65536), 3D=(16384, 16384, 16384)
  Maximum Layered 1D Texture Size, (num) layers 1D=(32768), 2048 layers
  Maximum Layered 2D Texture Size, (num) layers 2D=(32768, 32768), 2048
layers
  Total amount of constant memory: 65536 bytes
  Total amount of shared memory per block: 49152 bytes
  Total number of registers available per block: 65536
  Warp size: 32
  Maximum number of threads per multiprocessor: 1024
  Maximum number of threads per block: 1024
  Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
  Max dimension size of a grid size (x,y,z): (2147483647, 65535, 65535)
  Maximum memory pitch: 2147483647 bytes
  Texture alignment: 512 bytes
  Concurrent copy and kernel execution: Yes with 3 copy engine(s)
  Run time limit on kernels: Yes
  Integrated GPU sharing Host Memory: No
  Support host page-locked memory mapping: Yes
  Alignment requirement for Surfaces: Yes
  Device has ECC support: Disabled
  Device supports Unified Addressing (UVA): Yes
  Device supports Compute Preemption: Yes
  Supports Cooperative Kernel Launch: Yes
  Supports MultiDevice Co-op Kernel Launch: Yes
  Device PCI Domain ID / Bus ID / location ID: 0 / 175 / 0
################################################

Device 0: "NVIDIA GeForce RTX 3060 TI"
  CUDA Driver Version / Runtime Version 11.7 / 11.7
  CUDA Capability Major/Minor version number: 8.6
  Total amount of global memory: 7981 MBytes (8368357376
bytes)
  (38) Multiprocessors, (128) CUDA Cores/MP: 4864 CUDA Cores
  GPU Max Clock rate: 1665 MHz (1.66 GHz)
  Memory Clock rate: 9501 Mhz
  Memory Bus Width: 256-bit
  L2 Cache Size: 3145728 bytes
  Maximum Texture Dimension Size (x,y,z) 1D=(131072), 2D=(131072,
65536), 3D=(16384, 16384, 16384)
  Maximum Layered 1D Texture Size, (num) layers 1D=(32768), 2048 layers
  Maximum Layered 2D Texture Size, (num) layers 2D=(32768, 32768), 2048
layers
  Total amount of constant memory: 65536 bytes
  Total amount of shared memory per block: 49152 bytes
  Total number of registers available per block: 65536
  Warp size: 32
  Maximum number of threads per multiprocessor: 1536
  Maximum number of threads per block: 1024
  Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
  Max dimension size of a grid size (x,y,z): (2147483647, 65535, 65535)
  Maximum memory pitch: 2147483647 bytes
  Texture alignment: 512 bytes
  Concurrent copy and kernel execution: Yes with 2 copy engine(s)
  Run time limit on kernels: Yes
  Integrated GPU sharing Host Memory: No
  Support host page-locked memory mapping: Yes
  Alignment requirement for Surfaces: Yes
  Device has ECC support: Disabled
  Device supports Unified Addressing (UVA): Yes
  Device supports Compute Preemption: Yes
  Supports Cooperative Kernel Launch: Yes
  Supports MultiDevice Co-op Kernel Launch: Yes
  Device PCI Domain ID / Bus ID / location ID: 0 / 175 / 0
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Feb 28 2023 - 13:30:02 PST
Custom Search