| | | | | | |

Xeon Silver 4208 vs Ryzen 9 3900


Description
The Silver 4208 is based on Cascade Lake architecture while the 3900 is based on Zen 2.

Using the multithread performance as a reference, the Silver 4208 gets a score of 311.6 k points while the 3900 gets 687.5 k points.

Summarizing, the 3900 is 2.2 times faster than the Silver 4208. To get a proper comparison between both models, take a look to the data shown below.

Specs
CPUID
50657
870f10
Core
Cascade Lake-SP
Matisse
Architecture
Base frecuency
2.1 GHz
3.1 GHz
Boost frecuency
3.2 GHz
4.3 GHz
Socket
LGA 3647
AM4
Cores/Threads
8/16
12/24
TDP
85 W
65 W
Cache L1 (d+i)
8x32+8x32 kB
12x32+12x32 kB
Cache L2
8x1024 kB
12x512 kB
Cache L3
11264 kB
4x16384 kB
Date
April 2019
September 2019
Mean monothread perf.
38.61k points
74.97k points
Mean multithread perf.
311.57k points
687.5k points

Non-optimized benchmark
The benchmark in Mode 0 (FPU) measures cpu performance with non-optimized software. It uses the basic µinstructions from the i386 architecture with the i387 floating point unit. This mode is compatible with all CPUs so it's practical to compare very different CPUs
Monothread
Silver 4208
3900
Test#1 (Integers)
2.59k
4.39k (x1.69)
Test#2 (FP)
9.5k
16.99k (x1.79)
Test#3 (Generic, ZIP)
3.15k
7.74k (x2.46)
Test#1 (Memory)
5.9k
24.57k (x4.17)
TOTAL
21.13k
53.69k (x2.54)

Multithread

Silver 4208

3900
Test#1 (Integers)
21.09k
48.45k (x2.3)
Test#2 (FP)
80.93k
230.81k (x2.85)
Test#3 (Generic, ZIP)
30.76k
122.82k (x3.99)
Test#1 (Memory)
8.51k
54.73k (x6.43)
TOTAL
141.3k
456.8k (x3.23)

AVX2 optimized benchmark
The benchmark in mode III (AVX2), like AVX1, is optimized to used 256 bits registers beside the second version of the Advanced Vector Extensions (AVX). The first AVX2 compatible CPU was released in 2013.
Monothread
Silver 4208
3900
Test#1 (Integers)
16.86k
16.85k (x1)
Test#2 (FP)
12.08k
26.03k (x2.15)
Test#3 (Generic, ZIP)
3.37k
9.54k (x2.83)
Test#1 (Memory)
6.29k
22.55k (x3.58)
TOTAL
38.61k
74.97k (x1.94)

Multithread

Silver 4208

3900
Test#1 (Integers)
150.99k
229.41k (x1.52)
Test#2 (FP)
118.54k
292.81k (x2.47)
Test#3 (Generic, ZIP)
33.44k
128.48k (x3.84)
Test#1 (Memory)
8.6k
36.8k (x4.28)
TOTAL
311.57k
687.5k (x2.21)

Performance/W
Silver 4208
3900
Test#1 (Integers)
1776 points/W
3529 points/W
Test#2 (FP)
1395 points/W
4505 points/W
Test#3 (Generic, ZIP)
393 points/W
1977 points/W
Test#1 (Memory)
101 points/W
566 points/W
TOTAL
3666 points/W
10577 points/W

Performance/GHz
Silver 4208
3900
Test#1 (Integers)
5269 points/GHz
3920 points/GHz
Test#2 (FP)
3776 points/GHz
6054 points/GHz
Test#3 (Generic, ZIP)
1054 points/GHz
2218 points/GHz
Test#1 (Memory)
1966 points/GHz
5243 points/GHz
TOTAL
12065 points/GHz
17435 points/GHz

Monothread performance graph
Monothread performance graphics gives the performance vs time. They are useful to measure the time it takes to the CPU to reach the maximum performance.

Usually, CPU's performance will be steady during these tests but if it has a slow frequency strategy, the first samples will show a lower score.


Test#1 (Integers) [points vs time]

grafica bm.hardlimit.com


Test#2 (FP) [points vs time]

grafica bm.hardlimit.com


Test#3 (Generic, ZIP) [points vs time]

grafica bm.hardlimit.com


Test#1 (Memory) [points vs time]

grafica bm.hardlimit.com

Multithread performance graph
Multithread graphs measure the performance against a heavy load during certain time.

If CPU's TDP doesn't limit the frequency and the machine is properly cooled, performance should remain steady vs time. Otherwise, the performance score will oscillate or decrease over time.


Test#1 (Integers) [points vs time]

grafica bm.hardlimit.com


Test#2 (FP) [points vs time]

grafica bm.hardlimit.com


Test#3 (Generic, ZIP) [points vs time]

grafica bm.hardlimit.com


Test#1 (Memory) [points vs time]

grafica bm.hardlimit.com

Hardlimit Benchmark Central - Ver. 3.11.4