GIGABYTE MS03-CE0 + Intel Xeon Emerald Rapids EMR-SP

[JosefHrib]

I want to share my experience with Intel Xeon Emerald Rapids EMR-SP A0 and motherboard GIGABYTE MS03-CE0 with chipset C741.

Specs:
CPU: Intel Xeon Emerald Rapids EMR-SP Q2SR A0 64C/128T 1.7-4.0GHz 320MB L3 350W similar to Xeon Platinum 8592+
5120KB L1 (32 KB Instruction Cache per core, 48 KB Data Cache per core) 128MB L2 (2MB per core) 320MB L3 (5MB per core)
MB: GIGABYTE MS03-CE0 LGA4677 8Channel (BIOS R07->R08->R10->R11->R12), ME Firmware 18:6.1.4.47, BMC Firmware 13.06.10->13.06.13)
RAM: 256GB 8x32GB Micron 5600MHz 2Rx8 CL46 DDR5 RDIMM ECC REG MTC20F2085S1RC56BR
GPU: PALIT GeForce RTX 4070 Ti SUPER GameRock OmniBlack 16GB GDDR6X 256bit AD102
SSD: 4TB Samsung 990 Pro NVMe SSD
PSU: Seasonic PRIME TX 1600W ATX3 SSR-1600TR2
CPU Cooler: Noctua U12S DX-4677 2x Noctua NF-A12x25-FLX
Fans Hub: Noctua NA-FH1
Case Fans: 4x Phanteks M25G2-140 White D-RGB
Case: Phanteks Eclipse G400A White
Audio Device: -
WiFi Device: GIGABYTE GC-WIFI7 Revision 1.0 Qualcomm QCNCM865 WiFi 7
Bluetooth Device: Vention VWS-F97 Bluetooth 5.4
Monitor: Philips 32M2N6800M/00 Evnia 32" 4K 3840x2160 144Hz IPS MiniLED HDR1000 Ambiglow

BIOS modification required (noACM).

GIGABYTE website: BIOS link
GIGABYTE website: BMC Firmware link

After a installed Windows 11 Pro x64, I installed the following drivers:
GIGABYTE website: server_driver_intel_chipset_10.1.19485.8386 link
ASPEED website: v11503_windows link
INTEL website: QAT2.0.W.2.5.0-0007 link
INTEL website: Wired Driver link
NVIDIA website: GeForce Driver link
SAMSUNG website: Magician Software link

In BIOS I set:
Security - Secure Boot: Enabled
Security - Secure Boot Mode: Standart
Server Mgmt - System Event Log - SEL Components: Disabled
Chipset - Runtime Error Logging - System Errors: Disable
Chipset - Miscellaneous Configuration - Active Video: PCIE Device
Advanced - PCI Subsystem Settings - Re-Size BAR Support: Enabled
Chipset - Advanced Power Management Configuration - SOCKET RAPL Config - PL1 Time Window: 128 sec

I also tried the option:
Chipset - Power Policy - Power Policy Quick Settings: Best Performance
Chipset - Power Policy - CPU C6 report: Disable
Chipset - Power Policy - Enhanced Halt State (C1E): Disable
Chipset - Power Policy - Package C State: C0/C1 state

In Windows I set:
High Performance Mode

In Device Manager is all fine, no problem.

1.Tested CPU: Q2SR
Intel Xeon Emerald Rapids EMR-SP Q2SR A0 64C/128T 1.7-4.0GHz 320MB L3 350W similar to Xeon Platinum 8592+
Base 1.7GHz All core turbo 2.6GHz Max turbo 4.0GHz
With Windows High Performance Mode stable frequency +- 2.2-2.4GHz, under load all cores +- 2.6GHz, under load single core +- 3.9-4.0GHz, idle +- 2.2-2.3GHz (With Windows Balanced Mode Idle 800MHz)

Results with VMX Hypervision
Cinebench 2024.1.0 - Multi 3926 (4.image)
Cinebench R23.200 330542 - Multi 72533 (5.image)
CPU-Z 2.14.0 x64 - CPU Single Thread 649,7 (6.image)
CPU-Z 2.14.0 x64 - CPU Multi Thread 34455,1 (6.image)
AIDA64 7.60.7300 - Memory Read 282,21 GB/s Write 222,47 GB/s Copy 248,21 GB/s 6.image)

Results without VMX Hypervision
Cinebench 2024.1.0 - Multi 4045 (9.image)
Cinebench R23.200 330542 - Multi 73405 (8.image)
CPU-Z 2.14.0 x64 - CPU Single Thread 649,7 (7.image)
CPU-Z 2.14.0 x64 - CPU Multi Thread 34455,1 (7.image)
AIDA64 7.60.7300 - Memory Read 313,19 GB/s Write 251,39 GB/s Copy 278,45 GB/s (7.image)

 

[JosefHrib]

Added results from Cinebench R20,R23,2024.

 

[CyklonDX]

How are the thermals? powerdraw on mix load/full load. How about primegrid test?

 

[RolloZ170]

light better scores beause BIOS: PL1 Time Window = 128 seconds

 

[JosefHrib]

light better scores beause BIOS: PL1 Time Window = 128 seconds
View attachment 40171

Retested with PL1 128sec.

PL1 1sec - Cinebench R20 Multi 27077
PL1 1sec - Cinebench R23 Multi 69521
PL1 1sec - Cinebench 2024 Multi 3853
PL1 128sec - Cinebench R20 Multi 27759
PL1 128sec - Cinebench R23 Multi 72066
PL1 128sec - Cinebench 2024 Multi 3926

 

[JosefHrib]

Added results from 3DMark.

 

[arabus]

Nice!
Make please Aida memory benchmark.
Thank you!

 

[JosefHrib]

Nice!
Make please Aida memory benchmark.
Thank you!

Added results from AIDA64 Memory Benchmark.
Just a warning. Now I have only 4 sticks 32GB Micron DDR5 5600MHz CL46. So benchmark is in 4Channel. Now I cannot create benchmark in 8Channel. Additional 4 sticks I will have next year.

 

[arabus]

Many thanks.
Ok.I have lot of memory stick and coming moore.
Today i pay the processor and maybe 4 weeks.

 

[arabus]

light better scores beause BIOS: PL1 Time Window = 128 seconds
View attachment 40171

What about Q3W7?
This is really?

 

[RolloZ170]

What about Q3W7?

what should be with it ?
Q3W7 is 8592+ QS
buy and test it.

 

[arabus]

I think this.
I pay 130 Eur.
Test coming soon.

 

[arabus]

Lotto.

 

[Kizune]

Impressive! How about a single core performance? Do you have to do the core disabling trick to make it behave as it was with QYFS?

 

[RolloZ170]

Do you have to do the core disabling trick to make it behave as it was with QYFS?

what core disabling ? you need just some cores without load.
do you refering to this ?

[Operating Points]
CPU LFM (Minimum):    800.0 MHz = 8 x 100.0 MHz
CPU HFM (Base):    1700.0 MHz = 17 x 100.0 MHz
CPU Turbo Max:    4000.0 MHz = 40 x 100.0 MHz [Unlocked]
Turbo Ratio Limits - IA/SSE, Fused:_____40x (1-32c), 34x (33-48c), 27x (49-58c), 26x (59-64c)
Turbo Ratio Limits - IA/SSE, Resolved:__40x (1-32c), 34x (33-48c), 27x (49-58c), 26x (59-64c)
Turbo Ratio Limits - AVX2, Fused:_______38x (1-32c), 32x (33-48c), 26x (49-54c), 25x (55-64c)
Turbo Ratio Limits - AVX2, Resolved:____38x (1-32c), 32x (33-48c), 26x (49-54c), 25x (55-64c)
Turbo Ratio Limits - AVX-512, Fused:____35x (1-32c), 29x (33-48c), 25x (49-54c), 24x (55-64c)
Turbo Ratio Limits - AVX-512, Resolved:_35x (1-32c), 29x (33-48c), 25x (49-54c), 24x (55-64c)
Turbo Ratio Limits - TMUL, Fused:_______35x (1-32c), 29x (33-48c), 23x (49-54c), 22x (55-64c)
Turbo Ratio Limits - TMUL, Resolved:___ 35x (1-32c), 29x (33-48c), 23x (49-54c), 22x (55-64c)

 

[Kizune]

what core disabling ? you need just some cores without load.
do you refering to this ?

Indirectly. The problem was that I was not able to do tasks that require high single threaded performance reliably well unless I disable half of the cores. That’s mostly because windows scheduler loves spreading even low priority tasks to maximum number of cores so the single core turbo frequency was pretty low resulting, for example, in 3DMark performance dropping by 30-40 percent. Disabling half of the cores helped a lot in this regard.

 

[RolloZ170]

That’s mostly because windows scheduler loves spreading even low priority tasks to maximum number of cores so the single core turbo frequency was pretty low resulting, for example, in 3DMark performance dropping by 30-40 percent

if it is windows problem is will be the same. but e.g. CBR23 doesn't do(much) threadhopping in single thread bench.

 

[RolloZ170]

That’s mostly because windows scheduler loves spreading even low priority tasks to maximum number of cores so the single core turbo frequency was pretty low resulting, for example, in 3DMark performance dropping by 30-40 percent. Disabling half of the cores helped a lot in this regard.

Q2SR: cbr23 and cpu-z single core clock while running
edit: energy profile: high performance power plan
processor power management: minimum processor frq. 50%

 

[Kizune]

Q2SR: cbr23 and cpu-z single core clock while running

I see. That's a good news. Although for some reason the single core clock is the same as for the Sapphire Rapids Gold CPU and the multicore is like 20% more despite having twice the cores number. I thought we would see some IPC improvements on the Emerald Rapids. But that's still fine, at least you don't have to play with core disabling to get the decent single core performance.

 

[JosefHrib]

Added results from CPU-Z.