Exploring OC options and a viable overclock solution for the Z820

[Storm-Chaser]

I would like to delve into this for however long it may take to reach a satisfactory result.

I have two HP z820 workstations which I would like to overclock. I wanted to check in and see if there was any hope to possibly modify the base clock settings? Or possibly force the turbo multi to be used at all times? I am willing to pay you to find a solution here, because I really love the z820, and I want to keep it around for the last time. But in order to do that I need to bring more horsepower to the table. 

Has anyone been able to get setFSB to work with a z820? What else can be done, short of a bios customization, to enable base clock overclocking on this system? The CPUs obviously have locked multipliers so nothing can be done there. This would only be base clock overclocking. 

[Brian1965]

Can you please post your current Z820 configuration, e.g. CPU(s), GPU, RAM (speed, size/qty of DIMMs), SSD/HDD/M.2 drives, etc.?

You might also want to read this post.

https://h30434.www3.hp.com/t5/Business-PCs-Workstations-and-Point-of-Sale-Systems/z420-z620-overcklo...

[Storm-Chaser]

Here is a quick summary from userbenchmark.com - FYI this is a second gen z820 rig as confirmed by the later boot block date. Also note it has the higher output PSU at 1450 watts. In addition, this z820 has the upgraded factory liquid cooling system in place.

HP Z820 Workstation Performance Results - UserBenchmark

AND a little more specific information for you:

CPU Specs

         -CPU Type 2x 12-Core Intel Xeon E5-2696 v2, 3500 MHz (35 x 100)

         -24 cores / 48 threads in total

         -OEM only processor and has a single core turbo speed of 3.5GHz

         -All core turbo speed of 3.1GHz, 100Mhz higher than Intel's flagship 2600 series chip, the E5 2697v2     

         -This indicates that under heavy, sustained loads, the 2696 v2 is actually intel's best performing chip in the entire 2600              series lineup

Memory Specs

        -64GB Hynix DDR3 (16 x 4GB) PC3-14900 DDR3-1866MHz Registered ECC DIMMs 

       -Configured in 8 Memory channels for maximum bandwidth

       -All 16 memory slots are populated

HDD/SSD storage specs

         -1TB LSI Raid 0 logical volume

         -Disk configuration: 4 SSD RAID 0

         -4 x Intel DC545 series III SSDs (raid 0)

         -1 x SK Hynix Gold 512GB SSD (non raid)

GPU

        -MSI Radeon RX 5700 XT 8GB Gaming X

        -256-Bit GDDR6 14Gbps 

        -2560 Stream Processors

        -Core Clock: 1730MHz

        -Boost Clock: 1980MHz

[Brian1965]

I haven't heard of anyone on this forum who has modified their motherboards base clock speed. All the overclocking on these workstations is done using XTU, (or some other 3rd party overclocking software). Also, only specific E5-16xx CPU's have unlocked multipliers and can be overclocked using this method. There is no way to overclock a dual CPU Xeon system that I know, unless using a motherboard with BIOS overclocking features, e.g. ASUS, ASRock, etc.

You seem to have quite an unusual system configuration, i.e. 48-cores, 64GB RAM, gaming GPU? Optimizing a system for best performance is dependant on what your primary workload will be? e.g. gaming, VM's, video rendering, 3D CAD, etc. What applications do you mainly use?

[OSIRIS777]

I am not sure this will solve your problem but have you tried using the Intel Extreme Tuning Utility?  

[Storm-Chaser]

This z820 is my daily driver and sees a variety of workloads from day to day. In other words it's a hybrid system and wears many hats. Some gaming, some benchmarking, some video rendering and some folding, among other things. I'm a computer hardware enthusiast, so some of the projects I undertake aren't necessarily geared for a specific workload, more of my vision for what I want out of said build or what hardware I want to use or what I feel would go well together. "If you build it they will come" sort of theory, in part. The other z820 is part of my gp/gpu cluster and is paired with two HP ProLiant DL360p servers, The cluster also contains a dell PowerEdge R620 and a Dell PowerEdge C4130 GPU server (which contains four 24GB Tesla K80 GPUs for use with GP/GPU and some deep learning). Keep in mind this is a hobby for me and most of this stuff is not business critical. This is why I feel compelled to move forward with determination to find a viable methodology for overclocking the z820. In other words, I don't have to hold back for practical reasons. 

I have other GPUs for this z820 as well, but for the most part I am happy with keeping the RX 5700 XT in there. Futureproofing, in a sense. And pulls it's own weight, that's for sure. Very robust GPU and does well for itself even outside of gaming. Plus I am a huge AMD GPU fan and as soon as I heard about it's release I knew I had to get one.  The MSI Gaming X is one of the quietest GPUs on the market, both at idle and under full load and in all probability has one of the best heatsink designs available for any 5700 XT on the market at this time. Consequently, the GPU runs stone cold. Which is a must for me because I am pretty heavily GPU dependent when I want to be. 

Unfortunately, XTU does not work with the Xeon E5 2600 series family at all. And there is no hack to force max turbo mode either. I am aware that the CPU has a locked multi, that's why the only viable solution at this point is either setFSB or a custom BIOS so I can do it from a hardware level. That's what I really want to do, to be honest with you. Hardware level overclock is always best. I don't mind paying somebody to do this, but modding BIOS is a lost, dark art and not many people have the skills to pull something like this off. I have already posted a request like this at www.bios-mods.com forum in hopes they could help me with a custom BIOS with an unlocked base clock, but I didn't get any responses. 

That being said I will not be deterred. I will overclock this machine if it's the last thing I do. 

Tonight I am going to go through the PLLs on setFSB to see if I can roll the dice and find one that works. Highly doubt it.

Although, I was able to locate the PLL controller chip on the motherboard. Perhaps you know something about this. Here is the number: I have confirmed this is in fact the clock buffer/pll code that the z820 uses. 

ICS9ZX21901BKLF

I did happen to find some documentation surrounding the PLL itself which may prove helpful at some point. Take a look.

9ZX21901C Datasheet (renesas.com)

[BambiBoomZ]

Storm-Chaser,

Pursuing a high-level performance upgrade will be more successful if there is a focus on the hardware benefits relative to the priority applications.

The emphasis on processor overclocking potential suggests that the single-thread performance is the priority.  However, the subject system comprises 2X 12C @ 2.5 / 3.3GHz processors with locked multipliers, representing contrary indications.

As mentioned, no Xeon E5-26XX may be overclocked using Intel Extreme Tuning Utility  (XTU) , only the E5-1650 v2  E5-1660 v2, and E5-1680 v2.  CPU cooling is key to clock speeds. I would suggest that the highest clock speed obtainable on a z820 would be to use a Xeon E5-1660 v2 liquid cooled by an Alphacool Eiswand external tower cooler having 6X 120mm push /pull fans:

• 1x 360mm XT45 full-copper radiator
• 6x Eiswind fans 1100/700 RPM
• 1x XP3 CPU cooler
• 1x external power supply

Using XTU, overclock the E5-1660 v2 on all 6-cores to 4.6 or 4.7GHz, and set Power Options In Control Panel to High Performance which sustains the Turbo timing.  The all-core clock will 3.8+GHz.

Interestingly, in terms of total clock cycles per unit time (IPC) , the highest Passmark PerformanceTest mark for a 2X Xeon E5-2696 v2  on an HP system is  = 19189, whereas  an Xeon E5-1660 v2 on an ASUS Sabertooth X79, running at 4.9Ghz = 17875. That means that fastest 24C/48TH system is making only +7% of the 6-cores/12 thread system.   Of course, in certain full-threaded applications, the greater number of cores may have an advantage. Still, in my view, if possible, whatever the system, in today's use, there should be a core with a clock speed of 4.0Ghz or higher.  Single thread of course is the absolute ruler of gaming and in higher resolutions there, one needs 5.0Ghz+. On an HP system, the top 1660 v2 mark, running at 4.3GHz = 15240. One of the previous office systems here, z420_2, running at 4.2Ghz = 15129.  In terms of the Single Thread Mark (STM )performance, the Passmark average for  Xeon E5-2696 v2 = 1,712 and for Xeon E5-1660 v2 = 2,146.  The 4.9Ghz E5-1660 v2 system calculates to an STM of about  2,750.

Consider that a dual-CPU system running a single processor will have a slight performance disadvantage due to the, in effect, CPU link that synchronizes the threads between two processors.  The highest Passmark system rating for a z420 is the office's z420_3  Xeon E5-1680 v2 (8C@ 4.3GHz), rated at 6227, whereas the highest rated 2X 12C z820 (E5-2697 v2) = 4487, demonstrating the general principle that 2X CPU's mean no overclocking and more cores means lower clock speeds.

Consider a Xeon E5-1660 v2 and  RTX 3070 or RTX 3080.  For the dual-processor z820- unless the only system priority use is with completely scalar multi-threaded applications, consider changing to 2X E5-2667 v2 (8C @ 3.3/4.0GHz / STM= 2,029) (Xeon E5-2696 v2 STM = 1,712)  There would be a benefit in changing the single CPU system to a z620 chassis.

BambiBoomZ

HP z620_2 (2017) (R7) > Xeon E5-1680 v2 (8C@ 4.3GHz) / z420 Liquid Cooling / 64GB (HP/Samsung 8X 8GB DDR3-1866 ECC registered) / Quadro P2000 5GB _ GTX 1070 Ti 8GB / HP Z Turbo Drive M.2 256GB AHCI + Samsung 970 EVO M.2 NVMe 500GB + HGST 7K6000 4TB + HP/HGST Enterprise 6TB / Focusrite Scarlett 2i4 sound interface + 2X Mackie MR824 / 825W PSU / Windows 7 Prof.’l 64-bit (HP OEM) > 2X Dell Ultrasharp U2715H (2560 X 1440)

[ Passmark Rating = 6280 / CPU rating = 17178 / 2D = 819 / 3D= 12629 / Mem = 3002 / Disk = 13751 / Single Thread Mark = 2368 [10.23.18]

HP z420_3: (2015) (R11) Xeon E5-1650 v2 (6C@ 4.3GHz) / z420 Liquid cooling / 64GB (HP/Samsung 8X 8GB DDR3-1866 ECC registered) / NVIDIA GeForce GTX 1060 6GB/ Samsung 860 EVO 500GB + HGST 4TB / ASUS Essence STX + Logitech z2300 2.1 / 600W PSU > Windows 7 Professional 64-bit (HP OEM ) > Samsung 40" 4K

[Passmark System Rating: = 5644 / CPU = 15293 / 2D = 847 / 3D = 10953 / Mem = 2997 Disk = 4858 /Single Thread Mark = 2384 [6.27.19]

HP ZBook 17 G2: (2015 ) i7-4940MX Extreme (4C@3.1/ 4.0GHz) / 32GB / Quadro K3100M 4GB / Kingston 480GB SATA SSD > 17.3" LCD 1920 X1080 panel > HP docking station> video externally to HP 2711x 27" LCD + Dell 17" (2007) / Logitech 533 _2.1 speaker system

[Passmark System Rating: = 3980 / CPU = 10140 / 2D = 618 / 3D = 2779 / Mem = 2559 Disk = 4662 / Single Thread Mark = 2387 [1.3.20]

[SDH]

SetFSB takes a different approach than what we've been used to, and HERE  is a bit of insight.  I have no experience with this method so far:

https://www.tomshardware.com/reviews/overclocking-software,2059-3.html

The ICS PLL chips on the HP ZX20 motherboards seem to stay close to the central bar code label on the Z420/Z620/Z820.  There seem to be two, one rectangular and one square, and they are labeled U6 and U7 on the printing on the motherboards, respectively.  Here are some images:

[DGroves]

also i would like to point out that having 4 ssd's in raid 0 is pointless from a speed perspective as both the onboard SAS/SATA controller and most affordable add in raid cards that are reasonable priced are unable to support the necessary bandwidth required for such a setup however if your doing it simply from a available drive space reason that's a different usage although nowadays  a single 1TB ssd's is becoming cost effective vs 4 ssd's

[Storm-Chaser]

Thanks for the informative post. And yes, both Z820s are fitted with two CPUs each, the higher rated PSU at 1125w and factory liquid cooling system. In reference to the overclocking stuff: We are covering both ends of the spectrum here: one z820 will be geared toward single core and the other z820 will be geared toward multi core. In the case of the 2696 v2, multi thread performance is what I will be targeting, since it has a 3.1GHz all core turbo. This means that under heavy loads, the 2696 v2 will actually outperform Intels flagship, the 12 core, the 2697 v2, because the 2697 v2 has a lower all core turbo clock speed of 3.0GHz.  So in essence, I am overclocking based on the processors inherent strengths, and in this case, its all about core count. In the case of the other z820, we will go with less cores but a higher all core turbo, centered more around single core performance, so I am confident I made the right choice in picking up two Xeon E5 2673 v2 chips to go in my second z820 chassis. 

Yes both of my Z820s are loaded with two processors each. I did that for a number of reasons, and in this case, for example, I do a lot of memory bandwidth benchmarking with this rig, so with all 16 memory slots populated, you are effectively pushing memory performance to the theoretical maximum limit by addressing all 8 memory channels. Sure, latency will take a slight hit, but it's not so bad, as this z820 scores about 71ns in memory latency. And this result is about on par with about what you can expect to see from AMD's Ryzen lineup. 

I spent a lot of time deciding on processor choice for both z820s, in other words, I did not hastily choose my processors here and I am confident in the CPU choice I have already made (for both rigs)... So with that being said, let me try to give you some context here so you can see where I am coming from and why I chose the processors that I did.

First off, I have to disagree with the 2667 v2 recommendation (at least for my purposes). Let me explain my reasoning... the 2667 v2 is an 8 core retail chip, and while it has the desirable 4.0GHz single core turbo, it also comes into the game with a relatively high TDP of 130w. The 2687W shares nearly identical performance and has the same 4.0GHz single core turbo, but comes in at a staggering 150 W TDP. I can do without the extra heat. Which is why I chose the OEM only 2673 v2 processor to be the beating hearts of my second rig. With this OEM only (and incredibly rare) chip, I am effectively getting identical performance when contrasted to the 2667 v2 or the 2687W v2, but at a substantially lower TDP. All three processors have a single core turbo speed of around 4.0GHz, and an all core turbo speed of about ~ 3.6GHz. Only reason I am hesitant to pull the trigger on a 1660 v2 (for an added focus on peak single core) is because I would lose dual processor support IIRC.   

So while the 2667 v2 looks attractive at first  blush, the OEM 2673 v2 Xeon is actually the better processor choice here (and in most scenarios, actually). Why? Because the 2673 v2 has identical base/boost) turbo clocking, but does it with a 20 watt decrease in TDP (130 vs 110w).