Xeon Owners Club

[Edwired]

Crazy voltage! I would have dropped it right off. 1.25 to start off. 4.6ghz really shouldn't need 1.55v.

Yeah I noticed that too but that cpu he had is not listed anywhere it like a rabbit hole with that cpu he had heaps of issues with it barely getting to the operating system and what else is wrong with it. My guess is the microcode that needs updating before trying the cpu but he tried two different boards to test it. But it a full 6 core but no threads or hyper threads as I'm guessing like a engineering sample or one of those obscure cpu that must have falling in the wrong hands or worse case the cpu is borked to the point it unusable for a daily. But it booted to 4,60ghz on asus board

 

[lexluthermiester]

My guess is the microcode that needs updating before trying the cpu but he tried two different boards to test it.

This is my guess. No BIOS microcode supports that CPU as it was clearly an internal binned sample. It is unlikely that anyone outside the engineering teams at Intel have working microcode. It's very likely to be down to a hack job.

 

[Sora]

probably needs the boot vid and boot vccsa increased.

 

[Magic Cube]

Yesterday I changed my work PC's processor from E5-2678V3 to E6-2660V4. Also had to change memory from DDR3 64GB(16x4) 1866MHz to DDR4 16GB(4x4) 2133MHz. Huananzhi X99TF has both memoryslots. All this just because I had those parts laying around. First boot I had a memory error on post display, tried other set of memory, it did not help. So I removed 2660V4 and cleaned its contacts with electronic cleaning spray. After that computer started and has been working ok. I was suprised that it feels some how faster on desktop than all core turbo modded 2678V3.

Now I should hunt four cheap either 8GB 2400MHz memorysticks or 16GB 2400MHz memorysticks. At work I do have two Crucial DDR4 Registered 8GB 2666MHz sticks, but finding two similar sticks is probably more expensive than 4 sticks with lower speed. I could also get a faster and newer computer, but I don't use this computer that often. Only when I need more processor cores.

 

[lexluthermiester]

I was suprised that it feels some how faster on desktop than all core turbo modded 2678V3.

The IPC bump from the V3 gen to the V4 gen as well as the changes to the L1/L2/L3 caching schemes are likely why your desktop feels snappier.

 

[sk111]

Hi guys I came recently across this super useful thread as I was looking to overclock my system Dell Precision T5810 with the 1650 V3.
I opt to go for something stable with the stock cooling. I got the below settings where the temperature hovers around 75 C on a stress test with CPU-Z

For CPU Cache no change

I achieve a score of 11.491 though previously with more overclocking to 44T I got 10.767 (below) though I am happy with the 43T

I wanted to ask if my settings are correct and if there is further improvement to be done on the temperatures/performance without increasing much the temperature?

I run the system all day long at these settings (as I am typing for example) without issue (not obviously at heavy load)
I found that the system when I game is much more responsive and nippy.
I have the RX 6650 XT 8GB graphics card.
Photos of the PC here (done some mods)

Update:
I read somewhere that the VCCIN should be +0.4 over the core voltage so made it lower @1.6 (as otherwise you risk long term damage??) any ideas if this is correct?

This resulted in better result in TS Bench 11.194 and max PKG power increased to 141 from 126 before. Also CPU-Z result increased slightly... however I felt the response was not as good in games.

When I increase a little more the VCCIN to 1.65 this gives better response in games and lowers slightly the score in TS bench.
Am I right here or simply it is my feeling?

Results with these settings:

Please click the green human to continue - UserBenchmark

 

[Veseleil]

Hi guys I came recently across this super useful thread as I was looking to overclock my system Dell Precision T5810 with the 1650 V3.
I opt to go for something stable with the stock cooling. I got the below settings where the temperature hovers around 75 C on a stress test with CPU-Z
View attachment 295562


View attachment 295563
View attachment 295564
For CPU Cache no change

I achieve a score of 11.491 though previously with more overclocking to 44T I got 10.767 (below) though I am happy with the 43T
View attachment 295565


I wanted to ask if my settings are correct and if there is further improvement to be done on the temperatures/performance without increasing much the temperature?

I run the system all day long at these settings (as I am typing for example) without issue (not obviously at heavy load)
I found that the system when I game is much more responsive and nippy.
I have the RX 6650 XT 8GB graphics card.
Photos of the PC here (done some mods)
View attachment 295566
View attachment 295567

Update:
I read somewhere that the VCCIN should be +0.4 over the core voltage so made it lower @1.6 (as otherwise you risk long term damage??) any ideas if this is correct?
View attachment 295568

This resulted in better result in TS Bench 11.194 and max PKG power increased to 141 from 126 before. Also CPU-Z result increased slightly... however I felt the response was not as good in games.

View attachment 295569

When I increase a little more the VCCIN to 1.65 this gives better response in games and lowers slightly the score in TS bench.
Am I right here or simply it is my feeling?

Results with these settings:

Please click the green human to continue - UserBenchmark

Big like for the case mod.

 

[sk111]

Hey thanks
By the way I tried boosting it a bit more at 4.5 as a test

Please click the green human to continue - UserBenchmark

But I hit the following limits in a game. Any idea what I need to do?

 

[Veseleil]

For any throttle stop related questions, there is a specific forum section. You might get better feedback there.

 

[Edwired]

Just doing an update on the PC bought two crucial mx300 1tb and 750gb with very good health 93% on 1tb and 98% on 750gb as don't have it here yet but next week only paid €70 for the 1tb and €50 for 750gb which is not too shabby for the ssd. Had asked another seller who have crucial mx300 1tb for €95 but won't tell me the ssd current health and won't lower the price either so avoided this seller. Just need to expand the operating system as 250gb mx100 is limited by space available.
So did abit more tinkering with the pc and manage to fix strange stuttering with the crucial mx100 as it turned out I had to clone the ssd to a hard drive and boot from that to run crucial exclusive storage to run PSID revert and secure erase the ssd which fixes the stuttering issue as before when I ran bench test on hdtune it would run but speed drop randomly and not hitting the full speed across the whole test but since the fix is applied it hitting the full speed and not a single speed drop since and clone back the operating system back onto the ssd and smooth sailing since. Wished I did this from the beginning when I got the mx100 in the first place. As there's little information about fixing the stuttering issue on crucial ssd but a lot of hits about everything else. Hope this can be helpful to others running crucial ssd

 

[sk111]

Crazy voltage! I would have dropped it right off. 1.25 to start off. 4.6ghz really shouldn't need 1.55v.

I saw his cinebench results were really low 4190. Compared to a 1650 v3 overclocked to 4.4 stable @ around 1.21 volt, cinebench is 7500 and max power 132 Watt. Also I saw he is disabling speedstep whilst I achieved this with speedstep enabled and Throttlestop

 

[Edwired]

Playing round with the Marvell sata controller on my system as managed to find a driver that works with windows 10 without bsod since I have two hard drives that are matched in spec. So one of them is connected to Intel and one is connected to Marvell so did a speed test using atto benchtest to compare the speed between the two as Intel side seems to top out at about 78mb/s write and 77mb/s read and Marvell seems to top out about 70mb/s write and 69mb/s. As the test shows it only about 7 to 8mb/s slower between Intel and Marvell not only that Marvell lost the battle with the sata 2 mode enabled for the regular hard drive but just wondering will the sata 3 mode to be enabled on ssd would see these speed increase dramatically is a big question on my mind.

But interesting note when checking with Aida64 to check the ata menu between the two hard drives in the transfer that Intel selected mwdma 2 and Marvell selected udma 6 not sure why this is the case but I do know that Marvell sata controller is been gimped when connecting to PCI-E 2.0 x2 port #1 In Use @ x1 Marvell 88SE9128 SATA 6Gb/s Controller which is shared with PCI-E 2.0 x2 port #2 In Use @ x1 NEC uPD720200F1 USB 3.0 Host Controller.

Need to do more testing to see the outcome on marvell once the two ssd come here this week.

Secondly while testing the marvell in ahci mode in the bios does make it longer to post on screen something like 5 seconds before listing the hard drives before the window login screen pops up but when marvell is in ide mode it takes about 2 seconds to list the hard drives before the login screen

As for the bios loading I can decrease the loading time by disabling smart on all the sata ports as well changing the type to not installed except the ssd in the first sata port in auto as this is needed to load the operating system otherwise you can't boot from it if it is changed to not installed. The operating system can see all the ports no problem so no side effect there

Just as I figured it out the mwdma 2 can be changed to udma5 which involving editing the registry in couple areas between Intel and windows can't really list it here as it can be confusing and a possible chance it could brick someone's computer.

What it does is changing a registry value in specific key areas that will change the transfer mode for a particular sata controller that is currently used to make all hard drives on the same mode than mismatch mode between them. As for marvel it already at udma5 even with the registry change it don't change as well it seems like a real hardware mode

 

[Sora]

The native windows driver works best on the marvel controller, no need to go searching for anything specific.

Aida64 doesn't interpret DMA mode correctly for every controller/drive combo.

 

[Edwired]

Any differences in speed between Microsoft own driver compare to Marvell own driver as like to see comparison if anyone willing to upload pictures?

How you know when Aida64 isn't interpreting correctly when there's other bench test details that don't line up on what it said actual active or supported as some don't list it either

 

[lexluthermiester]

The native windows driver works best on the marvel controller

This is not correct. They may work "good enough", but to call them "best" is a very dubious statement. Drivers from Marvell themselves are best.

 

[80-watt Hamster]

This is not correct. They may work "good enough", but to call them "best" is a very dubious statement. Drivers from Marvell themselves are best.

Most drivers are provided to Microsoft by the device vendor. I don't know if that's true for the controller in question, but can't think why it wouldn't be. Though it is plausible that there's another version that works better for whatever reason.

 

[lexluthermiester]

Most drivers are provided to Microsoft by the device vendor.

It's not as simple as that. The drivers included with Windows are cut-down bare-bones and usually much older than those offered by the venders themselves. Using the drivers built into Windows can be hit or miss, but as a rule the actual drivers direct from the makers are best.

Though it is plausible that there's another version that works better for whatever reason.

This is actually more the case than not.

 

[Edwired]

Just as I speak one of the new operating system update came through and borked the bcd and anything booting related even I can see the hard drive even access it but as soon I was going to clone the ssd for a backup it went south pretty quick no matter how many times I did the bcd repair it refused to knowledge. I have most of the other important data on other hard drives. So downloaded the newest Win 10 22H2v1 so proceeded to install it the ssd as took ages to get the memory to load so once loaded to desktop just sad having to start all over again and proceeded to set up the usual tweaks and turning off things that I don't need. Next day went to turn the pc I was greeted with login password even I didn't set it up now I'm locked out of the operating system even without the password reset key that I didn't know that it was needed even the pc not even connected online yet. Yay another headache to deal with

 

[Sora]

This is not correct. They may work "good enough", but to call them "best" is a very dubious statement. Drivers from Marvell themselves are best.

Yeah, no.
Marvels drivers operate on legacy int only, usually have lower peak throughput than the standard microsoft driver, and break smart passthrough unless the app has raid-smart compatibilty.

and usually much older than those offered by the venders themselves.

lol, no. in most cases for older hardware, microsoft has the newest version of a vendor's raid driver available that can't even be found on the vendors site, and storahci's driver is misleading, microsoft uses those weird 2006 dates for the updates from the catalog to work correctly, the files themselves are newer than the last driver provided by say, marvel, especially for the 88SE9215.

 

[lexluthermiester]

Yeah, no.
Marvels drivers operate on legacy int only, usually have lower peak throughput than the standard microsoft driver, and break smart passthrough unless the app has raid-smart compatibilty.



lol, no. in most cases for older hardware, microsoft has the newest version of a vendor's raid driver available that can't even be found on the vendors site, and storahci's driver is misleading, microsoft uses those weird 2006 dates for the updates from the catalog to work correctly, the files themselves are newer than the last driver provided by say, marvel, especially for the 88SE9215.

Hey you do you, but stop giving people bad advice. Telling people to go with the default driver installed with Windows instead of running the driver supplied by the hardware makers is foolish and bad advice.

 

[Edwired]

Have abit of good news managed fix the login issue without reinstall or losing data and also managed to get the crucial mx500 1tb with 98% health remaining including mx300 750gb with 98% health remaining both including postage for €120 so will perform some tests on marvel once it here just to see what speed and transfer is available on it

 

[Edwired]

Arrived this morning had to cover important numbers

Both ssd are on the latest firmware including both drives are wiped so nothing is skewing the results

Just testing it on the Marvell controller on the board using ATTO bench test surprising to see the difference between Intel and Marvell

On Intel it tops out above 260mb/s read and write for MX100 on sata 2 mode
On Marvell it tops out above 230mb/s on both ssd but where it starts to shine is the the read it was topping out under 390mb/s for MX500 and above 360mb/s for MX300 on Marvell sata 3 mode in normal mode I say results reflects for what it can do but might show a different result once the Marvell is set to superspeed mode just to see the comparison. Will upload pictures later

 

[Sora]

Arrived this morning had to cover important numbers

Both ssd are on the latest firmware including both drives are wiped so nothing is skewing the results

Just testing it on the Marvell controller on the board using ATTO bench test surprising to see the difference between Intel and Marvell

On Intel it tops out above 260mb/s read and write for MX100 on sata 2 mode
On Marvell it tops out above 230mb/s but where it starts to shine is the the read it was topping out under 390mb/s for both MX300 and MX500 on Marvell sata 3 mode in normal mode I say results reflects for what it can do but might show a different result once the Marvell is set to superspeed mode just to see the comparison. Will upload pictures later

the marvel controller used on x58 boards is on a 1x link and can only do 250MB/s writes theoretical (before subtracting overheads) due to some duplexing weirdness with the controller, the theoretical maximum for SATA3 on 1x each way is around 450MB/s after overheads are subtracted.

The 9128 fixed this and has the same rate both ways, but i only know of the Rampage 3 Black implementing it.

 

[Edwired]

Yeah understandable that it been gimped on the Asus p6x58d-e as I know it is hardware limited linked with the usb 3 controller I don't care much about the write performance but do care about the read performance making it ideal for game storage and other things that require higher bandwidth without too much lag or stutter but only testing it for best balance between normal mode or superspeed to see which suits me as i like the idea of normal mode therefore 1.75tb of fast storage but dislike 750gb of superspeed losing 250gb of available space and running the risk of something breaking the partition losing the data in the process but it a trade off between the two as this is my honest opinion.

The thing about the board if Asus only had the idea to remove the last pcie slot and used the data lines from there to the Marvell controller then the bios has the option to switch the change the link speed between x8 to 2nd pcie and x8 to Marvell or 16x to 2nd pcie slot and x1 to Marvell as this would make perfect sense and it could have been a perfect board for balanced gaming and server based. But Asus decided against this but used the fixed version on the extremely high end board is how I see it

 

[Sora]

Yeah understandable that it been gimped on the Asus p6x58d-e as I know it is hardware limited linked with the usb 3 controller I don't care much about the write performance but do care about the read performance making it ideal for game storage and other things that require higher bandwidth without too much lag or stutter but only testing it for best balance between normal mode or superspeed to see which suits me as i like the idea of normal mode therefore 1.75tb of fast storage but dislike 750gb of superspeed losing 250gb of available space and running the risk of something breaking the partition losing the data in the process but it a trade off between the two as this is my honest opinion.

The thing about the board if Asus only had the idea to remove the last pcie slot and used the data lines from there to the Marvell controller then the bios has the option to switch the change the link speed between x8 to 2nd pcie and x8 to Marvell or 16x to 2nd pcie slot and x1 to Marvell as this would make perfect sense and it could have been a perfect board for balanced gaming and server based. But Asus decided against this but used the fixed version on the extremely high end board is how I see it

unfortunately, the marvel 912x is absolutely 1x only, it doesn't matter how many spare lanes the IOH has on it,
Wasn't till the 88SE9220/9230/9235 that it had 2 lanes and could handle near full speed on two devices on the host controller at the same time active.

​