Buffing up my old main i7, thoughts and advice

[Zare]

Hello everyone,

I decided its time to lay off retro a bit and invest into my main PC.
For various reasons I'm open to talking about, but really not necessary to explain them now, I won't be going for a new PC but have ordered a Xeon to replace the i7.

Board : MSI X99S MPOWER (MS-7885)
BIOS : 06/13/2019 M.D2
CPU : Intel Core i7-5820K

Memory disposition;

DIMM B1 : 4GB @ 1067 MHz
DIMM B2 : 4GB @ 1067 MHz
DIMM C1 : 16GB @ 1499 MHz
DIMM D1 : 4GB @ 1067 MHz
DIMM D2 : 4GB @ 1067 MHz

Ordered : Intel Xeon E5-2699 v4 (CM8066002022506)

Of course as you can see the memory is really suboptimal, 4x4GB was bought in 2015 and I've just dumped in a 16GB stick meanwhile without any regards for performance.
Some sellers have noted this CPU requires ECC RAM, but I'm unable to find ECC RAM on support list on MSI's site. So if someone could verify I don't need ECC that would be great.

Can I expect PC to just boot up with Xeon, or do I need to switch to faster RAM first? Normally I'd wait for chip to arrive to test it, before investing in new memory.

Third open question is M.2 support via PCI-X adapters - I have no experience here whatsoever.

Looking forward to hearing your opinions and experiences with Xeons on X99 consumer boards.

 

[GiGaBiTe]

The E5-2600v4 series will work with non-ECC memory, though I definitely wouldn't recommend it. I'd also recommend fixing your memory so all banks are equal and at the max DDR4-2400 speed so that the CPU can run in quad channel mode. You lose a considerable amount of performance with heterogeneous memory channels, or missing them entirely.

Can I expect PC to just boot up with Xeon, or do I need to switch to faster RAM first? Normally I'd wait for chip to arrive to test it, before investing in new memory.

You need to make sure you have the latest BIOS installed, version M9 according to MSI:

https://www.msi.com/Motherboard/X99S-MPOWER/support#cpu

You also need to have a VERY beefy CPU cooler installed. The crappy corn cob Intel cooler won't cut it. I'd recommend a dual tower heatpipe cooler because those 2699s consume an enormous amount of power, especially if MSI diddled the turbo expiry time to be infinite and allow max boost on all cores. You're looking at probably 300W+.

I have an E5-2690v4 at work, and I have to keep the minimum fan speed at 40%. Even then it feels like a space heater coming out the back.

Third open question is M.2 support via PCI-X adapters - I have no experience here whatsoever.

There is no such thing as a M.2 to PCI-X adapter. You mean M.2 to PCIe adapter, PCI-X is an entirely different slot of an entirely different bus. As for support "there be dragons." While it may "work", the BIOS supporting M.2 drives as a boot device is entirely up to whether MSI added that feature or not.

The weird Chinese LGA2011v3 clone boards do have M.2 slots and fiddled their stolen and hacked BIOS images to support booting from M.2 drives, but YMMV on your older legit board. I have a "KILLISRE" brand LGA2011v3 clone board, and it boots from M.2 SSDs fine, it even supports other weird devices, like a M.2 6 port SATA expander.

 

[Zare]

The E5-2600v4 series will work with non-ECC memory, though I definitely wouldn't recommend it. I'd also recommend fixing your memory so all banks are equal and at the max DDR4-2400 speed so that the CPU can run in quad channel mode. You lose a considerable amount of performance with heterogeneous memory channels, or missing them entirely.

Thanks for confirming, I am missing performance right now and will fix that.
I won't go for DDR4-2400 but for 3000, going to replace all RAM.

Why wouldn't you recommend non ECC for Xeon at home? I won't run HA server. I'm pretty sure I worked on a non-ECC Xeon workstation before at work.

You need to make sure you have the latest BIOS installed, version M9 according to MSI:

https://www.msi.com/Motherboard/X99S-MPOWER/support#cpu

The latest BIOS actually isn't listed here;

https://www.msi.com/Motherboard/x99s-mpower/support

Mine is at least one or two versions up. I believe those version(s) are just microcode update for CPU vulnerabilities. The 2019 BIOS is downloadable via the Live Update tool, MSI probably forgot to update the site for this old board.

You also need to have a VERY beefy CPU cooler installed. The crappy corn cob Intel cooler won't cut it. I'd recommend a dual tower heatpipe cooler because those 2699s consume an enormous amount of power, especially if MSI diddled the turbo expiry time to be infinite and allow max boost on all cores. You're looking at probably 300W+.

I have an E5-2690v4 at work, and I have to keep the minimum fan speed at 40%. Even then it feels like a space heater coming out the back.

I have a big Thermaltake I bought extra with this computer so that it can run completely silent under average load.
The 2699V4 consumes 5W max power more than my i7 per specs, 145 vs 140. If average power consumption is larger, I believe the vent will take it, it was mostly use for higher end AMD based gamer PCs back then. The board boasts itself with overclocking ability meaning a lot of it should be configurable.

On FreeBSD I also have an option of core affinity for the OS with manual affinity for running programs if I want. Maybe that would mitigate the regular power usage of running 22 cores at once.

Will post a photo later on, as the case is in the rack. What fan are you using on yours?

There is no such thing as a M.2 to PCI-X adapter. You mean M.2 to PCIe adapter, PCI-X is an entirely different slot of an entirely different bus. As for support "there be dragons." While it may "work", the BIOS supporting M.2 drives as a boot device is entirely up to whether MSI added that feature or not.

The weird Chinese LGA2011v3 clone boards do have M.2 slots and fiddled their stolen and hacked BIOS images to support booting from M.2 drives, but YMMV on your older legit board. I have a "KILLISRE" brand LGA2011v3 clone board, and it boots from M.2 SSDs fine, it even supports other weird devices, like a M.2 6 port SATA expander.

Yeah I know and I always make the typo. Ofc my board doesn't have PCI-X.

The BIOS boot support is great but I can live without it. I can keep boot partition of FreeBSD on SSD and move root to M.2.

 

[GiGaBiTe]

Thanks for confirming, I am missing performance right now and will fix that.
I won't go for DDR4-2400 but for 3000, going to replace all RAM.

DDR4-3000 is technically out of spec for the 2699v4. You're overclocking the IMC then.

Why wouldn't you recommend non ECC for Xeon at home? I won't run HA server. I'm pretty sure I worked on a non-ECC Xeon workstation before at work.

Because the more memory and the higher speed the memory is, the more likely it is you're going to have memory errors. ECC memory isn't that much more expensive, especially if you buy it on AliExpress, where it's usually dirt cheap. I've gotten a lot of memory from AliExpress over the years, and I haven't had any more issues from it than stuff bought from an in-country retailer. They use the same memory chips, because there are only a handful of memory chip manufacturers worldwide.

The few Xeon servers I've installed regular unbuffered memory in have always had more problems with stability than those with ECC.

Mine is at least one or two versions up. I believe those version(s) are just microcode update for CPU vulnerabilities. The 2019 BIOS is downloadable via the Live Update tool, MSI probably forgot to update the site for this old board.

Just remember those CPU vulnerability mitigation patches can have severe performance penalties on older processors. If you don't plan on doing any tasks on the machine that involve sensitive information, you may want to roll back to an older BIOS revision without them, and turn Windows' software patching off as well. I've noticed considerable performance gains after disabling them, especially on machines before 6th gen Intel.

The 2699V4 consumes 5W max power more than my i7 per specs, 145 vs 140. If average power consumption is larger, I believe the vent will take it, it was mostly use for higher end AMD based gamer PCs back then. The board boasts itself with overclocking ability meaning a lot of it should be configurable.

Intels' TDP figures are a joke, and have been for decades. You can expect at least double the power consumption vs the rated TDP, especially if MSI diddles the turbo expiry time and core boost tables. HWMonitor can show you the power consumption figures. My i9-10859k is a "125W" TDP part, but it regularly pulls 300W+ under full load. Even a 240mm AIO can't keep up with it.

Will post a photo later on, as the case is in the rack. What fan are you using on yours?

It's in a Dell Precision Tower 5810. The cooler is a dual tower I think 92mm with a single 92mm fan in the middle.

 

[Unknown_K]

I have a few LGA 2011-3 CPUs and under gaming they don't seem to run very hot even with just an old 212 EVO heatsink and fan. Currently I am using a i7-5930K, E5-2667 V2, and a E5-2643 V3 if I recall correctly (forget which ones are overclocked). Basically 6C/12T and 8C/16T which is all I tend to need.

When I first got into X99 CPUs I looked at the TDP and figured they would overheat with the Thermaltake 212's so I purchased a Vetroo U6 which does 200+W TDP and found it to be overkill.

2133P DIMMs are what was standard back in the day and can still be easily found and work well in 4 channel mode. Anything faster won't hurt and is probably easier to find.

 

[GiGaBiTe]

When I first got into X99 CPUs I looked at the TDP and figured they would overheat with the Thermaltake 212's so I purchased a Vetroo U6 which does 200+W TDP and found it to be overkill.

The E5-2699v4 is a 22 core / 44 thread part, it consumes far more power than the low end 6 and 8 core parts. There's an even higher E5-2699Av4, with a higher base clock (2.4 vs 2.2 GHz.) Even though Intel rates these parts as having a 145W TDP, it's much higher.

The highest official TDP figure of a 2600 series Xeon is the E5-2679v4, with a whopping 200W TDP for 20 cores, 40 threads with a 2.5 GHz base clock.

But even my 14 core E5-2680/90 chips are furnaces.

 

[Unknown_K]

If you really need 20+ cores a more modern CPU/platform would be better with higher core speeds and much less power required.

 

[GiGaBiTe]

Uh, not really. More performance? Certainly. Less power? Most definitely not. The E5-2699v4 is also considerably cheaper than having to invest in an all new platform, it's only a few hundred bucks compared to the over thousand dollars you'd have to spend on a modern high core count system.

High core count Intel chips have always been heat and power monsters. Their modern high end chips consume so much power and put out so much heat that even a 360mm AIO can't keep up with them. They also need to be delidded and direct die cooled because the IHS is junk. You can sometimes get up to a 20C reduction in package temperatures, which lets the CPU boost higher for longer.

AMD AM5 chips are plagued with the same problems, junk IHS. Then there was the whole "popcorn CPU" fiasco where motherboards were pushing so much current into chips that they were quite literally exploding and destroying the motherboard with them.

 

[krebizfan]

The Intel chips are only power draw monsters if one lets the chip run at the maximum clock before overheating instead of running at stock clocks. A lot of benchmarks don't show the higher clock speed yielding better results. Something else is bottle necking the system. I would be completely satisfied with projects taking a few percent longer especially if it keeps me from doubling my power consumption.

 

[Chuck(G)]

Mulling over dropping another hectobuck on one of the Chinese LGA2011 boards. I've got an unused tower case, power supply and cooler and display card, so I wouldn't have that much invested in it. The one I have purchased has been trouble-free with Linux.
Probably have to pick up a SATA DVD drive...

 

[GiGaBiTe]

The Intel chips are only power draw monsters if one lets the chip run at the maximum clock before overheating instead of running at stock clocks. A lot of benchmarks don't show the higher clock speed yielding better results. Something else is bottle necking the system. I would be completely satisfied with projects taking a few percent longer especially if it keeps me from doubling my power consumption.

The reason that Intel chips pull so much power is because Intel is pushing them far outside of their efficiency curve to brute force performance. Having Hyperthreading enabled also drastically increases power consumption. They want to be one step ahead of AMD, which is why they're throwing everything at the wall to see what sticks. Had AMD gone bankrupt back in 2011/2012 from Faildozer, we'd probably still be on 4C/4T for the desktop peasants. That decade of 4C/4T Intel kept desktop consumers on probably would have continued today.

If you disable turbo boost and HT, the chips use FAR less power.

 

[Unknown_K]

Uh, not really. More performance? Certainly. Less power? Most definitely not. The E5-2699v4 is also considerably cheaper than having to invest in an all new platform, it's only a few hundred bucks compared to the over thousand dollars you'd have to spend on a modern high core count system.

High core count Intel chips have always been heat and power monsters. Their modern high end chips consume so much power and put out so much heat that even a 360mm AIO can't keep up with them. They also need to be delidded and direct die cooled because the IHS is junk. You can sometimes get up to a 20C reduction in package temperatures, which lets the CPU boost higher for longer.

AMD AM5 chips are plagued with the same problems, junk IHS. Then there was the whole "popcorn CPU" fiasco where motherboards were pushing so much current into chips that they were quite literally exploding and destroying the motherboard with them.

The only issue I know about on AM5 is one motherboard manufacturer dumping way too much voltage into the CPU above what it should be rated (since fixed with BIOS patch). AM5 and AM4 CPUs use much less power than anything Intel puts out and have real cores to boot.

A simple Ryzen 7 5800X smokes that E5 in single core and even wins with multi core (with just 8 core and 16 threads):

https://www.cpubenchmark.net/compare/AMD-Ryzen-7-5800X-vs-Intel-Xeon-E5-2699-v4/3869vs2753

Under $200 for the chip new and AM4 platforms are cheap.

The only reason I bothered with the X99 boards is I get them for next to nothing and the CPU's are dirt cheap and make decent desktops and gaming machines for older games that I like. If I was doing any real work on them, I would buy an AM4 setup.

 

[Unknown_K]

Mulling over dropping another hectobuck on one of the Chinese LGA2011 boards. I've got an unused tower case, power supply and cooler and display card, so I wouldn't have that much invested in it. The one I have purchased has been trouble-free with Linux.
Probably have to pick up a SATA DVD drive...

Just dump the installer onto a USB 3 stick and don't bother with a DVD burner unless you need one. Spares are handy if you actually use that platform.

 

[GiGaBiTe]

The only issue I know about on AM5 is one motherboard manufacturer dumping way too much voltage into the CPU above what it should be rated (since fixed with BIOS patch).

Basically all DIY motherboard manufacturers have been overvolting CPUs for decades in an effort to get them to the top of the benchmark charts. They'll slightly overclock the FSB, abuse turbo expiry time or abuse core boosting tables and need the extra voltage and current to make their abuse stable and not get too many RMAs.

In AM5's case, it was multiple manufacturers that were popcorning CPUs. Asus was probably the most egregious, but Gigabyte and some others also had popcorned CPUs. It wasn't just the motherboard though, AMD was also at fault.

AM4 isn't immune from voltage issues either. Asus boards are known to overvolt CPUs. I have two Prime B450M-A/CSM boards that overvolt the Vcore, and can't properly detect what the Vcore is. I had to probe it with a multimeter manually to get a setting that was safe for the CPU. Default voltage being pushed to the CPU had it idling at 50-55C, but fixing it to run at the proper core voltage dropped idle temperatures down into the mid 30s.

 

[Zare]

If you really need 20+ cores a more modern CPU/platform would be better with higher core speeds and much less power required.

The only issue I know about on AM5 is one motherboard manufacturer dumping way too much voltage into the CPU above what it should be rated (since fixed with BIOS patch). AM5 and AM4 CPUs use much less power than anything Intel puts out and have real cores to boot.

A simple Ryzen 7 5800X smokes that E5 in single core and even wins with multi core (with just 8 core and 16 threads):

https://www.cpubenchmark.net/compare/AMD-Ryzen-7-5800X-vs-Intel-Xeon-E5-2699-v4/3869vs2753

I have a Ryzen running in secondary computer that clean beats my i7, that computer is also running M.2 which my main PC isn't.
The question is Xeon on X99; not about the current best-buy.

In any case we'll see how the current ventilation system deals with it, and non-ECC RAM.

I'm still curious is there anything extra about Xeon and non-ECC, or are we talking just about ECC points; sure you can't have memory consistency on normal RAM at server workloads and huge uptime. Is there anything extra?

Chinese boards are interesting, but my goal is to try a CPU replacement. New board = new system.

No success granted or necessary, it might be too hot and it might really work bad with non-ECC RAM and my board might not like M.2 for one reason or another, or the single thread performance might not be as i7 as advertised and so on; consider I bought a broken 286 for about the same money whose leakage is beyond my current time/skills, and Xeon is far more easy to resell at similar buying price than that old thing.

 

[GiGaBiTe]

I'm still curious is there anything extra about Xeon and non-ECC, or are we talking just about ECC points; sure you can't have memory consistency on normal RAM at server workloads and huge uptime. Is there anything extra?

Memory errors are a problem for any computer. Most of the time, memory errors are in places where it's not critical, but sometimes they are. An example would be a computer with a memory error doing a DNS lookup, can end up looking somewhere completely different and be compromised by a threat actor with registered bit flipped domains.

ECC memory solves far more problems than just uptime length.

 

[Chuck(G)]

Just dump the installer onto a USB 3 stick and don't bother with a DVD burner unless you need one. Spares are handy if you actually use that platform.

I've got binders full of CD-ROMs and DVDs going back may years. I'd have to either invest in a USB DVD player/burner or copy the contents of the CD/DVD to a USB stick. Simply much easier to buy a cheap DVD drive.

 

[Zare]

Sure in general, but as the board doesn't support ECC RAM I'll have to do without it as I did for now (at home).

Speaking of RAM the availability of 2400MHz sticks isn't good at all, will there be a problem with IMC if RAM is at 2666MHz?

 

[GiGaBiTe]

Sure in general, but as the board doesn't support ECC RAM I'll have to do without it as I did for now (at home).

If it supports Xeon CPUs, it supports ECC. A Xeon processor is a prerequisite of using ECC RAM.

Speaking of RAM the availability of 2400MHz sticks isn't good at all, will there be a problem with IMC if RAM is at 2666MHz?

I had no problem immediately finding DDR4 2400 sticks on Newegg, Amazon and Ebay.

Here's some ECC 2400 memory: https://www.ebay.com/itm/375345975936

Intel rates the IMC for a maximum speed of 2400 MHz. There is no guarantee the CPU will be stable over that speed. You can downclock faster memory modules to slower speeds.

 

[Zare]

The board's specs say otherwise, but the board supports many many more Xeons than it does i7s, so I bet those specs are wrong.
If you go to amazon.de, or set ebay shipping to Croatia, you'll see the availability shrink.

Found on ebay.de : 84euro with shipping for 4x16GB of Hynix HMA82GR7MFR4N-UH, seemed reasonable, ordered it.