XPost: alt.comp.hardware.pc-homebuilt

On 4/5/2024 3:56 AM, David Chmelik wrote:

We upgraded CPUs of several servers/workstations and are thinking of RAM. I'll give standard & overclock megahertz and ask advice. Numbers in parentheses aren't in guidebooks, just newer 'memory support' documents so don't know they're overclocks or merely faster RAM at slower speeds, and first of these doesn't mention overclocking at all in guidebook but newer 'memory support' document mentions faster RAM.

* ASUS Sabretooth 990FX: DDR3 1033MHz to 1866 (or overclock 2000?)
* ASUS P9X79 LE: DDR3 1066 to 1866 or overclock 2133 to 2400 (or 3000?)
* GigaByte GA-Z170XP-SLI: DDR4 2133 or overclock 2400 to 3466
* BioStar X470GTA: DDR4 1866 to 2667 or overclock 2933 to 3200 (or 4600?)

The ASUS use much slower RAM (faster was expensive then) so should I
upgrade to fastest standard speed or might slow-to-medium overclock not
take a year off system-/logic-/main-/mother-boards' remaining lifetime?

If we don't put Sabertooth 990FX in garage to run, it might be retired (storage/spare or give to PC shop) later this or early next year, but I
still want it faster remaining months/year.

I thought about overclocking P9X79 LE to 2400+ and GA-Z170XP-SLI to 3466
but guess for more hardware lifetime that's inadvisable rather than
fastest standard speed or slow-to-medium overclock?

The GigaByte is already on only/fastest standard speed and will probably
be used several more years though soon become a family spare PC (also number-cruncher for BOINC ( http://boinc.berkeley.edu ) and/or
cryptocurrency mining, which these all do when unused such as at night). GigaByte was no help advising on overclocking Z170XP-SLI nor Z270-HD3P I almost replaced with recently, just cautioning might cause problems/ instability (damage?).

I overclocked X470GTA to DDR4 3200MHz but mentioned after almost three
years a capacitor (beside RAM) popped off, so should I decrease to
2667MHz? It was my main PC and if/when repaired/replaced (similar/same system-board) will become family main PC... already thinking of upgrading (can't find adequate system-board) but don't want it to pop again in a few years due to overclock.

Even Sabretooth 990FX is fast enough if someone just wants to edit one or
two documents, or email, or browse maybe one website, and number-crunching until wears out, so what users want all four workstations for (so-called
'use case') is hardware lifetime, not overclocking unless won't
significantly shorten that, but nice if can overclock without
significantly shortening that, for faster number-crunching... Sabretooth 990FX had five-year warranty but we still use it a little 12 years later.

Of course if/when new Ryzen 9s and i9s or better are out, hopefully for full-to-extended ATX/SSI-CEB system-boards with one or two plain PCI slots (micro-ATX is inadequate for those plus large display/video/graphics card)
I might ask whether worth overclocking DDR5 SDRAM on any those or might
end up same situation (as X470GTA) after a few years it pops again.

Some people who know a lot about hardware said overclocking RAM certainly makes system-boards wear out faster, but that capacitors may wear out
around same time anyway... unsure what that means... maybe if not well- cooled, or system-boards may wear out by time one wants to upgrade, or
other parts not as fast but can't be replaced like capacitors can?

I asked a local electronics repair shop and searched online capacitor
shops for days but couldn't find anything I thought would be a good replacement. Details are the following.

5K238
C270 (C has dot so don't know it's really C or another symbol)
16V

I saw similar listed but might not either fit or last long, and repair
shop didn't reply if can get replacement, and after a couple weeks,
BioStar hadn't replied about getting just capacitor (and since warranty
page was hard to find and after buying one must register within a month-- about shipping time--my non-expired warranty is invalid).

* ASUS Sabretooth 990FX: DDR3 1033MHz to 1866 (or overclock 2000?)
* ASUS P9X79 LE: DDR3 1066 to 1866 or overclock 2133 to 2400 (or 3000?)
* GigaByte GA-Z170XP-SLI: DDR4 2133 or overclock 2400 to 3466
* BioStar X470GTA: DDR4 1866 to 2667 or overclock 2933 to 3200 (or 4600?)

At some point, you stop throwing money at the older systems. The
P9X79 would be about ten years old.

Manufacture of DDR3 stopped one year ago, March 2023. This means
for enthusiast RAM from reputable sources, they've stopped binning
it and testing for 2400. The DDR3-2400 was fine. 4x4GB SS DDR3-2400,
flip on XMP, it works. On the other hand, 8x8GB DS DDR3-2400
was not fine. Not the fault of the RAM. CPU didn't seem to like pushing
four channels fully loaded at that speed. Ended up running at 1866, so not particularly satisfying. Is currently error free, which is what
really counts on a RAM upgrade.

For the 990FX, I'd just leave that alone. Whatever is in there, is
what it's got :-) It's AMD. The "legs" on that, that vintage,
are unlikely to include heroics. If it's running and not causing a
problem, I'd just leave it. AMD might not like all slots filled,
and require turning it down and manual tuning. It might be hard
work tuning that, depending on what RAM is available and so on.
The P9X79, I have the experience with that, to say "don't use
8 DIMMs double sided with it at high speed", it'll run but
you'll have to do a lot of tweaking.

For the DDR4, you'd want to look around with regard to the
specific processor in each one, and see what the history is
like on various RAM choices. Fortunately, at the moment
they're still making DDR4. There should still be some
binned ones, and not the untested schlock that might be
advertised for sale for DDR3. I was surprised how seemingly
"drained" the retail channel was on DDR3, so quickly. I usually
expect to find "gems" hiding in plain sight for a few years.

On my DDR2 system that died, I went through three sets of RAM.
When the second set died (some not-very-nice Kingston, chips
ran warm and should not have run warm), I was expecting I
would not find anything nice and it would all be CAS6 garbage.
But the local computer store had a set of CAS5 and it
was fine until the Southbridge blew (nothing to do with RAM :-) ).
That really should not have been sitting at the store. Not
after so many years.

But for DDR3, I have my doubts there is a lot of 2400 sitting
around looking for a home. The retail channel just doesn't
behave the same any more. Enjoy your DDR4 and DDR5 choices,
which should be more predictable as far as "plug and play".
You still have to look the CPUs up, do some Googles,
to check for specific issues with such upgrades.

Since the memory controller is in the CPU, the part the
motherboard provides is copper tracks with controlled impedance.
The BIOS code has to understand how to program a specific
density. The Tsu/Th might be different on 4GB, 8GB, 16GB modules perhaps.
Maybe the 16GB modules have high density and low density modules
(8 chips on the first, 16 chips on the second), and the CPU
only happens to work on the 16 chip module, the low density one.
If an upgrade only reports half the RAM, you've made a density
mistake. On some of the processors, an adjustment of VCCSA or the
like, might cause a portion of the RAM to show up (a triple channel
12GB system that reports 8GB, the 4GB might come back after
a voltage adjustment). Good flaky fun, when weird things
like that happen. A lot of bald triple channel owners, from
the hair loss.

The DDR4 is the first RAM type that is "amazing". I have two
sample systems here, and both of them ran all slots full, DS sticks,
on XMP profile 2 (Command Rate 1). Which is ridiculous. That should
never have worked. I don't have any DDR5 systems here and cannot
comment on stuff like that. Whereas the DDR3, it tends to be
"tweak city" when you fill it up. The memory controllers
in that case, just don't seem to have the legs. On really
old DDR3 systems, their default is 1066 or something, and
"slow as molasses", so again, not a lot of joy. Spend
the money on upgrade, see nothing for results (yes, you got
more RAM perhaps, but no change in app behavior).

On my Optiplex 780 with DDR3, if I was blindfolded, I could
not tell the difference between the garbage RAM that arrived
in that refurb, versus the sweet RAM I tested in it. The
machine was unimpressed with my attempts. Made no practical
difference, whether the garbage RAM was there or not. Both sets
worked.

It takes a *huge* improvement in the RAM subsystem, to "feel"
a difference. One place we used to feel a difference, was
going from single channel memory on integrated graphics,
to dual channel memory. The integrated graphics were "snappy"
after an upgrade. For many other scenarios, you don't
even get to enjoy that. It's snappy either way.

If just the memory numbers in Task Manager changed, and
there was no other obvious difference, is it worth the
money ? You decide. It doesn't take gobs of memory to
write an email or edit a text file. You need a reason
for any kind of giant RAM config. Keeping 200 tabs open
in Firefox at 1GB each ? OK then :-)

Paul

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

XPost: alt.comp.hardware.pc-homebuilt

For Windows, I would never bother to use anything less than DDR-4.

Just wondering if you have anything with DDR-5 and if it's a noticeable improvement over DDR-4

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

XPost: alt.comp.hardware.pc-homebuilt

On 5/31/2024 5:25 AM, philo wrote:

For Windows, I would never bother to use anything less than DDR-4.

Just wondering if you have anything with DDR-5 and if it's a noticeable improvement over DDR-4

I don't have any DDR5 systems here. I "build to the trailing edge",
so I'm not paying an unnecessary premium.

https://www.techpowerup.com/review/intel-core-i9-12900k-alder-lake-ddr4-vs-ddr5/2.html

SuperPI

DDR4-4000 Gear2 378 sec
DDR4-4400 Gear2 373 sec
DDR4-3600 368 sec

DDR5-6000 Gear2 363 sec <=== ahead by 5 to 15 seconds

I would have preferred a 7ZIP benchmark, as 7ZIP is
quite dependent on cache and RAM performance. I would not take
these numbers to the bank, as the DDR5 today can stretch that
lead by a bit more. This is an early result, with slow DDR5.

https://images.anandtech.com/graphs/graph17047/127000.png

But other than 7ZIP, the improvement isn't much really.

*******

When you look at the Cinebench R23 Multi, I would have expected
the DDR5 to "win", but the results are almost independent of
RAM speed. The reason I was predicting that, is DDR5 has two channels
on a single stick, and that means two fetches could be happening
at the same time. But because it does not speed up, that hints
that perhaps even with two channels, Intel uses only one channel
on the DIMM at a time ? Just like on my 4930K with four channels,
it benches like a two channel design. The extra channels buy
you... nothing. Yet, the 7ZIP result looks promising, as there
is some boost present there. To me, the results seem in-consistent.

Paul

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

XPost: alt.comp.hardware.pc-homebuilt

On 5/31/24 9:14 AM, Paul wrote:

On 5/31/2024 5:25 AM, philo wrote:

For Windows, I would never bother to use anything less than DDR-4.

Just wondering if you have anything with DDR-5 and if it's a noticeable
improvement over DDR-4

I don't have any DDR5 systems here. I "build to the trailing edge",
so I'm not paying an unnecessary premium.

https://www.techpowerup.com/review/intel-core-i9-12900k-alder-lake-ddr4-vs-ddr5/2.html

SuperPI

DDR4-4000 Gear2 378 sec
DDR4-4400 Gear2 373 sec
DDR4-3600 368 sec

DDR5-6000 Gear2 363 sec <=== ahead by 5 to 15 seconds

I would have preferred a 7ZIP benchmark, as 7ZIP is
quite dependent on cache and RAM performance. I would not take
these numbers to the bank, as the DDR5 today can stretch that
lead by a bit more. This is an early result, with slow DDR5.

https://images.anandtech.com/graphs/graph17047/127000.png

But other than 7ZIP, the improvement isn't much really.

*******

When you look at the Cinebench R23 Multi, I would have expected
the DDR5 to "win", but the results are almost independent of
RAM speed. The reason I was predicting that, is DDR5 has two channels
on a single stick, and that means two fetches could be happening
at the same time. But because it does not speed up, that hints
that perhaps even with two channels, Intel uses only one channel
on the DIMM at a time ? Just like on my 4930K with four channels,
it benches like a two channel design. The extra channels buy
you... nothing. Yet, the 7ZIP result looks promising, as there
is some boost present there. To me, the results seem in-consistent.

Paul

I used to build all my machines but no longer do so because I get so
many cast-offs.

My newest one is an 8 core i7 3.8 ghz
It runs Win10 incredibly well and Win11 too...but I had to do the hack
to get Win11 installed>

My main Linux machine is a quad core i5

That particular machine was the *only* one I ever purchased already made
and I felt guilty. I was going to build one but this was offered on eBay
with no OS at a price too good to refuse.

It was packed wrong and the case was crushed in shipping but all else OK>

I notified the seller and since there was no way I could make a claim
with the shipper, he offered me a nice partial refund when I told him I
could re-case it.

It was a bit non- standard and I had to do a bit of modification to get
it re-cased...so I consider it a home built machine and no longer feel
guilty. LOL

Anyway as the the 8 core machine, though I do have Win11 installed on a
drive, I'm not using that...I'm fine with Win10

When support for Win10 ends, maybe by then I'll have a board that
officially supports it.


I use the Win10 machine for scanning and doing professional quality
prints and don't need to even keep the machine on-line.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

XPost: alt.comp.hardware.pc-homebuilt

On 5/31/2024 4:56 PM, philo wrote:

I used to build all my machines but no longer do so because I get so many cast-offs.

My newest one is an 8 core i7 3.8 ghz
It runs Win10 incredibly well and Win11 too...but I had to do the hack to get Win11 installed>

My main Linux machine is a quad core i5

That particular machine was the *only* one I ever purchased already made and I felt guilty. I was going to build one but this was offered on eBay with no OS at a price too good to refuse.

It was packed wrong and the case was crushed in shipping but all else OK>

I notified the seller and since there was no way I could make a claim with the shipper, he offered me a nice partial refund when I told him I could re-case it.

It was a bit non- standard and I had to do a bit of modification to get it re-cased...so I consider it a home built machine and no longer feel guilty. LOL

Anyway as the the 8 core machine, though I do have Win11 installed on a drive, I'm not using that...I'm fine with Win10

When support for Win10 ends, maybe by then I'll have a board that officially supports it.

I use the Win10 machine for scanning and doing professional quality prints and don't need to even keep the machine on-line.

By the time you got some "whizzy" machine, it would
already be obsolete :-) Some people learned that, by
buying a new laptop with a too-weak NPU in it, then Microsoft
announces they want a stronger NPU. That's the kind of
reward you get, for buying new.

I had two machine die (chipset), so that put me in the market
for a new one.

The techpowerup article, lists the 32M SuperPI 1.5 time as 354 second (a little less than six minutes) for my processor.

On my machine here, I get 358 seconds on Win10 (4 seconds slower)
and 374 seconds on Win11. The run to run variation is a lot higher than
I'm used to, and that's because of the closed loop control on the newer machines (turns down speed if it's too hot, basically). Even though the
cooler is rated at 250W, there are still signs that more cooling would
be better. On machines in the past, you only worried about cooling
as far as "Tmax" was concerned, but now you have to consider "what
the machine thinks of your setup", which is not the same thing.
If the machine seems "nervous" or "twitchy" in CPUZ, then the
cooling likely isn't good enough.

The RAM in the machine is DDR4-3200 CAS16, so pretty tame stuff.
It's possible the DDR4-3200 JEDEC is CAS22.

Paul

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

XPost: alt.comp.hardware.pc-homebuilt

On 5/31/24 5:41 PM, Paul wrote:

On 5/31/2024 4:56 PM, philo wrote:

I used to build all my machines but no longer do so because I get so many cast-offs.

My newest one is an 8 core i7 3.8 ghz
It runs Win10 incredibly well and Win11 too...but I had to do the hack to get Win11 installed>

My main Linux machine is a quad core i5

That particular machine was the *only* one I ever purchased already made and I felt guilty. I was going to build one but this was offered on eBay with no OS at a price too good to refuse.

It was packed wrong and the case was crushed in shipping but all else OK>

I notified the seller and since there was no way I could make a claim with the shipper, he offered me a nice partial refund when I told him I could re-case it.

It was a bit non- standard and I had to do a bit of modification to get it re-cased...so I consider it a home built machine and no longer feel guilty. LOL

Anyway as the the 8 core machine, though I do have Win11 installed on a drive, I'm not using that...I'm fine with Win10

When support for Win10 ends, maybe by then I'll have a board that officially supports it.


I use the Win10 machine for scanning and doing professional quality prints and don't need to even keep the machine on-line.

By the time you got some "whizzy" machine, it would
already be obsolete :-) Some people learned that, by
buying a new laptop with a too-weak NPU in it, then Microsoft
announces they want a stronger NPU. That's the kind of
reward you get, for buying new.

I had two machine die (chipset), so that put me in the market
for a new one.

The techpowerup article, lists the 32M SuperPI 1.5 time as 354 second (a little
less than six minutes) for my processor.

On my machine here, I get 358 seconds on Win10 (4 seconds slower)
and 374 seconds on Win11. The run to run variation is a lot higher than
I'm used to, and that's because of the closed loop control on the newer machines (turns down speed if it's too hot, basically). Even though the cooler is rated at 250W, there are still signs that more cooling would
be better. On machines in the past, you only worried about cooling
as far as "Tmax" was concerned, but now you have to consider "what
the machine thinks of your setup", which is not the same thing.
If the machine seems "nervous" or "twitchy" in CPUZ, then the
cooling likely isn't good enough.

The RAM in the machine is DDR4-3200 CAS16, so pretty tame stuff.
It's possible the DDR4-3200 JEDEC is CAS22.

Paul

Since most folks are now using their phones to do everything, my
computer repairs have dropped from having two or three machines on the
bench at all times, to now not even getting one computer repair a month

I also restore vacuum tube radios.

Someone had me check out a 98 year old Atwater Kent designed to run off batteries. After I built a supply for it.. amazingly it worked
...needing no repairs !

I think that was a first.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

XPost: alt.comp.hardware.pc-homebuilt

On 5/31/2024 11:42 PM, philo wrote:

On 5/31/24 5:41 PM, Paul wrote:

On 5/31/2024 4:56 PM, philo wrote:

I used to build all my machines but no longer do so because I get so many cast-offs.

My newest one is an 8 core i7 3.8 ghz
It runs Win10 incredibly well and Win11 too...but I had to do the hack to get Win11 installed>

My main Linux machine is a quad core i5

That particular machine was the *only* one I ever purchased already made and I felt guilty. I was going to build one but this was offered on eBay with no OS at a price too good to refuse.

It was packed wrong and the case was crushed in shipping but all else OK> >>>
I notified the seller and since there was no way I could make a claim with the shipper, he offered me a nice partial refund when I told him I could re-case it.

It was a bit non- standard and I had to do a bit of modification to get it re-cased...so I consider it a home built machine and no longer feel guilty. LOL

Anyway as the the 8 core machine, though I do have Win11 installed on a drive, I'm not using that...I'm fine with Win10

When support for Win10 ends, maybe by then I'll have a board that officially supports it.


I use the Win10 machine for scanning and doing professional quality prints and don't need to even keep the machine on-line.

By the time you got some "whizzy" machine, it would
already be obsolete :-) Some people learned that, by
buying a new laptop with a too-weak NPU in it, then Microsoft
announces they want a stronger NPU. That's the kind of
reward you get, for buying new.

I had two machine die (chipset), so that put me in the market
for a new one.

The techpowerup article, lists the 32M SuperPI 1.5 time as 354 second (a little
less than six minutes) for my processor.

On my machine here, I get 358 seconds on Win10 (4 seconds slower)
and 374 seconds on Win11. The run to run variation is a lot higher than
I'm used to, and that's because of the closed loop control on the newer
machines (turns down speed if it's too hot, basically). Even though the
cooler is rated at 250W, there are still signs that more cooling would
be better. On machines in the past, you only worried about cooling
as far as "Tmax" was concerned, but now you have to consider "what
the machine thinks of your setup", which is not the same thing.
If the machine seems "nervous" or "twitchy" in CPUZ, then the
cooling likely isn't good enough.

The RAM in the machine is DDR4-3200 CAS16, so pretty tame stuff.
It's possible the DDR4-3200 JEDEC is CAS22.

    Paul

Since most folks are now using their phones to do everything, my computer repairs have dropped from having two or three machines on the bench at all times, to now not even getting one computer repair a month

I also restore vacuum tube radios.

Someone had me check out a 98 year old Atwater Kent designed to run off batteries. After I built a supply for it.. amazingly it worked ...needing no repairs !

I think that was a first.

The batteries for those old radios, that would be one expensive habit.

That's the reason the battery radio I was given by someone, it
never got to run, because I couldn't afford the batteries. It took
two 45V batteries and some lump of a thing for filaments.

That could be why your set worked. It wasn't retired because it
was non-functional, it was retired because someone didn't want
to buy any more batteries for it. Mr.EverReady back then, must
have been rolling in the dough.

*******

When Windows 10 expires, I expect you'll be up to the ceiling
with "donated computers" :-) You won't lack for spare parts at
that point. There is a ChromeOS installer, if you want to put
ChromeOS on a machine. Or Linux, if the machine owner is
prepared for that. At this point, Linux can't be more complicated
than Windows is (KB5034441 they won't fix). My laptop, on a
Repair Install, put the emergency boot image (WinRE.wim)
on the wrong partition, but at this point, I've stopped caring
about "perfection". Fixing that is like "pushing on string".

Paul

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

XPost: alt.comp.hardware.pc-homebuilt

The 5v tube filaments ran from a 6v car battery which could be recharged
at the local (car) service station.
The high voltage batteries lasted a fairly long time due to low current requirements.


FWIW: my house was not electrified until 1932 so if the owner had a
radio, it would have been battery powered.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)