Post by PaulPost by PaulPost by David ChmelikI have a 5950X workstation, but would a 14900K one be an upgrade or
downgrade for computer programming/science code building/compiling/
making speed? Won't use newer AMD until there's a full or larger ATX
system-/ logic-/mother-/main-board with plain PCI (currently only on
micro-ATX).
Requires a miracle to get a PCI slot now.
Even though it is a one-chip solution to have a PCI slot (PCIe to PCI
bridge chip). [...]
https://www.cpubenchmark.net/singleThread.html
14900K 4,798
5950X 3,466
Multithreaded.
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i9-14900K&id=5717
14900K 61379
https://www.cpubenchmark.net/cpu_lookup.php?
cpu=AMD+Ryzen+9+5950X&id=3862
Post by Paul5950X 45780
[...]
Conclusion: Yes, 14800K is faster. But are you going to spend a lot of
money for that much improvement ? You have to be careful
about getting "trapped in incrementalism".
I was unaware of so many tests; thanks for details! Guess you meant
14900K (typographical error).
I usually re-read and fix the typing mistakes. Sorry about that.
Post by PaulPost by PaulA lot of lazy reviewers does not make this easy to evaluate.
I have a scaling test for the 5950x, but the OS sucks at it.
The study mostly shows strange scheduling choices. And Jim
Keller already complained that another benchmark did not
show off the best performance of his arch.
That leaves us with Passmark 61379 versus 45780.
A single point comparison, for your purchasing choice. [...]
What about lbench (multithreaded nbench replacement) on UNIX/GNU/Linux (I
also saw other tests available like Phoronix)? Those OS will be used.
So, random access memory (RAM) may affect test speeds. Maybe
relevant: AMD Ryzen 9 5950X central processing unit (CPU) is already on
this BioStar X470GTA system-board with RAM overclocked '3200' (PC3200?)
and--if possible--I'd use Intel Core i9 14900K CPU on SuperMicro X13SAE-F
system-board, which SuperMicro technical support hasn't been clear if
14900K (14th generation but X13SAE-F webpage says up to 13th generation)
or faster DDR5SDRAM even works (with BIOS update unless too many Watts) or
I should wait for possible new version/similar system-board by them, so
maybe no way to know maximum DDR5 speed yet...
My family has older 2016 personal computer (PC) local PC shop and I
built... randomly crashes/halts/reboots maybe once/week, so wondering if I
should replace BioStar with SuperMicro or if 14900K will be slower for my
usage, in contrast to 5950X, so should be my family's new one instead
(otherwise they get BioStar... so far very rarely crashing/halting). They
don't computer program nor are power users, so don't want fastest/brand-
new, but we use cores they don't, on average, and at night, number-
crunching ( http://boinc.berkeley.edu , cryptocurrency)... we both use
plain PCI slots and sometimes (number-crunching) extra display/video/
graphics cards so these are reasons for X13SAE-F or full-/extended-ATX,
even for them.
SuperMicro do not normally participate in overclocking.
This includes boosting RAM voltage. They don't normally boost RAM that way.
XMP requires boost (a boosted voltage is in the SPD table for it).
Biostar might do so.
The 5950X is at the top of the CPU support web page. But the turbo (power) value
is higher than the nominal power dissipation listed in the table here. My machine
draws 224 watts from the wall, flat out. SS-620GB Seasonic 80+ bronze. The CPU
may have a 142W limit by itself. The 105W number is an "indicator of class".
https://www.biostar.com.tw/app/en/mb/introduction.php?S_ID=955#cpusupport
The Biostar has a limited number of phases, and the VCore section got
a small heatsink. To engineer such things, you use a "power limiter", and
by limiting the power, that prevents thermal runaway on the MOSFETs.
You will likely find, that if you stick a finger on that heatsink,
while all cores are busy, it's getting hot. The temperature achieved,
the design engineer intended it.
The machine uses closed loop feedback, the way video cards have been
using closed loop feedback for a few generations. The hardware designer
can limit the power dissipated on the board, at the expense of the
performance of a high end processor. For example, the Biostar engineer could
ask the BIOS team to turn on "ECO mode" in order to keep the board cool
(no, they did not do that, that's just an example).
The 5950x is a "bucking bronco". An enthusiast board can deliver a lot
of power to it. The CPU temperature can soar, and the board could
be running with the limitation set by the CPU hitting 90C. This is why
you use Ryzen Master monitoring program, to see which operating domain
the system is in. (Linux users asked for a Linux version, but there
were no comments from AMD staff.)
You can install Windows, and use Ryzen Master for a few days, to
ascertain the control behavior of your setup.
I started with a mid-range motherboard, with no plan at the time to use
a 5950x. I started with a 5600G (6C 12T), which has tiny power consumption.
The feedback system on the motherboard is not taxed by this in the least,
and as a result, I was deceived as to the nature of the design. Fan speeds
do not modulate when the 5600G is running on the same board. You aren't even
aware there is a control system.
When I wanted to install the 5950X, the first thing was... "AMD recommends
water cooling". I kind of chuckled about this. I didn't believe them.
So I get the 5950X, fit a NH-C14S (blow-down cooler, some of the air
intended to cool the VCore heatsinks). Strictly speaking, that isn't enough
cooling (power capability not high enough). Well, I get the system running,
and I'm watching in Ryzen Master, and there are some "strange thermal transients".
The CPU was getting pretty warm, but every once in a while, it would shoot up
to 90C for a second or two, then return to the previous thermal regime. This
seemed to be some control loop artifact (the presentation seemed bang-bang
digital and not an analog ring or overshoot).
I could not figure out what the hell was going on :-) Rather than give Noctua
more money, I got a DeepCool AK620. It was $80 Canadian (perhaps a "sale" that day),
whereas a Noctua would have been a lot more.
https://www.deepcool.com/products/Cooling/cpuaircoolers/AK620-High-Performance-CPU-Cooler-1700-AM5/2021/13067.shtml
"The AK620 features a maximum heat dissipation power of 260W"
Based on volumetric limits, I was skeptical of the rating. But it does
seem to be a pretty capable cooler. While this is not "water cooling" that
AMD wanted, the thermal inertia with that thing on the system, the
control loop no longer allows "random 90C spikes". It can get hot, but
it no longer spikes.
My motherboard is likely still limiting the peak performance of the 5950X,
but I don't really want to push the thing too hard.
*******
Now, based on the "flavor" of that description, the 14900K is "more of the same".
To get the value for money, the motherboard has to feed it. Using a four phase
VCore main, for a thing like that, would be sinful. Certainly, with BIOS modification
(the so-called "ECO Mode" being an example), you can prevent these CPUs from
drawing too much power, and then an inadequate motherboard design can continue
to operate the board. But you aren't getting the value from the project, if you
buy an expensive CPU and then you dial it down to the 50% point so things
will not overheat.
You don't have to go crazy and buy one of the $1000 motherboards. But
on the other hand, a $90 motherboard is likely too far in the other extreme.
The $90 motherboards are a good match for the 65W CPUs, because it does
not take much to feed those. I have a 65W CPU and I've watched the
dynamics of that too.
To get the peak performance from the 14900K, you would want a decent
cooler on the thing, and a board that can provide turbo power. If SuperMicro
programs the power limitation, they'll set Turbo at 28 seconds or 56 seconds.
Some of the motherboard makers set the time to "Infinity" and they set
the power limiter to "4096 watts" (a register with 0xFFF kind of thing).
The VCore definitely does not deliver such power, but by removing a power limit
(VCore heatsink could overheat), the processor will be pushed a lot further.
Some of the enthusiasts like to do stuff like that. All I want, is for the
processor to not be "held back too much", rather than me attempting to set
a worlds record.
[Picture] Use "Download original" if the screen is not clear enough
Loading Image...
The motherboard might have been idling at 60W when first set up.
Today, for some reason, it idles at 36W. I have no good technical
explanation for what Windows 11 is doing... I have seen some bizarre
behaviors, and I don't like how the scheduler works, either.
The SuperPI 1.5XS run, is execution on a single core. Notice how
the system is drawing 50% of maximum system power when running on
one core. VCore is 1.43V. That is probably max-volts, so it can
run at 5GHz on one core.
The Decompression run, shows how the temperature can be highest, when
the system is not fully loaded. One silicon die may be hotter than
the other. The fan is at top speed.
The Compression run, the CPU is relatively busy, but both silicon
dies are being used to some degree, and the fan isn't actually running
as fast as in the previous case. Notice that when the system is power-limited
(last entry), the voltage is reduced to stay within the power-bounds.
In return, the fan does not have to spin as fast.
VCore Fan Condition
Idle 36W 27C 0.53V
OneCoreRailed 116W 51C 1.43V Volts or Clock limit
OneDieBusy 184W 75C 1.43V max-fan Definitely Volts limit (clock < max)
SmoothFullLoad 224W 65C 1.36V moderate fan System now power-limited (Vcore hot)
On Windows, 7ZIP can only take the whole CPU, if you use twice as
many threads as normal. To get the 5950X flat out, would require
a setting of 64 threads on the 7ZIP control panel. However, the
menu only offers 32 threads. On previous systems with lower core counts,
a 6C 12T processor running with 24 threads, fully occupies the
Task Manager CPU display. The 5950X 16C 32T, really needs a 64 thread
setting to work best, but the software does not offer that option.
*******
I see what you're after -- Intel dangling their half-assed ECC again.
I won't get fooled again on this. I see in the article notes, Intel
is playing the game they always play.
https://www.servethehome.com/supermicro-x13sae-f-intel-w680-motherboard-mini-review/
While there are 32GB DDR5 DIMMs, another option is the "asymmetric" DDR5
DIMM. This offers 48GB of memory. But this also requires clearance from
the BIOS designer, so again, it is hardly worth buying garbage hardware
only to discover some firmware or software stops it from working. The 48GB
DIMMs only work, "if the BIOS supports it".
Our lives are ruled by marketing geniuses. When things fail to work,
it is seldom the fault of the responsible design engineer :-/
While I admire your bravery going this route (to get ECC, to get one PCI slot),
you're giving up a lot to get there. Judging by the inductors in the first
picture on servethehome, it may be a 5+1+1 VCore design. Will that supply
253 watts during turbo ??? A phase can give on the order of 35W in best
case, so maybe 175W if you are lucky.
https://www.techpowerup.com/review/intel-core-i9-14900k-raptor-lake-tested-at-power-limits-down-to-35-w/
Dodging the whims of the marketing men, is an impossible task. If you
asked me for "a board with ECC and 253W power", I doubt the Venn diagram
has such an intersection point. The SuperMicro board comes close, in teasing
us with partial ECC support, and a PCI slot, I don't like the looks of the
VCore power design. At least the heatsink is "classically shaped" for
the cooling task, which is more than I can say for the "unoptimized blob
of aluminum" on my motherboard :-/
Do you remember this comparison ?
14900K 61379
5950X 45780
There is a good chance you and I will achieve neither of these operating
points, due to power and cooling limitations. Our actual numbers are
lower, because our infrastructure around the CPU, is not good enough.
The 5950X is "more likely from an architectural point" to support ECC.
What I don't know, is if any BIOS will switch that on for us. I have
stopped trying to get ECC completely, after a motherboard I purchased
which nominally supported ECC (the DIMMs fit and everything), the board
would not do ECC! That is why, for me, this now ceases to be a marketing carrot.
They can go fuck themselves.
Support for DIMM XMP option, is a handy feature for motherboards. I did not
have to make any BIOS adjustments, by using that switch in the BIOS. What
was *truly amazing* is:
XMP is only rated for two sticks, but the motherboard used the settings for four sticks.
XMP profile #1 is Command Rate 2 ("conservative"), profile #2 is Command
Rate 1 ("aggressive"). I had a board run four DIMMs at CR1 at DDR4-3200.
That's *impossible*. CR1 should *never* be used on 2DPC. Yet, it worked,
and it was passing memtest. On a previous HEDT board, four DIMMs would not
run XMP, and even at Command Rate 2, stability was only achieved at 3/4 of
rated memory speed (awful). This is how memory "normally behaves".
Whatever they are doing for the XMP setting, I just cannot believe
how much better they have done on this. I measured the memory speed,
and CR1 is faster than CR2, as you would expect, and the memory speed
difference hints that they did not cheat and use CR2 all the time
(which is an old trick they have used in the past).
Profile 1 (Command Rate 2) 38.4GB/sec \___ The difference Command Rate makes.
Profile 2 (Command Rate 1) 40.8GB/sec / My board was error-free with both!
If SuperMicro were to offer XMP, then you would select the first (conservative)
profile. The memory will be a degree or two cooler.
Good luck in your search for the "perfect motherboard".
Paul
