Trying to help the son get his desktop computer in a little better
shape. He's a 38 year old hard working guy who is trying to make it on
his own, which I really admire.
The core box is from this class

<https://support.hp.com/us-en/product/setup-user-guides/hp-z1-entry-tower-g5/model/28722929>

This computer is mainly used for CAD type work, specifically SolidWorks,
that he uses to program files to use on his Doosan CNC machine.

I found a few things when looking at it.
First, he only had a small 256 GB SSD drive, with no backup drive or
additional storage capacity.
It looks like it came with only 8 GB DDR4 ram, but was immediately
updated to 16 GB. I'd like it to go to at least 32 to speed things up.
Power supply should be big enough at 250 watts, though it does have a
dedicated video card which the SolidWorks sales guy recommended.

First, it has the 2nd slot available for a 1TB M.2 SSD. I figure I
could copy the 256 GB image to the new 1 TB drive, and erase the 256 GB
and use for storage once I've switched the connections for booting and
am sure it works.

There is a slot for a SATA spinner and appropriate cables from the power
supply. Not interested for now in an external drive, because for him
time is an important factor. I don't want him sitting around waiting
for files to be copied/written from/to a USB drive when he could be working.

It is rather intimidating looking for hardware, especially when I see
things from very knowledgeable people here talking about problems and
flaws I wouldn't even know existed. So my questions.

The new SSD should be straight forward, but I'm wondering if there are
things in particular I should look for? I think I want an 1 TB M.2 NVMe
drive. I know I have to check to make sure length and width are the
same. Anything else regarding capabilities I have to consider?

Ram should be straight forward for the most part. Biggest question is
if the board will recognize two additional 16GB sticks, or if I would
have to settle for additional 8 GB stick? I think it came with 2666
MHZ, and I see you can go faster. Would mixing the two speeds just
fallback to the 2666 MHz speed meaning it would be useless to get the
faster ram? Also, the CAD is very resource using things and I see one
of the sticks in it has the heatsinks on it. Are these really needed on
Ram or are they hype?

Installing a SATA spinner for backups has me wondering if 7200 rpm drive
is necessary, since most of the backing up will be middle of the night.
How much difference does cache make? I've seen cheaper ones with 64 MB
and what look like better at 256 MB and I believe expensive ones with
512 MB. I also figure to get him current, I should be looking for SATA
6 capable drives?

Any and all advice, as well as things to either look at or stay away
from really appreciated.

TIA
Sticks

--
Stand With Israel!
NOTE: If you use Google Groups I don't see you,
unless you're whitelisted and that's doubtful.

On 1/27/2024 1:43 PM, sticks wrote:
> Trying to help the son get his desktop computer in a little better shape.  He's a 38 year old hard working guy who is trying to make it on his own, which I really admire.
> The core box is from this class
>
> <https://support.hp.com/us-en/product/setup-user-guides/hp-z1-entry-tower-g5/model/28722929>
>
> This computer is mainly used for CAD type work, specifically SolidWorks, that he uses to
> program files to use on his Doosan CNC machine.
>
> I found a few things when looking at it.
> First, he only had a small 256 GB SSD drive, with no backup drive or additional storage capacity.
> It looks like it came with only 8 GB DDR4 ram, but was immediately updated to 16 GB.
> I'd like it to go to at least 32 to speed things up.

https://ark.intel.com/content/www/us/en/ark/products/191792/intel-core-i7-9700-processor-12m-cache-up-to-4-70-ghz.html

Max Memory Size (dependent on memory type) 128 GB
Memory Types DDR4-2666
Max # of Memory Channels 2 [Probably 4x32GB max, 2x32GB and remove the 2x8 might be a good choice)

> Power supply should be big enough at 250 watts, though it does have a dedicated video card
> which the SolidWorks sales guy recommended.

Yeah, the box is limited to a 75W video card. And some threads I could find,
some people pushing the Z1 as a "gaming machine". My ass. I have a 75W video card
right now, and I would not game on that. But the card the SolidWorks guy recommended
would be a Quadro with certified driver. That's what you're paying for is the
certified driver. Not for a gamers idea of performance. Video card drivers
are hobbled on purpose on gamer cards, to prevent people from doing
SolidWorks on them :-) That's one of the reasons you listen to the salesman,
for the explanation about card types and business practices.

>
> First, it has the 2nd slot available for a 1TB M.2 SSD.  I figure I could copy the 256 GB image to the new 1 TB drive, and erase the 256 GB and use for storage once I've switched the connections for booting and am sure it works.
>
> There is a slot for a SATA spinner and appropriate cables from the power supply.  Not interested for now in an external drive, because for him time is an important factor.  I don't want him sitting around waiting for files to be copied/written from/to a USB drive when he could be working.

Backup software like Macrium, computes a checksum as it goes and that is
a rate limiting step. Macrium lacks the threading for improved performance.

But for straight out Drag and Drop copies, this would be pretty fast.

I'm not very good at picking hardware, and there might be some really
expensive Apple ecosystem stuff you could use. We'll just offer a taste.
While there are currently two USB4 addon cards for sale, unlike previous
addons, nobody can use them. They require an I/O cable from some sort of
18 pin GPIO interface. Nothing in my room has the connector for that, so
no USB4 for me yet. Realtek is working on a USB4, but it looks like a
lab prototype, not a sale item. I can offer a half-rate solution instead.
This squeezes as much performance as I can manage, without doing anything
exotic.

2GB/sec add-on card for bottom slot of machine. $40 USB 3.2 Gen 2x2
(two 1GB paths on a USB-C connector = 2GB/sec) Connect up the Aux power connector.
May require a Y cable of some sort. You will need the Aux power, to run some
hot-snot portable SSD.

https://www.newegg.com/orico-pe20-1c-pci-express-to-usb-card/p/17Z-0003-00026

This is your storage device. $330. USB-C with USB3.2 Gen 2x2 2GB/sec (runs 1800MB/sec).

https://www.newegg.com/sandisk-extreme-pro-v2-4tb/p/N82E16820173501

You can use the pulldown menu here, and see the results of the storage consistency test of it.
Some flash drives run out of their "fake SLC cache" before the transfer is finished, and the
transfer rate drops to 1/2 or 1/3 of the original rate. It's possible the test used here,
just didn't run long enough to see the drive poop out.

https://www.anandtech.com/show/18734/samsung-portable-ssd-t7-shield-4tb-review-ip65-pssd-gets-a-capacity-upgrade/2

>
> It is rather intimidating looking for hardware, especially when I see things from very knowledgeable people here talking about problems and flaws I wouldn't even know existed.  So my questions.
>
> The new SSD should be straight forward, but I'm wondering if there are things in
> particular I should look for?  I think I want an 1 TB M.2 NVMe drive.  I know I
> have to check to make sure length and width are the same.  Anything else regarding
> capabilities I have to consider?

The motherboard needs the standoff and the screw, for mounting your M.2 .
If it were a vertical mount, like a toaster, it might need some support.

I keep an NVMe in a box. You can see this is a 2280.

[Picture]

https://i.postimg.cc/zvjWcgHN/Samsung-2280-NVMe.jpg

A PCIe Rev4 will run at Rev3 rates. The firmware on these should be fixed by now.
If there's an issue.

https://www.amazon.ca/Samsung-980-PRO-SSD-1TB/dp/B08GLX7TNT?th=1

A PCIe Rev3 will run at Rev3 rates. 3500MB/sec or so. These are
V-NAND 3-bit MLC, and they're not supposed to become mushy three months
after you write a file to them.

https://www.samsung.com/us/computing/memory-storage/solid-state-drives/ssd-970-pro-nvme-m2-1tb-mz-v7p1t0bw/#specs

It was originally announced that the width of the Rev5 ones would
be different, but I don't know what has become of the issue. There is
no particular reason to be even remotely interested in those. They've
been working on making those run at 14000MB/sec, which required waiting
for the flash chips to start yielding at the toggle rate required of the
chip interface. Some controllers have multiple channels, as part of the
bandwidth management.

>
> Ram should be straight forward for the most part.  Biggest question is if the board will recognize two additional 16GB sticks, or if I would have to settle for additional 8 GB stick?  I think it came with 2666 MHZ, and I see you can go faster.  Would mixing the two speeds just fallback to the 2666 MHz speed meaning it would be useless to get the faster ram?  Also, the CAD is very resource using things and I see one of the sticks in it has the heatsinks on it.  Are these really needed on Ram or are they hype?

You would buy 2x32GB and remove the 8GB ones and put them in the antistatic packaging
of the shipped items. There's no reason to be mixing things that far apart on
capacity.

The heatsinks are hype. Surface temperature may be 45C.

Back in RDRAM days, heatsinks were *riveted* to the module. The trick was, one of
eight chips would get hot, and dissipate 4 watts. That's because a serial interface was
involved, and a single chip could have a loop running over it and making it hot.
Having a heatsink to spread the 4W over a larger surface, was absolutely required.
You couldn't take chances.

Regular DRAM now, all chips share equal heating and the power is pretty low.
The processors have interfaces to memory, and the processors do not
allow truly excessive voltages to be used.

You used to be able to run Winbond 2.5V memory at 3.3V, to give an idea what
the fast-and-loose times were like. And those memories had forced air cooling
positioned over top.

>
> Installing a SATA spinner for backups has me wondering if 7200 rpm drive is necessary, since most of the backing up will be middle of the night. How much difference does cache make?  I've seen cheaper ones with 64 MB and what look like better at 256 MB and I believe expensive ones with 512 MB.  I also figure to get him current, I should be looking for SATA 6 capable drives?
>
> Any and all advice, as well as things to either look at or stay away from really appreciated.
>
> TIA
> Sticks

Cache wasn't really working as cache in the past. It tended to be abused by the
firmware as "scratch". For example, IBM used to keep a sector substitution table
in the SRAM at the time.

Once 512e (emulated drive, 4K physical, 512 byte virtual interface) came along,
the cache really had to work then. One side effect, is the drive can accept
1400 commands a second, when the platter only rotates 120 times per second.
Previously, the command rate would be less than the rotation rate or
less than 120 commands per second.

The 512e needs read-modify-write for updating 4K chunks on the platter, with
512 byte fractional operations. That is why the drive needed to cache some
materials and do the writes when it had the time.

The SATA AHCI has tagged queuing with 7 outstanding tags. This allows the
drive to plan the head movement, and pick the shortest path to getting things
done. The tags come back with an ACK, out of order. The cache can hold those
AHCI things until they are processed. As before, the drive has a limit on the
size of a transaction, like say, only doing 1MB of writes, at a time.

The cache can also work as a track buffer. If a request hits in the buffer, the
controller answers the query using the track buffer contents. This is in effect,
a "read ahead" capability. It's opportunistic and works if the track buffer
happens to have what you want.

On 1/27/2024 7:06 PM, Paul wrote:
> On 1/27/2024 1:43 PM, sticks wrote:
>> Trying to help the son get his desktop computer in a little better shape.  He's a 38 year old hard working guy who is trying to make it on his own, which I really admire.
>> The core box is from this class
>>
>> <https://support.hp.com/us-en/product/setup-user-guides/hp-z1-entry-tower-g5/model/28722929>
>>
>> This computer is mainly used for CAD type work, specifically SolidWorks, that he uses to
>> program files to use on his Doosan CNC machine.
>>
>> I found a few things when looking at it.
>> First, he only had a small 256 GB SSD drive, with no backup drive or additional storage capacity.
>> It looks like it came with only 8 GB DDR4 ram, but was immediately updated to 16 GB.
>> I'd like it to go to at least 32 to speed things up.
>
> https://ark.intel.com/content/www/us/en/ark/products/191792/intel-core-i7-9700-processor-12m-cache-up-to-4-70-ghz.html
>
> Max Memory Size (dependent on memory type) 128 GB
> Memory Types DDR4-2666
> Max # of Memory Channels 2 [Probably 4x32GB max, 2x32GB and remove the 2x8 might be a good choice)
>
>> Power supply should be big enough at 250 watts, though it does have a dedicated video card
>> which the SolidWorks sales guy recommended.
>
> Yeah, the box is limited to a 75W video card. And some threads I could find,
> some people pushing the Z1 as a "gaming machine". My ass. I have a 75W video card
> right now, and I would not game on that. But the card the SolidWorks guy recommended
> would be a Quadro with certified driver. That's what you're paying for is the
> certified driver. Not for a gamers idea of performance. Video card drivers
> are hobbled on purpose on gamer cards, to prevent people from doing
> SolidWorks on them :-) That's one of the reasons you listen to the salesman,
> for the explanation about card types and business practices.
>
>>
>> First, it has the 2nd slot available for a 1TB M.2 SSD.  I figure I could copy the 256 GB image to the new 1 TB drive, and erase the 256 GB and use for storage once I've switched the connections for booting and am sure it works.
>>
>> There is a slot for a SATA spinner and appropriate cables from the power supply.  Not interested for now in an external drive, because for him time is an important factor.  I don't want him sitting around waiting for files to be copied/written from/to a USB drive when he could be working.
>
> Backup software like Macrium, computes a checksum as it goes and that is
> a rate limiting step. Macrium lacks the threading for improved performance.
>
> But for straight out Drag and Drop copies, this would be pretty fast.
>
> I'm not very good at picking hardware, and there might be some really
> expensive Apple ecosystem stuff you could use. We'll just offer a taste.
> While there are currently two USB4 addon cards for sale, unlike previous
> addons, nobody can use them. They require an I/O cable from some sort of
> 18 pin GPIO interface. Nothing in my room has the connector for that, so
> no USB4 for me yet. Realtek is working on a USB4, but it looks like a
> lab prototype, not a sale item. I can offer a half-rate solution instead.
> This squeezes as much performance as I can manage, without doing anything
> exotic.
>
> 2GB/sec add-on card for bottom slot of machine. $40 USB 3.2 Gen 2x2
> (two 1GB paths on a USB-C connector = 2GB/sec) Connect up the Aux power connector.
> May require a Y cable of some sort. You will need the Aux power, to run some
> hot-snot portable SSD.
>
> https://www.newegg.com/orico-pe20-1c-pci-express-to-usb-card/p/17Z-0003-00026
>
> This is your storage device. $330. USB-C with USB3.2 Gen 2x2 2GB/sec (runs 1800MB/sec).
>
> https://www.newegg.com/sandisk-extreme-pro-v2-4tb/p/N82E16820173501
>
> You can use the pulldown menu here, and see the results of the storage consistency test of it.
> Some flash drives run out of their "fake SLC cache" before the transfer is finished, and the
> transfer rate drops to 1/2 or 1/3 of the original rate. It's possible the test used here,
> just didn't run long enough to see the drive poop out.
>
> https://www.anandtech.com/show/18734/samsung-portable-ssd-t7-shield-4tb-review-ip65-pssd-gets-a-capacity-upgrade/2
>
>>
>> It is rather intimidating looking for hardware, especially when I see things from very knowledgeable people here talking about problems and flaws I wouldn't even know existed.  So my questions.
>>
>> The new SSD should be straight forward, but I'm wondering if there are things in
>> particular I should look for?  I think I want an 1 TB M.2 NVMe drive.  I know I
>> have to check to make sure length and width are the same.  Anything else regarding
>> capabilities I have to consider?
>
> The motherboard needs the standoff and the screw, for mounting your M.2 .
> If it were a vertical mount, like a toaster, it might need some support.
>
> I keep an NVMe in a box. You can see this is a 2280.
>
> [Picture]
>
> https://i.postimg.cc/zvjWcgHN/Samsung-2280-NVMe.jpg
>
> A PCIe Rev4 will run at Rev3 rates. The firmware on these should be fixed by now.
> If there's an issue.
>
> https://www.amazon.ca/Samsung-980-PRO-SSD-1TB/dp/B08GLX7TNT?th=1
>
> A PCIe Rev3 will run at Rev3 rates. 3500MB/sec or so. These are
> V-NAND 3-bit MLC, and they're not supposed to become mushy three months
> after you write a file to them.
>
> https://www.samsung.com/us/computing/memory-storage/solid-state-drives/ssd-970-pro-nvme-m2-1tb-mz-v7p1t0bw/#specs
>
> It was originally announced that the width of the Rev5 ones would
> be different, but I don't know what has become of the issue. There is
> no particular reason to be even remotely interested in those. They've
> been working on making those run at 14000MB/sec, which required waiting
> for the flash chips to start yielding at the toggle rate required of the
> chip interface. Some controllers have multiple channels, as part of the
> bandwidth management.
>
>>
>> Ram should be straight forward for the most part.  Biggest question is if the board will recognize two additional 16GB sticks, or if I would have to settle for additional 8 GB stick?  I think it came with 2666 MHZ, and I see you can go faster.  Would mixing the two speeds just fallback to the 2666 MHz speed meaning it would be useless to get the faster ram?  Also, the CAD is very resource using things and I see one of the sticks in it has the heatsinks on it.  Are these really needed on Ram or are they hype?
>
> You would buy 2x32GB and remove the 8GB ones and put them in the antistatic packaging
> of the shipped items. There's no reason to be mixing things that far apart on
> capacity.
>
> The heatsinks are hype. Surface temperature may be 45C.
>
> Back in RDRAM days, heatsinks were *riveted* to the module. The trick was, one of
> eight chips would get hot, and dissipate 4 watts. That's because a serial interface was
> involved, and a single chip could have a loop running over it and making it hot.
> Having a heatsink to spread the 4W over a larger surface, was absolutely required.
> You couldn't take chances.
>
> Regular DRAM now, all chips share equal heating and the power is pretty low.
> The processors have interfaces to memory, and the processors do not
> allow truly excessive voltages to be used.
>
> You used to be able to run Winbond 2.5V memory at 3.3V, to give an idea what
> the fast-and-loose times were like. And those memories had forced air cooling
> positioned over top.
>
>>
>> Installing a SATA spinner for backups has me wondering if 7200 rpm drive is necessary, since most of the backing up will be middle of the night. How much difference does cache make?  I've seen cheaper ones with 64 MB and what look like better at 256 MB and I believe expensive ones with 512 MB.  I also figure to get him current, I should be looking for SATA 6 capable drives?
>>
>> Any and all advice, as well as things to either look at or stay away from really appreciated.
>>
>> TIA
>> Sticks
>
> Cache wasn't really working as cache in the past. It tended to be abused by the
> firmware as "scratch". For example, IBM used to keep a sector substitution table
> in the SRAM at the time.
>
> Once 512e (emulated drive, 4K physical, 512 byte virtual interface) came along,
> the cache really had to work then. One side effect, is the drive can accept
> 1400 commands a second, when the platter only rotates 120 times per second.
> Previously, the command rate would be less than the rotation rate or
> less than 120 commands per second.
>
> The 512e needs read-modify-write for updating 4K chunks on the platter, with
> 512 byte fractional operations. That is why the drive needed to cache some
> materials and do the writes when it had the time.
>
> The SATA AHCI has tagged queuing with 7 outstanding tags. This allows the
> drive to plan the head movement, and pick the shortest path to getting things
> done. The tags come back with an ACK, out of order. The cache can hold those
> AHCI things until they are processed. As before, the drive has a limit on the
> size of a transaction, like say, only doing 1MB of writes, at a time.
>
> The cache can also work as a track buffer. If a request hits in the buffer, the
> controller answers the query using the track buffer contents. This is in effect,
> a "read ahead" capability. It's opportunistic and works if the track buffer
> happens to have what you want.
>
> If your backups are only Macrium MRIMG files and nothing else, you don't
> need anything from the controller. As long as the drive has the sequential
> you want, you're done. However, if you ever try to install an OS on a
> 5900RPM lame drive, you will soon find out why it is only for backups.
>
> Paul

On 1/28/2024 11:51 AM, sticks wrote:

>
> Good stuff.  Thanks Paul.
>
> I took your advice and got two 32 GB sticks of ram.  I'll put the two 8 GB in my box.
> I did get a Samsung SSD, and two spinner drives, and 8GB for the kid and a 4GB for myself.
> Should get it next Friday and will try getting it in over next weekend. Interested to see if he gets a little improvement.
>
> sticks

Whether he can work faster, would depend to some extent on
how big the projects are. What the polygon count is like.

The mechanical engineer at work, was complaining about the
speed on his box. It wasn't a lack of money, more the
"orthodoxy" of the situation. Using a Sun Sparc. In his eight
hour work day, it was taking seven hours for a model to load.
Well, look at it. 100 BT Ethernet connection. Distant database server.
Sharing that server with others.

A look at the process execution, showed it was compute bound
as well. Needed a faster processor. Maybe some other box would
have been faster. Some other engineers tried out Windows PCs, but
they purchased the weirdest things (one machine had a bar bridge
in the bottom of it for cooling -- they must have had to search
pretty hard to find such a frightful contraption).

The graphics card on the Sparc, was the biggest one we could find, so no
help there. It wasn't even clear whether the acceleration
in the card was doing anything.

You know a 75W video card, can't be the top card. But unless you know
the complexity of the models being loaded, it's not time to panic yet.

Discrete (optional) graphics
AMD Radeon 520 1 GB Graphics Card
AMD Radeon RX 550X 4 GB Graphics Card
AMD Radeon RX 550 4 GB 2 DP and 1 HDMI Graphics Card
AMD Radeon RX 580 4 GB FH PCIe x16 (requires 500W chassis) <=== Solving vid card limit
AMD Radeon RX 580 8 GB FH GDDR5 (requires 500W chassis) <===
AMD Radeon R7 430 2 GB VGA and DP Graphics Card
AMD Radeon R7 430 2 GB GDDR5 64-bit 2 DP
AMD Radeon R7 430 2 GB 2 DP Graphics Card
NVIDIA GeForce RTX 2060 6 GB FH Graphics Card (requires 500W chassis) <===
NVIDIA GeForce RTX 2070 8 GB Graphics Card (requires 500W chassis) <===
NVIDIA GeForce RTX 2080 8 GB Graphics Card (requires 500W chassis) <===
NVIDIA Quadro P620 2 GB Graphics Card \_____ These two rows are
NVIDIA Quadro P400 2 GB Graphics Card / for the CAD guys! Why no rec. for 500W chassis ???

The Quadro P620 then, would have a "certified OpenGL driver" and
the card would "not be held back" when accelerating the calls.

https://www.videocardbenchmark.net/

These benchmarks aren't meaningful for CAD estimation, but the purpose is
to give some idea where in the spectrum of things, the card fits.

https://www.videocardbenchmark.net/video_lookup.php?gpu=Quadro+P620&id=3954

UHD 630 G3D 1,245 <=== The graphics chip inside the CPU, its rating

Quadro P620 G3D= 3,610 $ 574.55 <=== Presumably a 75W card (from the Z1 G5 list)

Quadro RTX 6000 G3D= 18,933 $ 6,300.00 <=== Priced for business (raytracing performance does not help)

GeForce RTX 4090 G3D= 38,842 $ 1,799.99 <=== best gamer card for comparison, to show what
a "certified OpenGL driver" is costing you.
1000W supply recommended (just for transient response!)

( https://www.videocardbenchmark.net/gpu.php?gpu=Intel+UHD+Graphics+630&id=3826 )

To fit an RTX 6000, might take a significant-sized power supply,
and that's without knowing whether the CPU versus video are
well-mated for the purpose (one having too much unusable horsepower for the other).

The "cooling equation" predicts how many CFM of exhaust fans on the
back of the PC it takes to dissipate 500W. Sometimes the results
either indicate a Hoover-like cooling solution, or that more fans
than will fit on the back, are required.

To measure percentage of P620 occupied, you could try GPU-Z while
CAD work is going on.

( Blue Download bar )

https://www.techpowerup.com/download/techpowerup-gpu-z/

These are the two cards in my machines.

[Picture] Furmark got the card to 100% usage

https://i.postimg.cc/rpzYV8QW/furmark-1050.gif

[Picture] Slightly stronger card is borderline power limited

https://i.postimg.cc/fyprrftk/GPUZ-furmark-1080.gif

You would be using that display, to measure "GPU Load (Percentage)"
to get some idea whether the CPU is using the GPU at any one instant
in time, and how much that matters to workflow.

Windows already has this display, percentage used, in the Task Manager
GPU pane. So adding no additional software at all to W10 or W11,
you have a measurement available while he works. But GPU-Z does
a better job of showing all measurements that are hooked up on the card
(temp, fan speed, slot power etc).

While conventional wisdom (at least, as promoted by a website) is that
more CPU cores are going to help, that generally isn't true. A higher clock
rate helps. But the companies offering product for sale, only
offer higher clock rates, on their slightly larger and more expensive CPUs.
You won't find a very desirable 2 core CPU running at 7GHz for sale
for example.

The CPU on this machine is an example. Only really helps for 7ZIP,
pretty well useless (you could chuck away 2/3rds of it and not notice
on core count). On a server, where a lot of separate executables
run in parallel, like an Exchange Server, a database server, a website
or three, are running, the cores do get used then. And then perhaps
the starvation point is memory bandwidth and how the cores are
connected (mesh or ring) matters. The other processor in the room, is
ring connected, and gives five cores of performance on a six core processor.
So starvation costs 16% of the expenditure. There's not enough memory
bandwidth for the sixth core. And they have laden as many as 28 cores
on stuff like that (so you can kinda guess you don't get 28 cores of
benefit from it).

You can use Maxxon Cinebench and increase the core count and benchmark
and see how badly the curve tips near the top end. That would allow you
to evaluate the internal arch of the CPU (mesh or ring).

Paul

On 1/28/2024 2:48 PM, Paul wrote:
> On 1/28/2024 11:51 AM, sticks wrote:
>
>>
>> Good stuff.  Thanks Paul.
>>
>> I took your advice and got two 32 GB sticks of ram.  I'll put the two 8 GB in my box.
>> I did get a Samsung SSD, and two spinner drives, and 8GB for the kid and a 4GB for myself.
>> Should get it next Friday and will try getting it in over next weekend. Interested to see if he gets a little improvement.
>>
>> sticks
>
> Whether he can work faster, would depend to some extent on
> how big the projects are. What the polygon count is like.
>
> The mechanical engineer at work, was complaining about the
> speed on his box. It wasn't a lack of money, more the
> "orthodoxy" of the situation. Using a Sun Sparc. In his eight
> hour work day, it was taking seven hours for a model to load.
> Well, look at it. 100 BT Ethernet connection. Distant database server.
> Sharing that server with others.

Though I had a hell of a time actually receiving the new SSD drives, I
did finally get his system all finished up. It did already have an SSD,
but at only 256GB, it was too small for my liking. But the real
improvement was that ram, put in as you suggested. I simply removed the
existing 16 GB, which was in the right slots, and put the 2 32GB sticks
in. Wow, what a difference that made. He had showed me the problem he
was having when he would force it to render the drawing after making
changes on certain variables. Like going from 0.01 to 0.0005. Before
the change, you would have to walk away and do something else, taking 10
to 15 minutes to get it done. After the 64 GB Ram upgrade, it would
take less than 30 seconds. He used to get funny lines in the rendered
drawing, and no matter what he did, he could not get them to appear. He
could grab and rotate the object around effortlessly without any
stuttering or freezing at all.

The new 1 TB disk does have better read/write speeds, but after
installing the ram, the improvement was not as noticeable. I'm sure it
is better, just not as important. The initial thought on that was he
simply needed more room, and a 2nd drive for a backup, which in this
case was a spinner doing daily backups.

>
> A look at the process execution, showed it was compute bound
> as well. Needed a faster processor. Maybe some other box would
> have been faster. Some other engineers tried out Windows PCs, but
> they purchased the weirdest things (one machine had a bar bridge
> in the bottom of it for cooling -- they must have had to search
> pretty hard to find such a frightful contraption).
>
> The graphics card on the Sparc, was the biggest one we could find, so no
> help there. It wasn't even clear whether the acceleration
> in the card was doing anything.
>
> You know a 75W video card, can't be the top card. But unless you know
> the complexity of the models being loaded, it's not time to panic yet.
>
> Discrete (optional) graphics
>
> AMD Radeon 520 1 GB Graphics Card
> AMD Radeon RX 550X 4 GB Graphics Card
> AMD Radeon RX 550 4 GB 2 DP and 1 HDMI Graphics Card
> AMD Radeon RX 580 4 GB FH PCIe x16 (requires 500W chassis) <=== Solving vid card limit
> AMD Radeon RX 580 8 GB FH GDDR5 (requires 500W chassis) <===
> AMD Radeon R7 430 2 GB VGA and DP Graphics Card
> AMD Radeon R7 430 2 GB GDDR5 64-bit 2 DP
> AMD Radeon R7 430 2 GB 2 DP Graphics Card
> NVIDIA GeForce RTX 2060 6 GB FH Graphics Card (requires 500W chassis) <===
> NVIDIA GeForce RTX 2070 8 GB Graphics Card (requires 500W chassis) <===
> NVIDIA GeForce RTX 2080 8 GB Graphics Card (requires 500W chassis) <===
> NVIDIA Quadro P620 2 GB Graphics Card \_____ These two rows are
> NVIDIA Quadro P400 2 GB Graphics Card / for the CAD guys! Why no rec. for 500W chassis ???
>
> The Quadro P620 then, would have a "certified OpenGL driver" and
> the card would "not be held back" when accelerating the calls.

This really amazed me when I first opened the box and you were
absolutely right in figuring out it would be a Quadro card. I took a
picture to give the model number and look up, but lost it, darn it.
Still, cool you knew what it would be.

---snip---

This one works well, he's happy, it is backing up properly, and I
appreciate the helpful advice!

--
Stand With Israel!
NOTE: If you use Google Groups I don't see you,
unless you're whitelisted and that's doubtful.

On 2/19/2024 7:32 PM, sticks wrote:
> On 1/28/2024 2:48 PM, Paul wrote:
>> On 1/28/2024 11:51 AM, sticks wrote:
>>
>>>
>>> Good stuff.  Thanks Paul.
>>>
>>> I took your advice and got two 32 GB sticks of ram.  I'll put the two 8 GB in my box.
>>> I did get a Samsung SSD, and two spinner drives, and 8GB for the kid and a 4GB for myself.
>>> Should get it next Friday and will try getting it in over next weekend. Interested to see if he gets a little improvement.
>>>
>>> sticks
>>
>> Whether he can work faster, would depend to some extent on
>> how big the projects are. What the polygon count is like.
>>
>> The mechanical engineer at work, was complaining about the
>> speed on his box. It wasn't a lack of money, more the
>> "orthodoxy" of the situation. Using a Sun Sparc. In his eight
>> hour work day, it was taking seven hours for a model to load.
>> Well, look at it. 100 BT Ethernet connection. Distant database server.
>> Sharing that server with others.
>
> Though I had a hell of a time actually receiving the new SSD drives, I did finally get his system all finished up.  It did already have an SSD, but at only 256GB, it was too small for my liking.  But the real improvement was that ram, put in as you suggested.  I simply removed the existing 16 GB, which was in the right slots, and put the 2 32GB sticks in.  Wow, what a difference that made.  He had showed me the problem he was having when he would force it to render the drawing after making changes on certain variables.  Like going from 0.01 to 0.0005.  Before the change, you would have to walk away and do something else, taking 10 to 15 minutes to get it done.  After the 64 GB Ram upgrade, it would take less than 30 seconds.  He used to get funny lines in the rendered drawing, and no matter what he did, he could not get them to appear.  He could grab and rotate the object around effortlessly without any stuttering or freezing at all.
>
> The new 1 TB disk does have better read/write speeds, but after installing the ram, the improvement was not as noticeable.  I'm sure it is better, just not as important.  The initial thought on that was he simply needed more room, and a 2nd drive for a backup, which in this case was a spinner doing daily backups.
>
>>
>> A look at the process execution, showed it was compute bound
>> as well. Needed a faster processor. Maybe some other box would
>> have been faster. Some other engineers tried out Windows PCs, but
>> they purchased the weirdest things (one machine had a bar bridge
>> in the bottom of it for cooling -- they must have had to search
>> pretty hard to find such a frightful contraption).
>>
>> The graphics card on the Sparc, was the biggest one we could find, so no
>> help there. It wasn't even clear whether the acceleration
>> in the card was doing anything.
>>
>> You know a 75W video card, can't be the top card. But unless you know
>> the complexity of the models being loaded, it's not time to panic yet.
>>
>> Discrete (optional) graphics
>>     
>> AMD Radeon 520 1 GB Graphics Card
>> AMD Radeon RX 550X 4 GB Graphics Card
>> AMD Radeon RX 550 4 GB 2 DP and 1 HDMI Graphics Card
>> AMD Radeon RX 580 4 GB FH PCIe x16 (requires 500W chassis)    <=== Solving vid card limit
>> AMD Radeon RX 580 8 GB FH GDDR5 (requires 500W chassis)       <===
>> AMD Radeon R7 430 2 GB VGA and DP Graphics Card
>> AMD Radeon R7 430 2 GB GDDR5 64-bit 2 DP
>> AMD Radeon R7 430 2 GB 2 DP Graphics Card
>> NVIDIA GeForce RTX 2060 6 GB FH Graphics Card (requires 500W chassis)  <===
>> NVIDIA GeForce RTX 2070 8 GB Graphics Card (requires 500W chassis)     <===
>> NVIDIA GeForce RTX 2080 8 GB Graphics Card (requires 500W chassis)     <===
>> NVIDIA Quadro P620 2 GB Graphics Card \_____ These two rows are
>> NVIDIA Quadro P400 2 GB Graphics Card /      for the CAD guys! Why no rec. for 500W chassis ???
>>
>> The Quadro P620 then, would have a "certified OpenGL driver" and
>> the card would "not be held back" when accelerating the calls.
>
> This really amazed me when I first opened the box and you were absolutely right in figuring out it would be a Quadro card.  I took a picture to give the model number and look up, but lost it, darn it. Still, cool you knew what it would be.
>
> ---snip---
>
> This one works well, he's happy, it is backing up properly, and I appreciate the helpful advice!
>

Thanks for the feedback.

I never would have guessed some CAD, would be such a memory pig :-)

I guess you could look in Task Manager, click the memory
tab and watch as it consumed RAM.

*******

I happen to have a 1050 Ti card in this computer, and the
article for the P620 claims it uses a similar GPU chip.

https://www.techpowerup.com/gpu-specs/quadro-p620.c3085

"512 shading units, 32 texture mapping units, and 16 ROPs

power draw is rated at 40 W maximum

Release Date Feb 1st, 2018
"

So that was released after the 1050Ti, presumably when the
driver was ready for it.

The power consumption of my 1050 Ti is so low, the fan
does not normally spin during a session. I have to think
up stuff to make the fan spin on it.

Paul

On 2/19/2024 8:12 PM, Paul wrote:
> On 2/19/2024 7:32 PM, sticks wrote:
>> On 1/28/2024 2:48 PM, Paul wrote:
>>> On 1/28/2024 11:51 AM, sticks wrote:
>>>
>>>>
>>>> Good stuff.  Thanks Paul.
>>>>
>>>> I took your advice and got two 32 GB sticks of ram.  I'll put the two 8 GB in my box.
>>>> I did get a Samsung SSD, and two spinner drives, and 8GB for the kid and a 4GB for myself.
>>>> Should get it next Friday and will try getting it in over next weekend. Interested to see if he gets a little improvement.
>>>>
>>>> sticks
>>>
>>> Whether he can work faster, would depend to some extent on
>>> how big the projects are. What the polygon count is like.
>>>
>>> The mechanical engineer at work, was complaining about the
>>> speed on his box. It wasn't a lack of money, more the
>>> "orthodoxy" of the situation. Using a Sun Sparc. In his eight
>>> hour work day, it was taking seven hours for a model to load.
>>> Well, look at it. 100 BT Ethernet connection. Distant database server.
>>> Sharing that server with others.
>>
>> Though I had a hell of a time actually receiving the new SSD drives, I did finally get his system all finished up.  It did already have an SSD, but at only 256GB, it was too small for my liking.  But the real improvement was that ram, put in as you suggested.  I simply removed the existing 16 GB, which was in the right slots, and put the 2 32GB sticks in.  Wow, what a difference that made.  He had showed me the problem he was having when he would force it to render the drawing after making changes on certain variables.  Like going from 0.01 to 0.0005.  Before the change, you would have to walk away and do something else, taking 10 to 15 minutes to get it done.  After the 64 GB Ram upgrade, it would take less than 30 seconds.  He used to get funny lines in the rendered drawing, and no matter what he did, he could not get them to appear.  He could grab and rotate the object around effortlessly without any stuttering or freezing at all.
>>
>> The new 1 TB disk does have better read/write speeds, but after installing the ram, the improvement was not as noticeable.  I'm sure it is better, just not as important.  The initial thought on that was he simply needed more room, and a 2nd drive for a backup, which in this case was a spinner doing daily backups.
>>
>>>
>>> A look at the process execution, showed it was compute bound
>>> as well. Needed a faster processor. Maybe some other box would
>>> have been faster. Some other engineers tried out Windows PCs, but
>>> they purchased the weirdest things (one machine had a bar bridge
>>> in the bottom of it for cooling -- they must have had to search
>>> pretty hard to find such a frightful contraption).
>>>
>>> The graphics card on the Sparc, was the biggest one we could find, so no
>>> help there. It wasn't even clear whether the acceleration
>>> in the card was doing anything.
>>>
>>> You know a 75W video card, can't be the top card. But unless you know
>>> the complexity of the models being loaded, it's not time to panic yet.
>>>
>>> Discrete (optional) graphics
>>>
>>> AMD Radeon 520 1 GB Graphics Card
>>> AMD Radeon RX 550X 4 GB Graphics Card
>>> AMD Radeon RX 550 4 GB 2 DP and 1 HDMI Graphics Card
>>> AMD Radeon RX 580 4 GB FH PCIe x16 (requires 500W chassis)    <=== Solving vid card limit
>>> AMD Radeon RX 580 8 GB FH GDDR5 (requires 500W chassis)       <===
>>> AMD Radeon R7 430 2 GB VGA and DP Graphics Card
>>> AMD Radeon R7 430 2 GB GDDR5 64-bit 2 DP
>>> AMD Radeon R7 430 2 GB 2 DP Graphics Card
>>> NVIDIA GeForce RTX 2060 6 GB FH Graphics Card (requires 500W chassis)  <===
>>> NVIDIA GeForce RTX 2070 8 GB Graphics Card (requires 500W chassis)     <===
>>> NVIDIA GeForce RTX 2080 8 GB Graphics Card (requires 500W chassis)     <===
>>> NVIDIA Quadro P620 2 GB Graphics Card \_____ These two rows are
>>> NVIDIA Quadro P400 2 GB Graphics Card /      for the CAD guys! Why no rec. for 500W chassis ???
>>>
>>> The Quadro P620 then, would have a "certified OpenGL driver" and
>>> the card would "not be held back" when accelerating the calls.
>>
>> This really amazed me when I first opened the box and you were absolutely right in figuring out it would be a Quadro card.  I took a picture to give the model number and look up, but lost it, darn it. Still, cool you knew what it would be.
>>
>> ---snip---
>>
>> This one works well, he's happy, it is backing up properly, and I appreciate the helpful advice!
>>
>
> Thanks for the feedback.
>
> I never would have guessed some CAD, would be such a memory pig :-)

The program being used is Mastercam. I think it's probably what most
people in the machining industry use.

>
> I guess you could look in Task Manager, click the memory
> tab and watch as it consumed RAM.
>
> *******
>
> I happen to have a 1050 Ti card in this computer, and the
> article for the P620 claims it uses a similar GPU chip.
>
> https://www.techpowerup.com/gpu-specs/quadro-p620.c3085
>
> "512 shading units, 32 texture mapping units, and 16 ROPs
>
> power draw is rated at 40 W maximum
>
> Release Date Feb 1st, 2018
> "
>
> So that was released after the 1050Ti, presumably when the
> driver was ready for it.
>
> The power consumption of my 1050 Ti is so low, the fan
> does not normally spin during a session. I have to think
> up stuff to make the fan spin on it.

I should check on this stuff just for kicks if nothing else. The power
usage is first on the list as it only has a 250 Watt supply. Would be
interesting to find out how much it actually draws. I could have left
the second SSD drive in it, but at only 256 GB, figured it was
unnecessary and every little bit of less draw would help.

--
Stand With Israel!
NOTE: If you use Google Groups I don't see you,
unless you're whitelisted and that's doubtful.

On 2/20/2024 1:37 PM, sticks wrote:
> On 2/19/2024 8:12 PM, Paul wrote:
>>
>> I never would have guessed some CAD, would be such a memory pig :-)
>
> The program being used is Mastercam.  I think it's probably what most people in the machining industry use.
>

Their recommended memory is 32GB.

https://www.mastercam.com/support/technical-support/system-requirements/

which to me, seems a bit ambitious. In fact, that's a danger signal
that we should look up how the software works.

https://www.linkedin.com/pulse/ten-design-softwares-every-molder-should-know-plastic-mould-ivy-cao-

"Mastercam is a CAD/CAM software based on PC platform developed by CNC Software Inc.
in United States. It integrates multiple functions such as two-dimensional drawing,
three-dimensional solid modeling, surface design, voxel stitching, <=== mentions voxel
numerical control programming, tool path simulation and realistic simulation.
"

A software could use polygons for representation of a piece of work. Or it could use voxels.

Polygons could "benefit from a good graphics card". Voxels, perhaps not as much of a card is required.

Voxels are basically a 3D pixmap, which is very
consumptive of memory, and increasingly so when you "crank" the
precision knob on the tool. That can multiply the memory
requirement... by a lot.

"He had showed me the problem he was having when he would force it to render
the drawing after making changes on certain variables. Like going from 0.01 to 0.0005.
"

In three dimensions, using voxels, that results in a factor of 8000. Now, the program
is hitting the pagefile (or, the functional equivalent of a pagefile).
Some companies do not like the behavior of the system paging, so they
write their own paging and have a separate file used for paging. On
Microsoft ICE, when working on a large panorama, they used a separate
2TB hard drive just for "private paging". If your job was big enough,
the prompt would change to "connect a 6TB hard drive to finish the job".

From the system requirements again:

"NVMe Drive with at least 20GB free"

While a program can pretend its paging algo is I/O limited, this is
seldom true. A lot of paging methods, are CPU bound, and the delays
involved don't always make sense (the delays can be a lot longer than
any reasonable person would think). While they can casually mention an
NVMe, again, this is material for Task Manager, and watching how
a "job execution" unfolds. You would watch Task Manager memory tab,
while changing your job from 0.01 to 0.0005, to see what the proportionality
factor was. You would also start watching the disk I/O tab, if you
see the storage LED coming on a lot while the job is rendering.
To get some idea whether the SATA SSD is flat out at 530MB/sec or
whatever. (Note that Windows ability to measure things, isn't
all that accurate.)

You can put an NVMe into a non-NVMe computer, but then it is generally
not the boot drive. For example, you could put the MasterCad paging
file on a separate NVMe data drive. But I would only be doing that,
if there was some indication that paging used "unlimited I/O speed".
I would estimate it's been about fifteen years, since I saw an I/O limited
paging method at play. Today, using 14000MB/sec PCIe Rev5 storage, isn't
going to make a dent in the problem. Just a waste of electricity.

(this is just to demonstrate, you can have NVMe on a separate card...
Don't actually buy one of these. The card may not show up in the BIOS,
which means it does not support booting. But can function as a data device.)

https://www.tomshardware.com/news/pcie4-card-21-m2-ssds-168tb-31gbps

I've seen computer paging algos that are so poor, the disk I/O
rate is 1.5MB/sec. And that would not even tax my year 2000 hard drive :-)

I used to have a subscription to Computer Graphics magazine, even though
I wasn't designing graphics hardware. Voxels were introduced a long time
ago, and well before their usage would have been practical. And even at the
time, they might discuss "how could we compress these, to reduce resource
requirements". That's why I cannot estimate the proportionality factor,
and I assume they've "bent the curve" a bit while inventing this stuff.

And generally speaking, large problems are not handled well by your
average piece of software. Usually, developers struggle to get accuracy
from a method, and performance "is, whatever it is". There usually isn't
time for optimization.

When Vista was released, the file copy operations hadn't been fully tested.
To give you some idea how "green" software can be :-) By Vista SP2, it was
in pretty good shape by then. So they called it Windows 7.

Paul