Building a gaming PC

+
Physics is unbreakable, but who said they need such increase in performance at the cost of such insane cooling? I certainly don't see it as a good idea. Let them progress without worsening the power draw and cooling solutions even if increase in performance won't look as good in their PR benchmarks.

One thing for sure - I'm not going to buy such ridiculously monstrous sized GPU. So some more reasonable lower level model will do if that will be needed.
You will have that .... It will be called the 4070 and it will likely run as well or better than a 3080 Ti while using less power

There are no laws forcing you to buy the flagship GPU and to me even a 3090 isn't really justifiable unless you are using it professionally or semi-professionally because you simply don't need 24 GB of memory to play games .... even half that is plenty so you are simply pissing away money for no gain. Now if you are video editing at 4K then that's a different story and that extra memory is useable
Post automatically merged:

This is what Nvidia's 12VHPWR connector looks like from the inside. There's only a thin piece of metal that connects the bridge plate to the six pins. This could be a problem for the outer wires when the cable gets bent.

Usually, each of the six pins would be crimped to one wire, without soldering.


Screenshots from Paul's Hardware,
in reference to
Igor's Lab.
There is really nothing wrong with the 12VHPWR connector itself **when properly implemented .... 8 1/3 amps per pin really isn't that much and I bet if you look at the governing body's specifications for that connector it says to use crimp connections on every pin

The problem is NOT the 12VHPWR connector it's Nvidia's improper implementation of the standard. Technically that isn't a 12VHPWR power connector it is a power adapter that doesn't properly follow the specs laid out by the governing body that writes the standards. They should have used 12 wires instead of 8 and used crimped connections which I'm sure the standard specifies. Nvidia needs to redesign those and replace the existing ones ASAP

Somewhere there is an Engineer at Nvidia telling one of the Management Beancounters "I told you so" .... This was a cost saving design by an MBA not a Electrical Engineer, I can guarantee that with 99% certainty
 
Last edited:
There are no laws forcing you to buy the flagship GPU and to me even a 3090 isn't really justifiable unless you are using it professionally or semi-professionally because you simply don't need 24 GB of memory to play games .... even half that is plenty so you are simply pissing away money for no gain. Now if you are video editing at 4K then that's a different story and that extra memory is useable
Semi true, depends on resolutions and settings. I saw over 16 gb "usage" on Cp2077 on my 3090 at times so theres games that uses more then the 6900xt for example can have. I would not pay for a 24 gb model without using it tho...
The problem is NOT the 12VHPWR connector it's Nvidia's improper implementation of the standard. Technically that isn't a 12VHPWR power connector it is a power adapter that doesn't properly follow the specs laid out by the governing body that writes the standards. They should have used 12 wires instead of 8 and used crimped connections which I'm sure the standard specifies. Nvidia needs to redesign those and replace the existing ones ASAP
Yea this is my understanding of the problem too. They have made a design error in the connector nvidia sends with the cards. The cablemod one seems fine with 8 seperate cables for power compared too the 6 nvidia ones. Also the soldering seems too be a issue, it seems they cheaped out too much? :D
 
This is what Nvidia's 12VHPWR connector looks like from the inside. There's only a thin piece of metal that connects the bridge plate to the six pins. This could be a problem for the outer wires when the cable gets bent.

Usually, each of the six pins would be crimped to one wire, without soldering.

View attachment 11326883View attachment 11326886
Screenshots from Paul's Hardware,
in reference to
Igor's Lab.
This is a picture-perfect example of what I'm talking about when I say that people buying into state-of-the-art tech are buying into unfounded, largely proprietary and experimental hardware and software.

("Picture-perfect". No pun intended. [Okay...pun intended. Pun definitely intended.])


I suppose such kind of messes is one of the reasons Evga broke up with Nvidia.
Probably stuff like this, yeah. I think their major gripe was that Nvidia would develop crazy things like this, then withhold the details and specs even from their partners until things were ready to launch. It left eVGA scrambling in the 10th hour to suddenly redesign and improve on totally unexpected curve balls thrown at their head. And I think, based on what I was reading, that when approached about how problematic this was, Nvidia basically found it funny. Their responses sounded ridiculously condescending -- especially when delivered to one of the longest-standing, leading brands supporting their product.

So, yeah, I'm sure eVGA was probably like, "You know what? Save your comments and market your own hardware. Go find someone else to solder the pins to your power supply cables! And, hint-hint, boss, maybe stop gluing your plates the PCB. We're out."
 
Last edited:
This is a picture-perfect example of what I'm talking about when I say that people buying into state-of-the-art tech are buying into unfounded, largely proprietary and experimental hardware and software.

("Picture-perfect". No pun intended. [Okay...pun intended. Pun definitely intended.])

I have a little saying you might like that I';ve been using for close to 20 years:

Those who live on the Bleeding Edge often die a Death by a Thousand Cuts

I just got a 5800X in May and a 3070 Ti in August and I don't plan on upgrading to Zen 4 or 40 Series Nvidia anytime soon .... Likely not until summer 2024 at the soonest and depending on conditions I may just skip this generation altogether and go with Zen 4+/5 and 50 series Nvidia ..... I can easily afford even Scalper prices but I just refuse to do so. I didn't become well off by wasting my money or foolishly trying to "Keep up with the Jones" and my debt load is zero. The last time I bought a prebuilt desktop computer it said Dell 486DX4-100 on it
 
Last edited:
I have a little saying you might like that I';ve been using for close to 20 years:

Those who live on the Bleeding Edge often die a Death by a Thousand Cuts

I just got a 5800X in May and a 3070 Ti in August and I don't plan on upgrading to Zen 4 or 40 Series Nvidia anytime soon .... Likely not until summer 2024 at the soonest and depending on conditions I may just skip this generation altogether and go with Zen 4+/5 and 50 series Nvidia ..... I can easily afford even Scalper prices but I just refuse to do so. I didn't become well off by wasting my money or foolishly trying to "Keep up with the Jones" and my debt load is zero. The last time I bought a prebuilt desktop computer it said Dell 486DX4-100 on it
Ditto. Plus, we're really getting to a performance plateau. Not so much the numbers, but the practical effect when compared to things like the power draw, the heat generated, the effect it will have on longevity...not to mention the out-of-control prices.

(Who in the world comes up with stuff like "Death by a Thousand Cuts"? I mean, it makes for really poetic sayings, but the fact that some maniac actually created the concept -- for real -- along with the Iron Maiden, the Bronze Bull, the Blood Eagle, etc. Seriously, humans?)
 
"Let's penny pinch on the power adapters. What could go wrong?", says the Basket Weaver. "Well, it could melt or catch on fire. We probably shouldn't penny pinch there.", says the Engineer.....
 
There are no laws forcing you to buy the flagship GPU and to me even a 3090 isn't really justifiable unless you are using it professionally or semi-professionally because you simply don't need 24 GB of memory to play games ..
I have one. Originally, I wanted a 3080 Ti, but with things unclear as to when it will come out, and the chip shortage creeping in, I've made a calculated risk and obtained the nearest available match at its near-MSRP, and that certainly wasn't an 8 GB 3080. After waiting for a few months, I got the 3090 (Strix OC) in early 2021. Given what followed with the GPU market, it was quite a bang for the buck, nothing to regret. At some point, that card was worth triple the price, and with no 3080 Ti on the horizon.

The only game able to fill up 24 GB of VRAM here is the CoD: MW, and it doesn't need all that VRAM. That part is true for gaming atm.
Post automatically merged:

I follow the market atm, pretty bad times with the uncertain state of those new connectors, with Intel's and AMD's thermals-dissipation-related challenges, with things in overall reforming between Intel and AMD, also in the GPU space, and with still high prices of DDR5.

Perhaps the next build is coming in 2024/25, and I am considering that it should sport a couple of those puppies, or alikes.
 
Last edited:
The only game able to fill up 24 GB of VRAM here is the CoD: MW, and it doesn't need all that VRAM. That part is true for gaming atm.
Yea i saw that too, shader cache is ftw :D i hade a fun issue at the start tho since it filled it up then dumped all and started over. Turns out i hade manually set my virutal memory and it did not like that.
 
Yea i saw that too, shader cache is ftw :D i hade a fun issue at the start tho since it filled it up then dumped all and started over. Turns out i hade manually set my virutal memory and it did not like that.

One of the biggest mistakes I see people do is manually setting the Page File size which really hasn't been a thing since Windows XP .... Just let Windows manage the Page File and you'll never run out (unless you run out of disk space) and it won't get fragmented. A fixed size Page File will actually get slower over time because it gets fragmented while when Windows manages the Page file it will shrink it when you quit out of a game or other memory intensive application and when it does that it moves everything to the front of the (smaller) Page File which in effect defragments it.
 
Haven't have a problem with setting it in Windows since I haven't used Windows in years ;)

Situation with VRAM on Linux is different though, since games generally use more VRAM than on Windows in case of running Windows games through Wine with let's say vkd3d-proton. So more VRAM is always welcome.
 
One of the biggest mistakes I see people do is manually setting the Page File size which really hasn't been a thing since Windows XP .... Just let Windows manage the Page File and you'll never run out (unless you run out of disk space) and it won't get fragmented. A fixed size Page File will actually get slower over time because it gets fragmented while when Windows manages the Page file it will shrink it when you quit out of a game or other memory intensive application and when it does that it moves everything to the front of the (smaller) Page File which in effect defragments it.
^ Yes.

Very important actually, especially for gaming, not to do this. The sheer speed of SSDs are utilized to amazing effect by modern virtual memory. Limiting it, locking it in place, or turning it off is literally like intentionally plugging a fuel injection pipeline in a car. You're just preventing the needed data from flowing now when it's needed.
 
Yea i saw that too, shader cache is ftw :D i hade a fun issue at the start tho since it filled it up then dumped all and started over. Turns out i hade manually set my virutal memory and it did not like that.
Virtual memory is only needed if your RAM can't match a threshold. I haven't explored in detail how much is needed, but, say, for 32 GB RAM is fine without virtual memory for me. A hypothesis can be that it won't ask for this if RAM is at least as big as VRAM or, as big as VRAM + some fixed value. A hypothetic idea behind it is that the game seems to want to be *able* (just able, not necessarily to also do it) to keep the copy of the same thing both in RAM and VRAM.
 
Virtual memory is only needed if your RAM can't match a threshold. I haven't explored in detail how much is needed, but, say, for 32 GB RAM is fine without virtual memory for me. A hypothesis can be that it won't ask for this if RAM is at least as big as VRAM or, as big as VRAM + some fixed value. A hypothetic idea behind it is that the game seems to want to be *able* (just able, not necessarily to also do it) to keep the copy of the same thing both in RAM and VRAM.
While that may be more or less true for normal programs Open World game engines are programmed to store a lot of information to the Page File mainly because of the way consoles work. Virtual Memory (AKA Page File or Swap File) and VRAM (Video Random Access Memory) are two different things. I use an 7" display to show system stats without the downside of using an overlay that looks like this
 

Attachments

  • New Sensor Panel.png
    New Sensor Panel.png
    61.7 KB · Views: 30
While that may be more or less true for normal programs Open World game engines are programmed to store a lot of information to the Page File mainly because of the way consoles work. Virtual Memory (AKA Page File or Swap File) and VRAM (Video Random Access Memory) are two different things. I use an 7" display to show system stats without the downside of using an overlay that looks like this
that's the size of the cashe, not what is in use, you will know if it is in use, because you will slow to a crawl, unless you have like a PCiE gen 5 SSD which is almost old ram speeds. nothing is actually stored in a cashe like that unless you are out of other options because it is so very slow. it would actually happen in the old days (90's) and it was unusable. even then it was better to disable it and just have stuff crash rather than pretend it could run on that.
 
RAM is used when a program assigns addresses to hold data that will be utlized by the application itself. By design, programs will not simply use any and all RAM available -- they'll work withing a RAM budget. Not all data can be loaded by a program simultaneously (or rather, not all programs can be 100% loaded in to RAM.) And most importantly, not only 1 program is ever going to be used by a system at a given moment, there are usually between 30-50 individual processes running on Windows that take up RAM in a "permanent" way.

When you're actively running an application, it's not simply going to blindly use every last stitch of system resources available -- it's going to stay within its budget. 64-bit uses significantly more, but it's still not infinite. Your virtual memory is used to quickly load and unload data assigned to specific RAM addresses in contiguous blocks: a single arrangement of data in one place on a drive that will stay put until it's needed again. This will work both ways, being written to and dumped from RAM.

1.) I arrange data that's needed so that it's ready to go when called for by a program. I write it to the swap file as one block.
2.) When the time comes, I copy the data out of RAM and save it to the swap file as one block as I write the existing block to the active RAM address.
3.) I save that block of data until the the program is done with the other block, then write the original data right back into RAM. Fast.
4.) I look at the next call the program is making, and go arrange the data for that.

Do I NEED to do it this way? No. But it's much, much faster. Even with modern SSDs being so much faster than spinning HDDs, it's still going to be more efficient to handle things with Virtual Memory active.

A drive does not store information for a program together in one place. Different types of similar files are stored together on a drive, regardless of what program they're associated with. That means that a program still needs to go get the information it needs from multiple different places in order to write it to RAM whenever called for. That takes time. Even if it's only a matter of milliseconds, it's still possible to increased the time this takes immensely. (If something takes 48 milliseconds when it could take 2...yes we're talking only milliseconds here...but that's still 24x slower.)

Or, the busing analogy. I have 20 people that need to get to work at the same location, but they all live in different spots. Virtual memory is like having them all go to the same bus stop at the same time, get on the same bus, and arrive at work together. If virtual memory is off, that bus needs to drive around to 20 different stops to pick up each person individually before taking them to work. Even if my bus is really fast, it still needs to make every stop.

If you're seeing some sort of system slowdown with virtual memory, then there is an issue with your system. Some of the most common being that the drives are too full or the swap file is active on an HDD rather than an SSD. I would recommend disabling virtual memory on spinning drives if you have an SSD available and ensuring that you save at least 10% of your free space on all drives. (I usually do 20%)
 
Last edited:
One of the biggest mistakes I see people do is manually setting the Page File size which really hasn't been a thing since Windows XP .... Just let Windows manage the Page File and you'll never run out (unless you run out of disk space) and it won't get fragmented. A fixed size Page File will actually get slower over time because it gets fragmented while when Windows manages the Page file it will shrink it when you quit out of a game or other memory intensive application and when it does that it moves everything to the front of the (smaller) Page File which in effect defragments it.
Yea i learned fast. Was fun too see the texture quality go from very low blocky too ultra 4k over and over tho :D First time ive hade an issue with it since i have enough ram for it too "never" get used.. Except for when its forced ofc :D Its going too be interesting too se what happends with the whole directstorage and so on. I think more vram will be better for that tech too work and run properly.
 
If you're seeing some sort of system slowdown with virtual memory, then there is an issue with your system. Some of the most common being that the drives are too full or the swap file is active on an HDD rather than an SSD. I would recommend disabling virtual memory on spinning drives if you have an SSD available and ensuring that you save at least 10% of your free space on all drives. (I usually do 20%)
oh i don't have any swap file at all and don't see slow down, because it is only there as a back up, if you have enough ram you never use it at all, nor does the OS or games need it for anything. I've been disabling it for 20 plus years because once in the 90's on my first computer i saw it used and it was a slide show of programs crashing.

Mostly however you are toally correct that it is not nessercry, it is managed perfectly well by the OS and will not effect perfromance at all, i just don't like it eating big chunks of space to maybe, perhaps, need to mirror what is loaded in to ram.
 
oh i don't have any swap file at all and don't see slow down, because it is only there as a back up, if you have enough ram you never use it at all, nor does the OS or games need it for anything. I've been disabling it for 20 plus years because once in the 90's on my first computer i saw it used and it was a slide show of programs crashing.

Mostly however you are toally correct that it is not nessercry, it is managed perfectly well by the OS and will not effect perfromance at all, i just don't like it eating big chunks of space to maybe, perhaps, need to mirror what is loaded in to ram.
Yeah, this is probably one of the most convoluted and mishandled aspects of computers...ever. It's never been "necessary" from the standpoint of "what is absolutely required to make a computer work". And, yes, during the earlier days of Windows, it was handled really, really badly. I also had numerous systems through Windows XP where I would configure things to either a manually set variable or totally fixed size at 1.5x RAM to avoid the hideous thrashing it could cause with gaming. I even had set virtual memory settings for different games, at times.

If you have enough RAM overhead that the system and whatever program can actively detect and utilize when it's needed, then no -- technically -- you'll be very unlikely to run into any issues. The programs just want the data when they call for it; they don't care how they get it.

When people are likely to run into trouble is when their system gets into a situation where a RAM address just isn't accessible quickly enough (often because of some other program or a background process kicking on at an inopportune moment). It's going to be more of a factor for people that do more multitasking, have lots of stuff open in web browsers constantly, are actively streaming + recording + gaming + using social media at the same time, that sort of thing. Simply having it available can be the difference between smooth sailing and a full system crash.

In general, it will improve overall stability by making the operating system more robust at all times. The days of virtual memory needing to be micromanaged to avoid sweeping issues are long gone. Windows 7 basically ironed it out very nicely, and the advent of SSDs becoming standard put the a nice coat of varnish on it.

Everyone can do whatever they wish, of course, but if players are seeing an issue and fiddling around with virtual memory seems to help, that's far more likely to be indicative of a RAM issue, rather than a virtual memory issue. It's actually not all that uncommon for RAM to be configured incorrectly in the BIOS, creating stutters, hitching, crashing, hanging, etc.
 
Top Bottom