I’m assuming you are running Windows. Open up Task Manager (Win 8/10) or Resource Monitor (pre Win 8 ) and you will see the bottlenecks in your system (i.e. persistently 100%).
Usually they are HDD/SSD, then RAM, then CPU so upgrading in that order can help, but may just shift the bottleneck. My rule of thumb is that the CPU is the most expensive so it should be adequately resourced, i.e. the bottleneck. If you upgrade it you need to factor in the costs of adequately resourcing it.
On using different RAM. It is usually OK but will run at the lowest common factors, which is why matching is preferable. There can be inconsistencies, to the standards, in cheaper RAM but a good tester (MemTest etc.) should find these.
More RAM or Better CPU?
-
- KVRian
- 652 posts since 2 Mar, 2015 from UK
I wouldn't buy another CPU while the new ones are coming out. There's another Ryzen out in 2018. Switching sockets is expensive though.
-
- KVRian
- 700 posts since 19 Jan, 2008
The thing with AMD is they aren't changing sockets like Intel. AM4 platform will be supported until 2020 and every CPU until then will be based on AM4. You'll just have to update bios and will be able to use any newer CPU.
-
- KVRian
- 652 posts since 2 Mar, 2015 from UK
AMD TR4 looks like a good socket too the Threadripper processor has two CPU dies + two dummy ones to support the heat spreader so if AMD decides, the TR4 socket could get a 32core Threadripper.
-
- KVRian
- 716 posts since 23 Dec, 2011
This is worth repeating. A slow drive, will tax RAM (swap), and certainly slow everything (app launchin / execution), and disk access. Pricing for SSDs has plummeted over the last year or two. It's an investment that can be stuck into the next system you purchase too.Burillo wrote:get an SSD if you still haven't.
JDoo
18 years Intel corp, retired 2 years ago.. the SSD thing is something we didn't want to tout too strongly
-
- KVRAF
- 12555 posts since 7 Dec, 2004
My system uses 3 x 8 gb cards and is currently (while semi-idle, its typical usage with browsers + IDEs + media player + etc open) using 3.5 gb out of the 24 gb available.
This memory cost me only $250 when I bought it and I'd expect prices are similar now. There is absolutely no limit on the amount of memory you can place in a system: but you need to be aware of three key things!
PC hardware isn't too complicated, but it is definitely complex with many working parts interacting in non-obvious ways. It isn't as simple as just plugging in a new stick of RAM or a CPU: the whole system needs to be considered all together just to get mediocre results as opposed to having it blow up in your face.
The apparent knowledge level regarding PC hardware on a forum like here on KVR isn't too high. Please keep that in mind. In other words if you honestly need to ask "more RAM or a better CPU?" you should be asking yourself another question like "should I do a bit more research and learn more before I make a good cost vs. benefit decision?"
Regarding #1 (PSU/wattage): consider for example the cheaper memory brands which tend to have big heatsinks on them. Compare to more expensive memory with smaller or no heatsink. Think about what that means. If you look at the specs you might find the expensive cards without heatsinks use a smaller transistor size or a variation on one of many modern "3d/stacked" semiconductor architectures, which end up requiring far less power to accomplish the same thing.
So in that case you might think "wow, those heatsinks are HUGE! SO AWESOMEEEEE, they must be the best ever! THINK OF ALL THE HEAT THEY CAN SINK!". That would be wrong. The biggest heatsink is in fact demonstrating the shortcomings of an inferior product.
This memory cost me only $250 when I bought it and I'd expect prices are similar now. There is absolutely no limit on the amount of memory you can place in a system: but you need to be aware of three key things!
- The WATTs required for that memory. Modern systems are highly efficient but more memory will require more power, even during idle/suspend! This can require doubling the capability of your PSU ($500-$1000) to cope with a few extra memory cards ($50-$200).
- The type of memory controller and system configuration for timing, the type of memory, bus speeds and so on. There is not much point to buying high speed memory cards and plugging them into a system where they're forced to run at 1/8th the speed. Everything needs to be carefully tuned to get optimum performance including which cards you use, how they're mixed and even the order you place them in the slots on your mainboard in addition to the mainboard configuration for memory.
- You need to be aware of the type of software you're using, most important the OS you're running. Various systems have various limitations on the amount of RAM they can handle. Plugging 64 gb into a Windows version that only handles 8 gb won't do much good. Increasing the total memory available on your system won't help much if you're running a 32-bit host with 32-bit plug-ins or a 32-bit OS. Even with 64-bit across the board, in 99% of cases 8 gb is more than enough for most applications.
PC hardware isn't too complicated, but it is definitely complex with many working parts interacting in non-obvious ways. It isn't as simple as just plugging in a new stick of RAM or a CPU: the whole system needs to be considered all together just to get mediocre results as opposed to having it blow up in your face.
The apparent knowledge level regarding PC hardware on a forum like here on KVR isn't too high. Please keep that in mind. In other words if you honestly need to ask "more RAM or a better CPU?" you should be asking yourself another question like "should I do a bit more research and learn more before I make a good cost vs. benefit decision?"
Regarding #1 (PSU/wattage): consider for example the cheaper memory brands which tend to have big heatsinks on them. Compare to more expensive memory with smaller or no heatsink. Think about what that means. If you look at the specs you might find the expensive cards without heatsinks use a smaller transistor size or a variation on one of many modern "3d/stacked" semiconductor architectures, which end up requiring far less power to accomplish the same thing.
So in that case you might think "wow, those heatsinks are HUGE! SO AWESOMEEEEE, they must be the best ever! THINK OF ALL THE HEAT THEY CAN SINK!". That would be wrong. The biggest heatsink is in fact demonstrating the shortcomings of an inferior product.
-
- KVRAF
- 1929 posts since 4 Nov, 2004 from Manchester
Old Intel X58?aciddose wrote:My system uses 3 x 8 gb cards and is currently (while semi-idle, its typical usage with browsers + IDEs + media player + etc open) using 3.5 gb out of the 24 gb available.
Prices have been rising steadly for the past year or two due to a shortage coming out of the factories. Although if it is an X58 system, prices might just about be back on a parallel with 2010 right now after dropping from 2011 - 2015!aciddose wrote: This memory cost me only $250 when I bought it and I'd expect prices are similar now.
Oh, the absoutely is. The physcial slots and the support written into the BIOS for them are key. Not every board supports using bigger than 8GB sticks for instance.aciddose wrote: There is absolutely no limit on the amount of memory you can place in a system
A stick of memory takes a couple of watts at most, I've never seen anyone need to upgrade their PSU for a couple of sticks.aciddose wrote: [*]The WATTs required for that memory. Modern systems are highly efficient but more memory will require more power, even during idle/suspend! This can require doubling the capability of your PSU ($500-$1000) to cope with a few extra memory cards ($50-$200).
Yeah, It's normally the other way around, in that memory has heat spreaders stuck on and an additional 20% to the price to make it a premium product. Agreed with you about not needing them, they don't really do anything unless you're overvolting your memory, and you may as well just buy faster memory than do that because cheap and slow memory rarely if ever overclocks to any real degree these days.Regarding #1 (PSU/wattage): consider for example the cheaper memory brands which tend to have big heatsinks on them. Compare to more expensive memory with smaller or no heatsink. Think about what that means. If you look at the specs you might find the expensive cards without heatsinks use a smaller transistor size or a variation on one of many modern "3d/stacked" semiconductor architectures, which end up requiring far less power to accomplish the same thing.
So, in that case, you might think "wow, those heatsinks are HUGE! SO AWESOMEEEEE, they must be the best ever! THINK OF ALL THE HEAT THEY CAN SINK!". That would be wrong. The biggest heatsink is in fact demonstrating the shortcomings of an inferior product.
-
- KVRAF
- 12555 posts since 7 Dec, 2004
No, I'm not running Intel X58, whatever that is.
You might want to look in to the actual increase in power consumption on various mainboards based upon the memory and bus speeds. For example running at 1333 is not at all the same as a higher speed like 2400: there is a reason these will have heatsinks while the modern 1333 likely have none. Back in ~2008 when such memory first became available it tended to have massive heatsinks and much larger transistors with higher power consumption and waste. Due to issues with the underlying properties of these systems there are significant diminishing returns as clock rates and bus speeds are increased. We've for the past five years been in a lull with the upcoming ddr5 (~2021) possibly using smaller transistors in improved 3d patterns, but without significant changes to the entire architecture it simply isn't possible to improve bandwidth and density much further than small incremental steps.
(See discussion on "the end of the von Neumann architecture".)
The important thing is you'll never find such info anywhere in the advertising or technical specifications and datasheets for such cards. You will need to identify which factory actually manufactured the chips (often re-branded, not branded by the maker) and find matching datasheets and do a whole lot of math to actually work out all the real numbers.
Alternatively as most people who have experience in this field tend to do, it is possible to over-spec the entire system and plug in various cards while measuring the total PSU load during benchmarking. You can often find these numbers (totally unreliable, but trivial to measure in that unreliable way) published by those who do such benchmark tests which are unfortunately some of the most reliable numbers available.
You might want to look in to the actual increase in power consumption on various mainboards based upon the memory and bus speeds. For example running at 1333 is not at all the same as a higher speed like 2400: there is a reason these will have heatsinks while the modern 1333 likely have none. Back in ~2008 when such memory first became available it tended to have massive heatsinks and much larger transistors with higher power consumption and waste. Due to issues with the underlying properties of these systems there are significant diminishing returns as clock rates and bus speeds are increased. We've for the past five years been in a lull with the upcoming ddr5 (~2021) possibly using smaller transistors in improved 3d patterns, but without significant changes to the entire architecture it simply isn't possible to improve bandwidth and density much further than small incremental steps.
(See discussion on "the end of the von Neumann architecture".)
The important thing is you'll never find such info anywhere in the advertising or technical specifications and datasheets for such cards. You will need to identify which factory actually manufactured the chips (often re-branded, not branded by the maker) and find matching datasheets and do a whole lot of math to actually work out all the real numbers.
Alternatively as most people who have experience in this field tend to do, it is possible to over-spec the entire system and plug in various cards while measuring the total PSU load during benchmarking. You can often find these numbers (totally unreliable, but trivial to measure in that unreliable way) published by those who do such benchmark tests which are unfortunately some of the most reliable numbers available.
-
- KVRian
- 778 posts since 21 Apr, 2016
If you're maxing out your RAM, I'd just toss another 8 gig stick in.
Nobody, Ever wrote:I have enough plugins.
-
- KVRAF
- 1929 posts since 4 Nov, 2004 from Manchester
As far as I'm aware, the only chipset I could think of that was ever optimized for 3 memory sticks.aciddose wrote:No, I'm not running Intel X58, whatever that is.
I couldn't work out why you were running 3 sticks of memory, when boards are usually designed and optimized to use even numbers of RAM sticks.
You don't need heatsinks for 2400 memory, which is kind of my point. 2400MHz server memory which is designed to run 24/7 in cramped server racks, which absolutely kick out a ton of heat, never have heat spreaders.aciddose wrote: You might want to look in to the actual increase in power consumption on various mainboards based upon the memory and bus speeds. For example running at 1333 is not at all the same as a higher speed like 2400:
2400 consumer memory that sits in well-ventilated cases? Yeah, heatsinks on most of them.
Why? Nobody seemed to need them until everyone and their dog decided to put windows on the case side panels, now you can't seem to buy a pack without them.
Yes, if you go outside of the controller spec and insist on jacking up the voltage in order to hit some clock speed that the board considers overclocking it may pull more voltage and the sinks might help, or you could just buy better memory to start with. Going back to DDR 3 all those kits were 1.35v and the current DD4's are 1.2v so they should be kicking out less heat anyway and very, very few of those DDR3 packs ever had sinks on them.
Putting heatsinks on everything has just become expected, and its getting a bit out of hand now; even more so that they've start splattering LEDs all over them.
-
- KVRian
- 652 posts since 2 Mar, 2015 from UK
G.Skill do the best looking RAM ever seen that is proper RGB LED.Kaine wrote:As far as I'm aware, the only chipset I could think of that was ever optimized for 3 memory sticks.aciddose wrote:No, I'm not running Intel X58, whatever that is.
I couldn't work out why you were running 3 sticks of memory, when boards are usually designed and optimized to use even numbers of RAM sticks.
You don't need heatsinks for 2400 memory, which is kind of my point. 2400MHz server memory which is designed to run 24/7 in cramped server racks, which absolutely kick out a ton of heat, never have heat spreaders.aciddose wrote: You might want to look in to the actual increase in power consumption on various mainboards based upon the memory and bus speeds. For example running at 1333 is not at all the same as a higher speed like 2400:
2400 consumer memory that sits in well-ventilated cases? Yeah, heatsinks on most of them.
Why? Nobody seemed to need them until everyone and their dog decided to put windows on the case side panels, now you can't seem to buy a pack without them.
Yes, if you go outside of the controller spec and insist on jacking up the voltage in order to hit some clock speed that the board considers overclocking it may pull more voltage and the sinks might help, or you could just buy better memory to start with. Going back to DDR 3 all those kits were 1.35v and the current DD4's are 1.2v so they should be kicking out less heat anyway and very, very few of those DDR3 packs ever had sinks on them.
Putting heatsinks on everything has just become expected, and its getting a bit out of hand now; even more so that they've start splattering LEDs all over them.
-
- KVRist
- 114 posts since 2 Dec, 2015
It does amuse me that PC building has become so much about the look of internal components. And the manufacturers are of course happy to feed into it. "Looking for a video card to match the colour scheme of my motherboard" is the kind of thing people are seriously asking now.
Don't know if anyone remembers RAMBUS memory? Intel briefly had a line of chipsets using it in the early 2000's. That genuinely needed heatsinks - it ran very hot.
Don't know if anyone remembers RAMBUS memory? Intel briefly had a line of chipsets using it in the early 2000's. That genuinely needed heatsinks - it ran very hot.
