GDDR vs pc ram vs ssd

[Deleted member 194470]

Hello,

I am curious about the changes in memory speed and what people here think about it. I have ddr 4 sdram 2400 with a mt/s speed of 2400 and the top speed of ddd4 is 3200. While the top speed of graphics memory is 14,000 mt/s(geforce 1660 super) and the lowest 16 series card the 1650 has a speed of 8000 mt/s. So this means graphics memory is more advance and faster than pc ram. Why has pc ram not caught up? Would it be possible for hardware companies to reengineer graphics memory so it can be used as pcram? I was not able to figure out the equivalent of mt/s to mb/s but I have a wd sn750 with 3,500 read and 2,600 write speeds. Is this about as fast as my ram? In the future as ps5 proposes ssd can be used instead of ram?

It seems that pc-ram is the slower type of storage now or at least is not as advance as graphics memory, where in the past it was the fastest to make up for slower storage speeds(hard drives).

Thanks let me know what you think.

 

[BEP]

GPU memory is different from CPU memory because they do different jobs and are connected differently. GPU GDDR memory is based on DDR memory and contrary to what you think even older versions of DDR (GDDR4 and 5 is based on DDR3) so if I were to judge them based solely on that fact, DDR is superior to GDDR because GDDR is based on earlier versions of DDR right? Or if I were to look at raw speed GDDR is faster and thus GDDR is better than DDR right?

Well, as I said they do different jobs and are connected differently, GPU memory is connected to PCIe bus and have more bandwidth than CPU and are differently interconnected for the job they do, think of it as brute raw bandwidth, lots of chunks of data, less precision, latency, accuracy, but more sheer speed, power, bandwidth so that's why they are faster, CPU memory, on the other hand, have better latency, precision, but less raw speed/bandwidth because they work with many small files rather than big chunks, that's the gist of it and the main reason of differences anyways.
Also, this is the reason GPUs are used for large-scale brute force attacks to, let's say, do data breaches and steal user info (usernames, passwords, credit card info,etc.) their goal is not to tweak around 1 user try to guess his password or whatever (in which CPUs would be more useful), nope, their goal is to attack and get as many results as possible while imputing random words (well not entirely random, they use common words, phrases, combinations they also keep from other attacks on other sites) till they get the password right so they are not using precision, accuracy, latency or whatever, just sheer raw brute force (in this case the bandwidth GPUs have) instead of CPUs so for hacking brute force, yes GPU superior, but in general it depends on what you need to do, and CPU RAM is not inherently 'superior' or 'inferior' to GPU RAM, just different.

And that was it about CPU RAM (DDR) vs GPU RAM (GDDR), and now for storage, well, of course, there's HDDs limited by being mechanical (rotating platter, moving heads), SSD's which are better because they don't rely on mechanical movement, and when this SSDs are connected to SATA 2, they suffer some performance loss and when connected to SATA 3, they are no longer bottlenecked by the connections, however, even faster than these are NVMe SSDs that are connected using PCIe bus that we've talked about earlier, so more bandwidth right? Well, these NVMe SSDs use that to their advantage to reach even higher speeds than normal SATA 3 SSDs by using that bandwidth like how GPU RAM is faster when compared to CPU RAM.
So that bandwidth, as earlier explained, doesn't make the GPUs superior when compared to CPUs, they each have their own different job to do, however, more bandwidth does make NVMe SSDs > SATA 3 SSDs > SATA 2 SSDs > SATA 2 HDDs > SATA 1 HDDs.

Hope I helped a bit, and even though I wrote quite a bit I tried keeping it as compact as I could, and if this interests you more, do look more into it because what I wrote are just summaries.

 

[Vya Domus]

The difference isn't that big, GPUs simply have wider buses. GDDR memory is optimized for throughput and DDR for latency, that's the gist of it.

 

[Rei]

As @Vya Domus said about wider bus. 64-bit has been the cap for system RAM,& the current way to expand that is by multi-channeling RAM. Wider bus is possible on GPU due to the RAM built into the GPU board among other things.

MT/s & MB/s is not really the same thing. MT/s has yet to take into account the bus width which will make up the total speed in MB/s or GB/s. So your system RAM's total speed is 19.2GB/s in single channel or 38.4GB/s in dual channel mode. That is still hella faster than any SSD to date.

 

[Deleted member 194470]

Oh I see thanks for the detailed response Bep. I was wondering if it had to do also with the architecture of the memory and how it is used. Isn't part of the reason graphics cards are good for bruteforce too because of the cuda cores and equivalent for amd? I never really understood what they are exactly, are they actually 'cores', I was reading about it once and it just seemed to be how they described it, they are really shaders? Do games actually use the cuda or its just for applications. I would like to use the full power of my video card which is impressive considering all that it but I just use video games.

Ram is faster than the ssd, that's what I thought. I couldn't find the equivalent storage speed I sought. Maybe in any case developres will find a better use for ssd, until I updated my storage I just saw hard drives as the slowest part of the computer and there was nothing you could do about it. Also isn't the add in pci cards faster than nvme? LIke intel optane or is that just the type of memory but I guess if it plugs into the pci-e slot then it is the same type of interface. Optane is the fastest ssd right?

 

[BEP]

It's all about specialized stuff being better at specialized things like for example, you don't NEED a separate PC to stream or host a game server, for example, especially if you have a good PC you can host/stream on just one PC, BUT it would still help to have 1 server PC 1 stream PC 1 game PC, segregating jobs to different specialized builds and making sure you have the internet bandwidth, of course, will help. So that's why GPU's appeared in the first place, first there were just CPUs, for command-line OS-es and the like or even very simple menu-based, but as they got more advanced they started using "2d graphics accelerators" connected to AGP (Accelerated Graphics Port) and the likes, my first card was a Matrox with 8Mb VRAM. So basically CPU did most stuff and the "graphic accelerators" helped with rendering the desktop and took that specialized job over from CPUs shoulders. And skip to today where GPUs are much more than just helpers for desktop visuals, they now have specialized bits like you mentioned: Cuda Cores/ Streaming Processors, ROPs (Raster Operators), Texture Units, and whatever else. Each having a specialized role in GPUs that are a specialized role in your PC. And yes, these are used wherever needed, but developers also have a role in how or what they use (bot apps and games or whatever else) ROPs at rasterization, texture units at resizing and placing textures on 3d objects, Cuda Cores / Streaming Processors (Shaders) at light and colors, etc.

Yes RAM > SSD by design RAM is primary memory while SSDs/HDDs are secondary, by placement they are second place, what about VRAM though? Well if RAM is primary memory and Storage is secondary memory, VRAM is a buffer between CPU and display, but it's specialized just for the graphics job that's why it's hard to tell especially since this "primary/secondary" denomination is just for explaining that RAM is primary memory because its what CPU works with and is volatile memory, doesn't store data permanently, just for quick calculations and stuff, so secondary is storage, where you keep files/programs/etc. So VRAM is the same volatile memory, so in that sense, it's primary memory because you don't store programs in it, however, it's not primary in the sense that CPUs don't really need it for computing primary stuff, just displaying related stuff is offloaded to GPU so yeah.
And yes HDD slowest part of PC, that's why thanks to PCs the best upgrade for old PC is to buy an SSD for primary OS storage, that and making sure you running RAM in dual channel. (If you buy an already-built PC and it comes with 1 stick of RAM, buying a second identical one)

And "add in pci cards" are one of NVMe SSDs the other being M.2, while there are also add in U.2 SSds like the optane mentioned and all NVMe as earlier explained, uses PCIe bus instead of SATA bus thats why they are better. And yeah Intel Optane 905P up there in top with Samsung 970 Pro.

 

[Deleted member 194470]

my first video card was Riva TNT 2. I thought that was an amazing card and I overclocked and it started smoking. Remember voodoo. I liked the video cards back then in terms of competition, now you just have two. Maybe 3 with intel's new cards but their graphics chips always suck.

Is it possible to combine or make the os use graphics and ram. Like for example windows can use page file or a usb stick(although I know that is not used anymore). I know there is a program where someone installed a video game on ram and used it. I guess for me it doesn't really matter, my most intensive games might be use 3-4 gbs of ram and I have windows 7/10 using about 2 gb of ram as I slimmed it down. I never use anyway close to 16 gb, but it would be an interesting idea. But as your saying its a separate architecture so I guess not but is that hardware based or software, maybe you could have software that makes use of the ram for non-graphics related things or offload ram used for graphics onto the card assuming not all the ram is being used for non-display things.

Actually I think if I am not mistaken, aorus pcie gen4 nvme is the fastest ssd now. I did not know about dual channel when I made a new computer(it was a while since my last one) and so I just bought one chip but when I looked at benchmarks of dual channel versus non, the difference was miniscule. For games for example it was 1 fps. Although I made the ram faster by changing some settings in the bios, I forget what settings now, but I didn't overclock it. It made it about 10 percent faster.

 

[BEP]

Yeah all you could do along those lines is ramdisk, if you have tons of ram you can make part of them act as a disk, and would only be useful for, idk, if you have a workbench with lots of ram and wanna preview videos faster while working on editing them is one example. But that for ram, none for vram, afaik at least.