So apparently a Ryzen 3900X is basically just two 3600X's put together. There's like 2 die sets called chiplets or something.... How is this different then the "modules" amd was using 4 or 5 years back? They had chips that were like 4 modules 8 cores and only 1 instruction could be in the modules pipeline at a time meaning 4 of the cores were useless..... Can anyone explain this? Maybe this was all wrong but I can't find anything that clearly explains how this works and.... And also what's the difference between dual-6-core chiplets and 1 12 core Chiplet? Anyone????
- Oct 15, 2014
- 18,739
- 2,732
- 104,090
sounds like you are trying to put a "one is better than the other" label on this. it's really not that simple.
every design has strengths and weaknesses. it's a matter of maximizing the strengths and minimizing the weaknesses. none of them are "better" designs but rather simply different. AMD has learned a lot about their method and make great strides in maximizing the strengths of their design. intel/arm and the others do it differently to play to the strengths of their designs.
in the end the important thing is how they perform. if you were into car racing and won all the time, would it matter if you did it with a v8 vs a v12?? or a single turbo vs a twin turbo design vs a massive supercharger?? would it matter if you liked larger tires than others? and so on and so on.....
what matters is how it does on the track and how it performs vs the competition. you're trying to second guess what AMD is doing for some reason. they chose what they chose and have caught up to what intel has been doing. so does it seem like their design is inherently weaker? must not be since they can do what the competition can and some.
are you trying to convince yourself that intel is better and are reaching for something to grasp onto for it to be true? so despite what the reviews are showing you can still say intel is better? it sure seems like this is where you are going since your trying to convince yourself that amd's approach is a bad design despite the performance facts........
every design has strengths and weaknesses. it's a matter of maximizing the strengths and minimizing the weaknesses. none of them are "better" designs but rather simply different. AMD has learned a lot about their method and make great strides in maximizing the strengths of their design. intel/arm and the others do it differently to play to the strengths of their designs.
in the end the important thing is how they perform. if you were into car racing and won all the time, would it matter if you did it with a v8 vs a v12?? or a single turbo vs a twin turbo design vs a massive supercharger?? would it matter if you liked larger tires than others? and so on and so on.....
what matters is how it does on the track and how it performs vs the competition. you're trying to second guess what AMD is doing for some reason. they chose what they chose and have caught up to what intel has been doing. so does it seem like their design is inherently weaker? must not be since they can do what the competition can and some.
are you trying to convince yourself that intel is better and are reaching for something to grasp onto for it to be true? so despite what the reviews are showing you can still say intel is better? it sure seems like this is where you are going since your trying to convince yourself that amd's approach is a bad design despite the performance facts........
mitch074
Splendid
- Mar 17, 2006
- 3,010
- 577
- 22,040
actually, it could be FASTER since they are throwing a crapload of L3 cache on each chiplet, cache that can only be directly addressed by up to 8 cores at a time. Something they would probably not do on a big monolithic die in fear of really horrendous yields.
The thing about using chiplets is to increase yields - limited room per wafer. If you make 16-core chiplets, you can make, say, 60 CPUs per wafer. if your wafer has 12 major defects, you only get 48 16-core CPUs (80% yield). If, on the other hand, you are making 8-core chiplets, that same wafer will give you (120-12) / 2 = 54 8*2 cores CPUs (90% yield).
If, on top of that, you take out the elements that don't really belong in the CPU (i.e. the I/O part of the chip, which actually used to be a separate component), you get a modular design with very few drawbacks and a crapload of advantages (modularity for one, better yields second, and the ability to manufacture a CPU across several plants without having to move them to the latest and greatest process).
You have to realize that the very same chiplet design covers the 6-core Ryzen R5 up to the 64-core EPYC Rome, only the I/O chip changes between all of these.
To do the same thing, Intel has to rely upon almost a dozen different core designs.
cryoburner
Judicious
- Oct 8, 2011
- 5,790
- 1,405
- 34,840
Actually, AMD isn't even using chiplets with 6 cores. Their current chiplets all have 8 cores, and they just disable a portion of each chiplet for the 6 and 12 core parts, whereas the 8 and 16 core parts use complete chiplets.
One advantage to this is that they only need one chiplet design to cover everything from their $200 Ryzen 3600 up to their multi-thousand dollar 64-core Epyc server processors. They can test the chiplets and put the best-performing or most efficient ones in the higher-end parts, and the ones that don't perform quite as well in the lower-end parts where those factors are not as important. Plus it gives them flexibility to manufacture a bunch of chiplets in advance while being able to put them into the products that are most in demand.
And if they were to go with a higher number of cores per chiplet, they would need to disable a large part of the chip to create the most common consumer parts, where 12+ cores isn't likely to be a necesity for some time, which would result in increased waste.
The chiplets themselves are primarily just the cores and cache for each processor, while other parts of the processor that allow it to do things like communicate with system memory and other components in the system have all been moved to a separate chip in the processor. One advantage of this is that AMD is able to use the older 14nm manufacturing process for that IO chip, since doing so doesn't have as much of an impact on performance. This allows them to save money and avoid shortages, since the 7nm manufacturing process is new and production is likely limited at this time, and allows them to focus on using that 7nm process where it makes the most difference, in the cores themselves.
One advantage to this is that they only need one chiplet design to cover everything from their $200 Ryzen 3600 up to their multi-thousand dollar 64-core Epyc server processors. They can test the chiplets and put the best-performing or most efficient ones in the higher-end parts, and the ones that don't perform quite as well in the lower-end parts where those factors are not as important. Plus it gives them flexibility to manufacture a bunch of chiplets in advance while being able to put them into the products that are most in demand.
And if they were to go with a higher number of cores per chiplet, they would need to disable a large part of the chip to create the most common consumer parts, where 12+ cores isn't likely to be a necesity for some time, which would result in increased waste.
The chiplets themselves are primarily just the cores and cache for each processor, while other parts of the processor that allow it to do things like communicate with system memory and other components in the system have all been moved to a separate chip in the processor. One advantage of this is that AMD is able to use the older 14nm manufacturing process for that IO chip, since doing so doesn't have as much of an impact on performance. This allows them to save money and avoid shortages, since the 7nm manufacturing process is new and production is likely limited at this time, and allows them to focus on using that 7nm process where it makes the most difference, in the cores themselves.
This question has NOTHING to do with intel. Im sitting on $3000 in parts, including an AMD motherboard, and waiting for my 3900X to be in stock ANYWHERE so I can finish the job. I'm just asking because I was taught some things many years ago, and maybe this was wrong. But I was told that when AMD first introduced their "module" system, around 5 years ago, that what was happening is that they would have 4 modules with 2 cores in each, and that only 1 instruction could be in the pipeline at a time, in each module, which would render 4 of the cores useless. The result is that these 8 core processors were really no better than standard quad cores. Perhaps what I was told at that time was incorrect. It makes sense that it could have been correct, but since changed. I'm trying to understand what they are doing differently now as apposed to at that time. I'm trying to understand if theres a performance difference between 1 chiplets, 12 cores, and 2 chiplets, 6 cores. Im trying to figure out the pros and cons of each. Im trying to figure out why AMD is doing this and other companies aren't.
- Dec 26, 2012
- 30,759
- 4,553
- 142,790
The answer is connections. In a block 2 cores wide, there's 2 ways to access it, 1 side and underneath. Doesn't matter if it's 4 cores deep, you still have access to at least 1 side. In a block 3 cores wide, at 9 cores you have a 'landlocked' center core. The only possible way to have communications is through other cores or from underneath, which restricts speed. A single die with 12 cores will have 2x center cores landlocked, plus all those cores have no other option but to radiate heat into other areas. Far better to have 2x seperate 6core blocks with easy access to communications than a single 12core block with 2cores getting the brunt of the radiated heat from the surrounding cores.
I'd not be surprised if Intel 10th gen opted for multiple chiplet designs when trying to go beyond a 2x4 core block instead of a 3x4 or 4x4 core block. The i9 10900x will most likely be a 2x 2x4 giving 16 cores/32 threads and still pulling the same 250w output.
I'd not be surprised if Intel 10th gen opted for multiple chiplet designs when trying to go beyond a 2x4 core block instead of a 3x4 or 4x4 core block. The i9 10900x will most likely be a 2x 2x4 giving 16 cores/32 threads and still pulling the same 250w output.
- Jun 19, 2011
- 33,688
- 1,700
- 159,590
AMDs modules from their older architectures and Ryzens chiplets are 2 different technologies. You are trying to compare apples and oranges. You have already been given several possible reasons why they have chose to use multiple dies vs single.
- Mar 16, 2013
- 179,651
- 22,020
- 184,590
Because ohter companies have different designs and architecture to achieve the same goals (mostly).
Because AMD does it this way does not make it the One True Way.
Nor does it make it substandard compared to any other mainstream CPU chip.
Simply...different.
TCA_ChinChin
Distinguished
- Feb 15, 2015
- 648
- 336
- 19,490
The chiplet approach is not that related to the module approach that AMD used with Bulldozer. The reason Bulldozer was so terrible was because of the way they handled the FPU, where two cores would share a single FPU, which would starve the CPU significantly in many workloads (as you mentioned). Ryzen and the chiplet approach does not have that flaw. The only drawback to the chiplet approach that AMD is currently using is increased latency between the chiplets themselves and also the I/O die. However, AMD has demonstrated that they have minimized this significantly and it does not have a noticeable impact in most applications (See any Ryzen 3000 review). The pros of using chiplets versus other companies (Intel) is that AMD can offer the same or more amount of cores for significantly lower prices, with only minimal impact to latency.
TRENDING THREADS
-
Question No POST on new AM5 build - - - and the CPU & DRAM lights are on ?
- Started by Uknownflowet
- Replies: 13
-
-
-
-
Question Spilled liquid on a ThinkPad P1 Gen 3 keyboard
- Started by Corobori
- Replies: 6
-
Facebook
Twitter
Instagram