What are the performance and architectural differences between PCIe and QPI? - cpu-architecture

PCIe 3.0 x16 and QPI 1.1 (20 lanes) have identical effective bandwidth (16 GB/s). So, I wanted to get a rough picture about the differences between the two.
What are the differences between the two in terms of latency and message rate (number of packets or TLPs per second)? For latency, my ballpark numbers are 20 ns for QPI and 200 ns for PCIe 3.0. Are these good estimates? If yes, why is PCIe's latency so much higher - is it due to the wire length?
Are there significant architectural differences between the two, apart from the fact that QPI provides cache snooping? To my knowledge, both use a layered protocol: transport layer through physical layer.

The two have fairly different messaging types due to their different roles. QPI is directly concerned with implementing cache coherency via the MESIF protocol and NUMA via a distributed directory. PCIe has no such notions, although they share in common memory read and write and completion message types (see here for some PCIe basics). They have similar power states and a priority scheme implemented through virtual channels. Both use credit-based flow control but there's no guarantee of any commonality in what sort of credits are maintained by QPI versus PCIe endpoints (as far as I know, the specifics of QPI credits are an Intel trade secret).
Message rate for each is usually expressed in GT/s. Typical QPI rates are 4.8, 6.4 and 8 GT/s, and 5 or 8 GT/s for PCIe.
Your latency estimates for both are probably low. QPI is on the order of a few hundred ns per hop. Note that 4+ socket systems may have more than one QPI hop between pairs of sockets. PCIe might be closer to 500ns, although again, that depends on the system topology. Latency between a processor socket's main memory and a PCIe card hanging directly off that socket (the PEG slot) is going to be lower than between that same memory and a card hanging off the south bridge.

Related

How many clock cycles do the stages of a simple 5 stage processor take?

A 5 stage pipelined CPU has the following sequence of stages:
IF – Instruction fetch from instruction memory.
RD – Instruction decode and register read.
EX – Execute: ALU operation for data and address computation.
MA – Data memory access – for write access, the register read at RD state is
used.
WB – Register write back.
Now I know that an instruction fetch, for example, is from memory which can take 4 cycles (L1 cache) or up to ~150 cycles (RAM). However, in every pipelining diagram, I see something like this, where each stage is assigned a single cycle.
Now, I know of course real processors have complex pipelines with over 19 stages and every architecture is different. However, am I missing something here? With memory accesses in IF and MA, can this 5 stage pipeline take dozens of cycles?
Classic 5-stage RISC pipelines are designed around single-cycle latency L1d / L1i, allowing 1 IPC (instruction per clock) in code without cache misses or other stalls. i.e. the hopefully common / good case. Every stage must have a worst-case critical path latency of 1 cycle, or trigger a stall.
Clock speeds were lower back then (even relative to 1 gate delay) so you could get more done in a single cycle, and the caches were simpler, often 8k direct-mapped, single port, sometimes even virtually tagged (VIVT) so TLB lookup wasn't part of the access latency.
First-gen MIPS, the R2000 (and R3000), had on-chip controllers1 for its direct-mapped PIPT split L1i/L1d write-through caches, but the actual tags+data were off-chip, from 4K to 64K. Achieving the required single-cycle latency with this setup limited clock speeds to 15 MHz (R2000) or 33 MHz (R3000) with available SRAM technology. The TLB was fully on-chip.
vs. modern Intel/AMD using 32kiB 8-way VIPT L1d/L1i caches, with at least 2 read + 1 write port for L1d, at such high clock speed that access latency is 4 cycles best-case on Intel SnB-family, or 5 cycles including address-generation. Modern CPUs have larger TLBs, too, which also adds to the latency. This is ok when out-of-order execution and/or other techniques can usually hide that latency, but classic 5-stage RISCs just had one single pipeline, not separately pipelined memory access. See also Cycles/cost for L1 Cache hit vs. Register on x86? for some more links about how performance on modern superscalar out-of-order exec x86 CPUs differs from classic-RISC CPUs.
If you wanted to raise clock speeds for the same transistor performance (gate delay), you'd divide the fetch and mem stages into multiple pipeline stages (i.e. pipeline them more heavily), if cache access was even on the critical path (i.e. if cache access could no longer be done in one clock period). The downside of lengthening the pipeline is raising branch latency (cost of a mispredict, and the amount of latency a correct prediction has to hide), as well as raising total transistor cost.
Note that classic-RISC pipelines do address-generation in the EX stage, using the ALU there to calculate register + immediate, the only addressing mode supported by most RISC ISAs build around such a pipeline. So load-use latency is effectively 2 cycles for pointer-chasing, due to the load delay for forwarding back to EX.)
On a cache miss, the entire pipeline would just stall: those early pipelines lacked scoreboarding of loads to allow hit-under-miss or miss-under-miss for loads from L1d cache.
MIPS R2000 did have a 4-entry store buffer to decouple execution from cache-miss stores. (Apparently built from 4 separate R2020 write-buffer chips, according to wikipedia.) The LSI datasheet says the write-buffer chips were optional, but with write-through caches, every store has to go to DRAM and would create a stall without write buffering. Most modern CPUs use write-back caches, allowing multiple writes of the same line without creating DRAM traffic.
Also remember that CPU speed wasn't as high relative to memory for early CPUs like MIPS R2000, and single-core machines didn't need an interconnect between cores and memory controllers. (Although they maybe did have a frontside bus to a memory controller on a separate chip, a "northbridge".) But anyway, back then a cache miss to DRAM cost a lot fewer core clock cycles. It sucks to fully stall on every miss, but it wasn't like modern CPUs where it can be in the 150 to 350 cycles range (70 ns * 5 GHz). DRAM latency hasn't improved nearly as much as bandwidth and CPU clocks. See also http://www.lighterra.com/papers/modernmicroprocessors/ which has a "memory wall" section, and Why is the size of L1 cache smaller than that of the L2 cache in most of the processors? re: why modern CPUs need multi-level caches as the mismatch between CPU speed and memory latency has grown.
Later CPUs allowed progressively more memory-level parallelism by doing things like allowing execution to continue after a non-faulting load (successful TLB lookup), only stalling when you actually read a register that was last written by a load, if the load result isn't ready yet. This allows hiding load latency on a still-short and fairly simple in-order pipeline, with some number of load buffers to track outstanding loads. And with register renaming + OoO exec, the ROB size is basically the "window" over which you can hide cache-miss latency: https://blog.stuffedcow.net/2013/05/measuring-rob-capacity/
Modern x86 CPUs even have buffers between pipeline stages in the front-end to hide or partially absorb fetch bubbles (caused by L1i misses, decode stalls, low-density code, e.g. a jump to another jump, or even just failure to predict a simple always-taken branch. i.e. only detecting it when it's eventually decoded, after fetching something other than the correct path. That's right, even unconditional branches like jmp foo need some prediction for the fetch stage.)
https://www.realworldtech.com/haswell-cpu/2/ has some good diagrams. Of course, Intel SnB-family and AMD Zen-family use a decoded-uop cache because x86 machine code is hard to decode in parallel, so often they can bypass some of that front-end complexity, effectively shortening the pipeline. (wikichip has block diagrams and microarchitecture details for Zen 2.)
See also Modern Microprocessors
A 90-Minute Guide! re: modern CPUs and the "memory wall": the increasing mismatch between DRAM latency and core clock cycle time. DRAM latency has only dropped a little bit (in absolute nanoseconds) as bandwidth has continued to climb tremendously in recent years.
Footnote 1: MIPS R2000 cache details:
An R2000 datasheet shows the D-cache was write-through, and various other interesting things.
According to a 1992 usenet message from an SGI engineer, the control logic just sends 18 index bits, receiving a word of data + 8 tags bits to determine hit or not. The CPU is oblivious to the cache size; you connect up the right number of index lines to SRAM address lines. (So I guess a line-size of one 4-byte word?)
You have to use at least 10 index bits because the tag is only 20 bits wide, and you need tag+index+2(byte-in-word) to be 32, the physical address-space size. That sets a minimum cache size of 4K.
20 bits of tag for every 32 bits of data is very inefficient. With a larger cache, fewer tag bits are actually needed, since more of the address is used up as part of the index. But Paul Ries posted that R2000/R3000 does not support comparing fewer tag bits. IDK if you could wire up some of the address output lines to the tag input lines, to generate matching bits instead of storing them in SRAMs.
A 32-byte cache line would still only need 20-bit tags (at most), but would have one tag per 8 words, a factor of 8 improvement in tag overhead. CPUs with larger caches, especially L2 caches, would definitely want to use larger line sizes.
But you're probably more likely to get conflict misses with fewer larger lines, especially with a direct-mapped cache. And the memory bus can still be busy filling a previous line when you encounter another miss, even if you have critical-word-first / early-restart so the miss latency wasn't worse if the memory bus was idle to start with.

Does memory copying on APUs (e.g. apple m1 mac) use GPU-specific wide vector instructions?

I was reading this article Why mmap is faster than system calls, where the main difference appeared to be mmap's ability to use vector instructions like AVX-2, something system calls can't.
I understand that the SIMD instructions used by GPUs tend to be much wider. A Nvidia warp of size 32 operating on float32 = 1024 bits (?) vs 256 bits of AVX-2. So potentially a 4x speedup. I guess this is not used in traditional discrete gpu settings as host-to-device (and back) copy would outweigh any benefit from wide registers.
However in APUs, GPU shares memory with CPU, eliminating the need for these expensive copies. I was wondering if those GPU instructions can therefore be used to accelerate mmap like vector operations further (numpy is another example). Has it already been done (in M1 mac or any CPUs with integrated graphics)? or can you please detail the architectural issues that prevent this?
You're kind of asking 2 separate questions: whether an OS (or user-space standard libraries?) can use GPGPU to speed up reading from the pagecache (into user-space memory with a read system call, or from an mmaped region). And separately whether GPGPU on normally-allocated process memory (and/or the pagecache) can avoid a copy to memory dedicated to the GPU.
For the 2nd part Apple has said the answer is yes for MacOS on M1 thanks to making the integrated GPU's memory accesses cache-coherent with the CPU. I think AMD made similar suggestions that copying could be avoided in graphics or GPGPU drivers on their APUs (Fusion IIRC?), but IDK if software ever took full advantage.
For the first part; doubtful. Large memory copies are bottlenecked by DRAM bandwidth, not CPU-core <-> L1d cache bandwidth (which scales with SIMD register width). On x86, an AVX2 loop on a single core can come pretty close to maxing out the DRAM bandwidth of an Intel "client" chip (quad-core or similar, not a big xeon with a higher-latency interconnect). Single-core bandwidth (to L3 or DRAM) tends to be limited by the number of outstanding cache misses that a core can track, not by doing the copy with fewer instructions. That mostly helps in terms of seeing farther with the same size out-of-order execution window, to start page walks sooner across page boundaries and stuff like that. See Why is std::fill(0) slower than std::fill(1)? for SSE (16-byte) vs. AVX (32-byte) vectors.
GPU offload would thus not help for large copies. It could only possibly help for small copies, and then it would not leave the copy result hot in L1d cache of the CPU. And/or not be able to take advantage of the source or destination already being hot in L1d cache of a CPU working with the data.
Also, setup overhead (to communicate with the GPU, going outside the current core) would dominate any faster copying for small copies.

perf power consumption measure: How does it work?

I noticed that perf list now has the option to measure power consumption. You can use it as follows:
$ perf stat -e power/energy-cores/ ./a.out
Performance counter stats for 'system wide':
8.55 Joules power/energy-cores/
0.949871058 seconds time elapsed
How accurate is this measurement, and how does perf estimate the power consumption?
The power/energy-cores/ perf counter is based on an MSR register called MSR_PP0_ENERGY_STATUS, which is part of the Intel RAPL interface (Intel seems to call each individual RAPL MSR a RAPL interface). A complicated model based on system activity events is used to estimate (static and dynamic) energy consumption. The MSR register name has PP0 in it, which refers to power plane 0, which is one of the RAPL domains that contains all the cores of the socket including the private caches of the cores. PP0, however, excludes the last-level cache, the interconnect, the memory controller, the graphics processor, and everything else that is in the uncore. It's impossible to measure the accuracy of MSR_PP0_ENERGY_STATUS because there is no other way to estimate the energy consumption of power plane 0 only.
It's possible to measure the accuracy of other RAPL domains though. These include the Package, DRAM, and PSys domains. For example, the accuracy of the Package domain energy estimation can be measured by comparing against the energy consumption of the whole system (which can be measured using a power meter) and running a workload that keeps the energy consumption of everything outside the package a known constant as much as possible. The accuracy of MSR_PKG_ENERGY_STATUS and MSR_DRAM_ENERGY_STATUS have been measured in different ways by different people on many different processors. You can refer to the recent paper entitled RAPL in Action: Experiences in Using RAPL for Power Measurements for more information, which also includes summaries of previous works. The paper covers Sandy Bridge, Ivy Bridge, Haswell, and Skylake. The conclusion is that MSR_PKG_ENERGY_STATUS and MSR_DRAM_ENERGY_STATUS appear to be accurate on Haswell and Skylake (the implementation has changed on Haswell, see : An Energy Efficiency Feature Survey of the Intel Haswell Processor). But this is not necessarily true on all kinds of workloads, P states, and processors. So the accuracy does not just depend on the microarchitecture.
The RAPL interface is discussed in Section 14.9 of the Intel Manual Volume 3. I noticed there are errors in the section. For example, it says client processors don't support the DRAM domain, which is not true. The client Haswell processor I'm using to write this answer supports the DRAM domain. The section is probably outdated and applies only Sandy Bridge and Ivy Bridge processors. I think it's better to read the datasheet of the processor on which you want to use RAPL.
The power/energy-pkg/ perf counter can be used to measure energy consumption of the package domain. This is the only domain that is known be supported on all Intel processors starting from Sandy Bridge.
On x86 systems, these values are based on RAPL (Running Average Power Limit) - an interface that provides built in CPU energy counters. While originally designed by Intel, AMD also provides a compatible interface on Zen systems.
The accuracy depends on the actual microarchitecture. Originally, RAPL was backed by a model with certain biases. On Intel CPUs since the Haswell architecture, it is based on measurements which are quite accurate. As far as I know there is no good understanding of the accuracy on AMD's Zen RAPL implementation.
One important thing you have to consider is the scope of the measurements. On most systems, only package and DRAM is covered1. So if you need to know how much power / energy your entire system consumes - you usually cannot easily answer that with RAPL.
Also note that RAPL is updated every 1 ms, so short workloads will have significant errors from the update rate.
1 - Skylake Desktop systems can implement a full-system RAPL. It's accuracy depends on the manufacturer.

Why are CPU registers fast to access?

Register variables are a well-known way to get fast access (register int i). But why are registers on the top of hierarchy (registers, cache, main memory, secondary memory)? What are all the things that make accessing registers so fast?
Registers are circuits which are literally wired directly to the ALU, which contains the circuits for arithmetic. Every clock cycle, the register unit of the CPU core can feed a half-dozen or so variables into the other circuits. Actually, the units within the datapath (ALU, etc.) can feed data to each other directly, via the bypass network, which in a way forms a hierarchy level above registers — but they still use register-numbers to address each other. (The control section of a fully pipelined CPU dynamically maps datapath units to register numbers.)
The register keyword in C does nothing useful and you shouldn't use it. The compiler decides what variables should be in registers and when.
Registers are a core part of the CPU, and much of the instruction set of a CPU will be tailored for working against registers rather than memory locations. Accessing a register's value will typically require very few clock cycles (likely just 1), as soon as memory is accessed, things get more complex and cache controllers / memory buses get involved and the operation is going to take considerably more time.
Several factors lead to registers being faster than cache.
Direct vs. Indirect Addressing
First, registers are directly addressed based on bits in the instruction. Many ISAs encode the source register addresses in a constant location, allowing them to be sent to the register file before the instruction has been decoded, speculating that one or both values will be used. The most common memory addressing modes indirect through a register. Because of the frequency of base+offset addressing, many implementations optimize the pipeline for this case. (Accessing the cache at different stages adds complexity.) Caches also use tagging and typically use set associativity, which tends to increase access latency. Not having to handle the possibility of a miss also reduces the complexity of register access.
Complicating Factors
Out-of-order implementations and ISAs with stacked or rotating registers (e.g., SPARC, Itanium, XTensa) do rename registers. Specialized caches such as Todd Austin's Knapsack Cache (which directly indexes the cache with the offset) and some stack cache designs (e.g., using a small stack frame number and directly indexing a chunk of the specialized stack cache using that frame number and the offset) avoid register read and addition. Signature caches associate a register name and offset with a small chunk of storage, providing lower latency for accesses to the lower members of a structure. Index prediction (e.g., XORing offset and base, avoiding carry propagation delay) can reduce latency (at the cost of handling mispredictions). One could also provide memory addresses earlier for simpler addressing modes like register indirect, but accessing the cache in two different pipeline stages adds complexity. (Itanium only provided register indirect addressing — with option post increment.) Way prediction (and hit speculation in the case of direct mapped caches) can reduce latency (again with misprediction handling costs). Scratchpad (a.k.a. tightly coupled) memories do not have tags or associativity and so can be slightly faster (as well as have lower access energy) and once an access is determined to be to that region a miss is impossible. The contents of a Knapsack Cache can be treated as part of the context and the context not be considered ready until that cache is filled. Registers could also be loaded lazily (particularly for Itanium stacked registers), theoretically, and so have to handle the possibility of a register miss.
Fixed vs. Variable Size
Registers are usually fixed size. This avoids the need to shift the data retrieved from aligned storage to place the actual least significant bit into its proper place for the execution unit. In addition, many load instructions sign extend the loaded value, which can add latency. (Zero extension is not dependent on the data value.)
Complicating Factors
Some ISAs do support sub-registers, notable x86 and zArchitecture (descended from S/360), which can require pre-shifting. One could also provide fully aligned loads at lower latency (likely at the cost of one cycle of extra latency for other loads); subword loads are common enough and the added latency small enough that special casing is not common. Sign extension latency could be hidden behind carry propagation latency; alternatively sign prediction could be used (likely just speculative zero extension) or sign extension treated as a slow case. (Support for unaligned loads can further complicate cache access.)
Small Capacity
A typical register file for an in-order 64-bit RISC will be only about 256 bytes (32 8-byte registers). 8KiB is considered small for a modern cache. This means that multiplying the physical size and static power to increase speed has a much smaller effect on the total area and static power. Larger transistors have higher drive strength and other area-increasing design factors can improve speed.
Complicating Factors
Some ISAs have a large number of architected registers and may have very wide SIMD registers. In addition, some implementations add additional registers for renaming or to support multithreading. GPUs, which use SIMD and support multithreading, can have especially high capacity register files; GPU register files are also different from CPU register files in typically being single ported, accessing four times as many vector elements of one operand/result per cycle as can be used in execution (e.g., with 512-bit wide multiply-accumulate execution, reading 2KiB of each of three operands and writing 2KiB of the result).
Common Case Optimization
Because register access is intended to be the common case, area, power, and design effort is more profitably spent to improve performance of this function. If 5% of instructions use no source registers (direct jumps and calls, register clearing, etc.), 70% use one source register (simple loads, operations with an immediate, etc.), 25% use two source registers, and 75% use a destination register, while 50% access data memory (40% loads, 10% stores) — a rough approximation loosely based on data from SPEC CPU2000 for MIPS —, then more than three times as many of the (more timing-critical) reads are from registers than memory (1.3 per instruction vs. 0.4) and
Complicating Factors
Not all processors are design for "general purpose" workloads. E.g., processor using in-memory vectors and targeting dot product performance using registers for vector start address, vector length, and an accumulator might have little reason to optimize register latency (extreme parallelism simplifies hiding latency) and memory bandwidth would be more important than register bandwidth.
Small Address Space
A last, somewhat minor advantage of registers is that the address space is small. This reduces the latency for address decode when indexing a storage array. One can conceive of address decode as a sequence of binary decisions (this half of a chunk of storage or the other). A typical cache SRAM array has about 256 wordlines (columns, index addresses) — 8 bits to decode — and the selection of the SRAM array will typically also involve address decode. A simple in-order RISC will typically have 32 registers — 5 bits to decode.
Complicating Factors
Modern high-performance processors can easily have 8 bit register addresses (Itanium had more than 128 general purpose registers in a context and higher-end out-of-order processors can have even more registers). This is also a less important consideration relative to those above, but it should not be ignored.
Conclusion
Many of the above considerations overlap, which is to be expected for an optimized design. If a particular function is expected to be common, not only will the implementation be optimized but the interface as well. Limiting flexibility (direct addressing, fixed size) naturally aids optimization and smaller is easier to make faster.
Registers are essentially internal CPU memory. So accesses to registers are easier and quicker than any other kind of memory accesses.
Smaller memories are generally faster than larger ones; they can also require fewer bits to address. A 32-bit instruction word can hold three four-bit register addresses and have lots of room for the opcode and other things; one 32-bit memory address would completely fill up an instruction word leaving no room for anything else. Further, the time required to address a memory increases at a rate more than proportional to the log of the memory size. Accessing a word from a 4 gig memory space will take dozens if not hundreds of times longer than accessing one from a 16-word register file.
A machine that can handle most information requests from a small fast register file will be faster than one which uses a slower memory for everything.
Every microcontroller has a CPU as Bill mentioned, that has the basic components of ALU, some RAM as well as other forms of memory to assist with its operations. The RAM is what you are referring to as Main memory.
The ALU handles all of the arthimetic logical operations and to operate on any operands to perform these calculations, it loads the operands into registers, performs the operations on these, and then your program accesses the stored result in these registers directly or indirectly.
Since registers are closest to the heart of the CPU (a.k.a the brain of your processor), they are higher up in the chain and ofcourse operations performed directly on registers take the least amount of clock cycles.

Communication between processor and high speed perihperal

Considering that a processor runs at 100 MHz and the data is coming to the processor from an external device/peripheral at the rate of 1000 Mbit/s (8 Bits/Clockcycle # 125 MHz), which is the best way to handle traffic that comes at a higher speed to the processor ?
First off, you can't do it in software. There would be no way to sample the digital lines at a sufficient rate, or to doing anything useful with it.
You need to use a hardware FIFO buffer or memory cell. When a data burst comes in, it can be buffered in the high speed FIFO and then read out as needed by the processor.
Drop in high speed FIFO chips are surprisingly expensive (though most are dual ported). To cut cost, you would be best off using an SRAM chip, and a hardware adder to increment the address lines on incoming data.
This is not an uncommon situation for software. semaj said the right word. This is a system engineering issue. Other folks have the right answer too. If you want to look at or process that data with the 100MHz processor, it is not going to happen, dont bother trying. You CAN look at snapshots of it or have the hardware filter out a specific percentage of it that you are looking for. At the end of the day though it is a systems issue, what does the hardware provide, where does it put this data, what is the softwares task for this data, does it see X buffers of data come in on the goesinta, and the notify the goesouta hardware that there are X buffers ready to go? Does the hardware examine and align the buffers so that you can look at a header, and then decide where to route the hardware? Once you do your system engineering you will know if you can use that processor or not, and if you can use it what its job is and how to do it.
Your direct question. What is the best way to handle it. The best way to handle it is to have hardware (fpga, asic, etc) move it into and out of some storage device (ram of some sort probably). Not necessarily the same ram the processor runs out of (DMA is a good thing to avoid). The hardware is something the software can talk to but you cannot examine all of that data so dont try. Without knowing what kind of data this is, what form, what the software looks at how much work you are willing to force the hardware to do, etc determines the rest of the answer. If you expect a certain (guaranteed) percentage to be bad or not belong to this processor, etc have the hardware filter that out and then what is left you can process.
Networking is a good example of this, PCs have gige ports but cannot process GigE line rate data. That is why we use switches now instead of hubs, hardware slices out a percentage of the data so the pc can handle it, the protocols take care of the data that cannot be processed by resending it later. And the switches processors dont look at all of the data, the hardware slices it up so the software can examine just the header. Or sometimes the software simply manages tables that drive the hardware and the hardware does all the work of processing the data.
Do your system engineering the answers will simply fall out.
You buffer it. Typically data from a device is written to a memory buffer (circular queue) using DMA (no cpu involved). The cpu reads from the memory buffer at a constant rate. Usually devices send data in bursts. This keeps the buffer from filling up. If there is too much data, buffer overflow.
DMA (direct memory access) is possibly the solution, however, it seems unlikely that the memory bus could run faster than the processor core, so the receiving peripheral would have to accept data into a larger register than 8 bit because 125MHz could not be sustained. For example a 16bit register would allow memory writes at 62.5MHz which may be achievable. Also the receiving device would have to be able to accept an external clock that is both faster and asynchronous to the core clock. Also of course the receiving peripheral must have support for DMA.
Unless you are more specific about your hardware and the communication protocol it is difficult to give anything other than a general answer.