Wikipedia:Reference desk/Archives/Computing/2018 November 7
Computing desk | ||
---|---|---|
< November 6 | << Oct | November | Dec >> | November 8 > |
Welcome to the Wikipedia Computing Reference Desk Archives |
---|
The page you are currently viewing is a transcluded archive page. While you can leave answers for any questions shown below, please ask new questions on one of the current reference desk pages. |
November 7
editIntel Xeon versus i7
editComparing a single core, how would the performance of a 3.4GHz Intel Sandy Bridge i7 compare to a 2.4GHz Xeon E5-2695v2 ? This would be on integer operations only, often with a lot of reading and writing to memory. The i7 has a higher clock speed but the Xeon is a newer design with more memory cache. Bubba73 You talkin' to me? 05:29, 7 November 2018 (UTC)
- This can't really be answered without more information. Critical information would be: (1) the exact ratio between arithmetic operations (internal to the CPU) and memory accesses, (2) whether both systems have the same L2 cache size and design, (3) how the size of the working set of the program compares to that of the L1 and L2 caches, (4) if the working set is larger than the L1 cache, some analysis of the program's locality of reference would be needed to understand how much benefit the caches provides. This type of analysis is in most cases fairly difficult to do analytically; it's much easier to just test the program on both architectures. CodeTalker (talk) 16:23, 7 November 2018 (UTC)
- I can't really directly compare the two because I have some Sandy Bridge i7s but I'm thinking about buying a Xeon workstation. The mix of integer operations and memory operations will vary from program to program, but the memory-intensive stuff mainly accesses the memory in sequence. The i7s have the standard caches. The workstation I'm considering has two 12-core Xeon E5-2695v2 running at 2.4GHz, 64GB of ECC RAM, but I don't know about the memory cache sizes. Bubba73 You talkin' to me? 02:16, 8 November 2018 (UTC)
- Depends what you are doing but I've generally found passmark (cpubenchmark.net) to be reasonably predictive. 173.228.123.166 (talk) 20:49, 10 November 2018 (UTC)
- Thanks - is that Passmark score running on one core or all of the CPU's cores? It gives 8,196 for the Sandy Bridge i7-2600 and 15,847 for the Xeon E5-2695 v2. That is 1.93x for the Xeon, but it has 12 cores compared to 4. If that is per core, that is great for the Xeon, but if it is for the whole CPU, that isn't so great for the Xeon (which has 3x as many cores). That 1.93x ratio is pretty close to 2.4*3/3.4 = 2.12, which would indicate that a Xeon doesn't quite core doesn't quite match a i7 core, per GHz, and the Xeon runs at a lot lower clock speed. Bubba73 You talkin' to me? 04:06, 14 November 2018 (UTC)
- Well, I found this, which says that it runs on all available CPUs. Which means that the 12-core Xeon E5-2695 v2 is a little less than the equivalent of 8 Sandy Bridge i7 cores. Bubba73 You talkin' to me? 04:12, 14 November 2018 (UTC)