I've been getting the upgrade itch very badly recently. An 11400f or a 10600k, would certainly be a nice uplift from my 4790k. With my sensible hat on however, I wouldn't be able to make any use of it with my current 980Ti holding me back as it is, let alone with a new CPU. On top of that, I think the current crop of motherboards are way overpriced and I don't want to join the DDR4 club just before everyone else leaves it for DDR5 which is just round the corner. So, I'm going to put off upgrading my gaming PC for now. Instead, I decided to satisfy the upgrade itch by giving my X79 living room PC a bit of love. So, this evening, I installed this 8 core, 16 thread bad boy: It replaces the 6 core 12 thread E5-1650V2 (4930k equivalent) I had previously. To my pleasant surprise, it runs perfectly stable at exactly the same settings too. Meaning it's running at 4.2Ghz with 32GB of 2133Mhz RAM. I've been running some tests in games and I'm very surprised by the results. So the way I've done it is to lower the resolution down to 720p, find a particularly CPU limited area in the game and record the FPS using the frame counter built into my monitor. The only program running in the background is MSI afterburner to check that the GPU is running at less than 100% utilisation, ensuring there is a CPU bottleneck. I first ran the E5-1650v2 and then the E5-1680v2. First game is Skyrim, the original 2011 version. A game notorious for only using 2 threads. These readings were taken in Whiterun, standing at the top of the steps outside Dragonsreach and looking down on the rest of the town. E5-1650v2 --- 67-69 fps E5-1680v2 --- 74-76 fps Next up, Star Trek Online. This was taken standing inside the earth space dock, standing on the tip of the Starfleet logo outside the exchange and looking towards club 47. E5-1650v2 --- 82-83 fps E5-1680v2 ---83-84 fps Shadow of the tomb raider, probably the most multi threaded game on the list. This was taken standing in in Kuwaq Yaku. Screenshot for exact location: E5-1650v2 --- 69-71 fps E5-1680v2 --- 82-83 fps On the 1680v2, the CPU bottleneck was almost eliminated, I lowered the resolution scaling down further, and achieved 84-85fps with GPU usage at 96%. The newest game on the list is Star Was: Jedi: fallen order. That was taken just as you arrive on the planet Kashyyyk. E5-1650v2 --- 67-70 fps E5-1680v2 --- 77-80 fps Another old favourite next, Civilization V. This was taken using using an old save game file of a very late game and zooming out as far as possible. E5-1650v2 --- 147-150 fps E5-1680v2 --- 159-162 fps Last game, the ever popular GTA V. Taken somewhere in Los Santos. E5-1650v2 --- 120-130 fps E5-1680v2 --- 139-143 fps The 8 core seems to be a bit more stable too with less variation in the frame rate. I honestly wasn't expecting there to be much of a difference between the two, apart from in heavenly threaded workloads. But Skyrim, a game that only used two threads, saw a 10% improvement, and Shadow of the Tomb Raider saw a 17% improvement and became GPU limited at 720p. So what's going on? The only other difference between the two CPUs is cache. The 1650 has 12MB to the 1680's 25MB. Let's test the cache: E5-1650v2: (reading taken a few months ago with slightly different memory timings) E5-1680v2: As we can see, not only is there an increase in the size of L3 cache, but there is a very significant uplift in the speed of L1, L2 and L3 cashes too. This is where the performance increase in lightly threaded scenarios is coming from. Thanks for reading everyone. I'm off to play some games!
If you test Bf5, odyssey or vallhalla we'll probably see the performance increase is coming from higher core count as well. 6c/12 cpus have quite a high cpu usage in some games. Cyberpunk comes in mind. Anyway good stuff im also waiting for ddr5 goodness and probably zen4. My R7 2700x 4.35ghz is still good enough.
most 1680v2's will do 4.4ghz at 1.24v or lower (vccsa might need tweaking for aiming for 4.5 or 4.6, expect the voltage scaling to be dramatic. i don't know how 2 more cores, 4 more threads at the same clock speed would result in this More cache certainly would have little to no impact.
Well the only difference, other than the two extra cores, is the cache. I can see how a doubling in cache size and an increase in cache bandwidth would produce these results. Have a look at AnandTech's retrospective of Broadwell. It does extremely well in games compared to its contemporaries due to its extra cache. https://www.anandtech.com/show/1619...ective-review-in-2020-is-edram-still-worth-it
There is nothing particularly special about L4 cache that means that it will make a CPU instantly better in games. All cache is essentially the same thing. L4 just means that it is the fourth level of memory away from the cpu core. L3 cache is just a little closer to the CPU, L2 closer still and so on. The closer it is, the faster the access times. If added L4 cache improves performance then added L3 cache will do the same thing.
Just noticed that my STO results are invalid. I had the resolution up at 1440p so it was GPU bound. I thought this game would be completely CPU bound due to its age but apparently not. This is what the 1680v2 is doing at 720p No idea what the 1650v2 was doing, but I thought Id test my 4790k running at 4.7Ghz. Can't quite believe it myself but I've checked and rechecked this: How can this be happening? Surely the 4790K would thrash the 1680v2 just with its IPC advantage alone. Further investigation is required here I think...
So I've done quite a few tests now comparing the 4790k to the E5-1680v2. It seems that the 4.7Ghz Haswell i7 gets beaten by the 4.2Ghz Ivy-Bridge Xeon, or at least draws with it, in every game. Has anyone else had similar results? Seems bonkers to me. I've re-ran some synthetic tests (cinebench, CPU-z, cpumark99, SuperPi) and the Haswell is performing very well in those, smashing the E5-1680v2 in single threaded IPC, but for some reason games are running better on the xeon.
if any of these are avx optimized, keep in mind that IBE doesn't have an avx offset like haswell+ does.