With the release of P55 and Lynnfield, it begged the question of how they stack up to X58 and Bloomfield. Are socket 1156 Core i7s worthy of being called Core i7?
The curiosity was too much and I decided to find out, since I had one of each setup on hand. I began running benchmarks with the idea of doing a direct comparison of how Core i7 P55/1156 would compare if Core i7 X58/1366 were “crippled” to run at Lynnfield limits, namely in dual-channel and at 18x QPI.
There’s nothing very scientific here, just some comparative tests started on a whim. What it’s not, is a comparison of newly released hardware against some tests in the bank from many months ago, when the X58s and 920s were new. Some of these were done back-to-back, the others within a few days. The important thing is that they were all done on the same cooling and ambient temps, HDD, OS, video card(s), memory, etc. The only things that changed between setups were the motherboards and CPUs.
| Socket 1156 | Socket 1366 |
CPU | Core i7 870 | Core i7 920 |
Motherboard | Gigabyte P55-UD4P | Gigabyte EX58-UD4P |
BIOS | F4m | F7 |
Same hardware/software used in both setups | ||
DDR3 | 2x2GB (and 3x2GB) from a Kingston KHX16000D3ULT1K3/6GX kit | |
Video card(s) | Gigabyte GTX 260 SP216 – stock cooling | |
HDD | VelociRaptor 300GB | |
PSU | Odin 1200W | |
OS | Windows XP SP3 32-bit / Windows 7 RC Build 7100 32-bit (for 3DMark Vantage only) | |
Video Drivers | Forceware 190.62 | |
Cooling | NexXxos XP waterblock, 2x BIP II radiators in series, 1x MCP-355 pump |
Many comparisons run 1156 and 1366 systems at defaults. The goal of this testing was to match Lynnfield and Bloomfield systems as closely as possible and see how they really compare on a level field. To do so, the P55 and X58 systems were overclocked equally. Both systems were setup as follows:
- 21x 200 BCLK boot & bench (920 = 20x + Turbo, 870 = down clocked to 21x). This gave a final CPU clock of 4.2 GHz each with matched BCLK
- All cores + HT were kept on for all benches
- Uncore set to 18x on both. This is the highest setting for P55/870 and lowest setting for X58/920
- QPI set to 18x on both. This is fixed on P55/870 and an underclock from the default 19x on X58/920 (and adjustable lower and higher)
- Memory had manually set primary timings and tRFC, everything else was left on AUTO. While few secondary timings differed between boards, the differences were minor and Roundtrip Latency was identical on both with the BIOSes used: Channel A=54 and Channel B=56. For tri-channel testing on the X58, RTL still remained on AUTO, which defaulted to A=57, B=59, C=60
- All EIST, Speed Step, C-State functions, etc, were disabled
- All voltages (including Vcore, Vdimm, Vtt, IOH/PCH, CPU PLL, etc.) were left on AUTO;
- PCIe frequencies were left on AUTO
- Video card clocks were the same saved profile and clocks were kept purposely low (700/1509 (linked)/1200) to avoid heat causing inconsistencies on their stock cooling
Here are the head-to-head results in a nutshell. This is as closely matched as both setups can be (save from also manually matching all memory secondary timings). For the average user, when X58/Bloomfield are matched to P55/Lynnfield limits, both setups are indeed very close in performance.
P55 Lynnfield 870 | X58 (@ P55 limits) Bloomfield 920 | |
Mobo/CPU setup | 21x 200, 18x Uncore, 18x QPI | |
Memory | Dual-channel, 800 MHz 6-7-6-20 1T | |
3DMark03 | ||
3DMark05 | ||
3DMark06 | ||
3DMarkVantage | ||
SuperPI 32M | 8m 49.437s (529.437s) | 8m 43.218s (523.218s) |
wPrime 32M | ||
wPrime 1024M |
Click on links for images
It is fairly clear that even when reduced to P55/Lynnfield limits, X58/Bloomfield still maintains a performance edge. One possible cause may be the immature P55 BIOS. Gigabyte has always been good about continuously updating BIOS and it is no different for P55. Even since this testing was started, new beta BIOS releases for the P55-UD4P have increased performance and narrowed the differences seen here.
The newest beta BIOS was NOT used for these tests however, since they had already been started with the older beta. While the performance improvement was not huge (ie. 1.5s faster in 32M and +90 points higher in 3D06), it is measurable and a step in the right direction. Future improvements may narrow the gap even further.
The difference in SuperPi 32M is one of particular interest as the benchmark relies almost entirely on CPU speed and bandwidth/latency. Given the same CPU speeds and memory speed/timings (and same OS, HDD, etc.) on both setups, a nearly 6 second difference is fairly big and indicative that “something” in X58/Bloomfield is just inherently faster.
One possibility is internal timings. Another indication that P55/Lynnfield may be a bit “looser” is that both the Gigabyte P55-UD4P and the P55-UD6, as well as the Core i7 860 and 870 that I have tried in them can very easily out-clock the X58/Bloomfield in the memory department. Even in dual-channel mode, it takes a fair amount of work to get 1000MHz 7-8-7 benchable with these sticks in the EX58-UD4P, where it is effortless in both P55 boards.
More to the point, P55/Lynnfield can clock the memory much, much higher with some extra voltage. This is a good thing since P55/Lynnfield performance definitely benefits from the extra memory speed/bandwidth.
Crossfire / SLI
P55-UD4P PCIe lanes are 16x for the primary slot with a single card and 8x+8x when running cards in first and second PCIe slots. X58-UD4P is, of course, 16x for a single card in the primary PCIe slot, but remains 16x+16x for two cards. As proven time and again, on mid-range video cards (like the GTX 260s used here), the extra lanes on the X58 make little or no difference, but if multi-GPU cards were used for Crossfire or SLI, the limited PCIe lanes of the P55 would be a bottleneck.
If you plan on running multi-GPU video cards in Crossfire or SLI, X58 is the obvious choice to get all the video card performance you are paying for. For us mere mortals with mid-range single-GPU cards, P55 scales just as well as X58.
Click on links for images
As seen in the chart above and as expected, SLI scaling is similar between both systems. For the most, X58/Bloomfield carry over their lead from the single card results in overall points, but there is some difference in scaling. The 3DMark06 run just might have been a little “off” since it’s the only one X58 isn’t ahead in. There is no reason to think P55 scales better in only that bench and is much more likely that a normal variation in score bounced in favor of the P55. Either way, it’s close and for all intents and purposes, they scale pretty much the same.
It is already evident that even cut down to P55/Lynnfield limits, the X58/Bloomfield still performs better overall, if not by much in most cases. You are paying for the added performance features with X58/Bloomfield, so let’s find out how much the margin increases taking advantage of what X58 offers.
In the following tests, as before, the same hardware and overclocks are used, but this time tri-channel memory was used in the X58 and the QPI was kept at the default of 19x instead of the 18x limit of P55. The memory was clocked to the same speed and timings (800 MHz 6-7-6-20), however, as a reminder, in tri-channel mode Roundtrip Latency was reduced to Channel A=57, Channel B=59 and Channel C=60 by the AUTO settings.
The X58 default configuration benches were run with SLI only due to time constraints. Scores from all tests are shown in the following chart for easy comparison:
| P55 Lynnfield 870 | X58 (@ P55 limits) Bloomfield 920 | X58 (default config) Bloomfield 920 |
Mobo/CPU setup | 21x 200, 18x Uncore, 18x QPI | 21x 200, 18x Uncore, 19x QPI | |
Memory | Dual-channel 800 MHz 6-7-6-20 1T | Tri-channel 800 MHz 6-7-6-20 1T | |
3DMark03 |
| ||
3DMark03 SLI | |||
3DMark05 |
| ||
3DMark05 SLI | |||
3DMark06 |
| ||
3DMark06 SLI | |||
3DMarkVantage |
| ||
3DMark Vantage SLI | |||
SuperPI 32M | 8m 49.437s (529.437s) | 8m 43.218s (523.218s) | 8m 34.312s (514.312s) |
wPrime 32M | |||
wPrime 1024M |
Click on links for images
These results are what you could expect if you took a P55 Core i7 system (first results column) and an X58 Core i7 (last results column), clocked them identically and took advantage of the tri-channel and higher default QPI offered on the X58/Bloomfield system. Naturally, the X58 increases its margin in this scenario and is just meant to show the extra performance you are paying for with it.
Miscellaneous
Prior to even laying my hands on P55/Lynnfield, I (and probably most of you) might have heard stories of them beating X58/Bloomfiled in game benchmarks. Looking at the results from all of the synthetic benchmarks that I ran, that seemed odd. Since there wasn’t time to add gaming benchmarks to the battery of tests that I did, I cheated and just went and looked at other comparisons of the P55/X58.
It seems that for every claim of P55/Lynnfield beating X58/Bloomfield in game benches, they were run at default clocks on both systems, usually pitting a 2.8GHz Core i7 860 against a 2.66 GHz Core i7 920. Mystery solved. Regardless, it appears that neither system was ahead or behind much, which would fall in line with what the 3DMark benches show here.
In terms of overclocking, the 1156 Core i7s that I have seem very similar to 1366 Core i7s (at least D0 steppings) in that 4 GHz seems like it will be easily doable with “livable” temperatures on aftermarket air cooling, even more so on water. How much over 4 GHz one may see will come down to cooling, a little luck and how much voltage one is willing to risk.
Conclusion
Everyone will need to reach their own conclusions on this one. The results might make it easier or more difficult for someone to decide between P55/Lynnfield and X58/Bloomfield as their next upgrade. When clocked identically, they are close in performance as one might expect. The tough decision will be how much extra will you be willing to shell out for the added performance?
Overall, P55/Lynnfield is cheaper (even now), performs very similar and uses less power to do it. It will likely become even more of a bargain the further we move away from the recent launch and prices decline.
On the other hand, X58/Bloomfield is still king if you need every last drop of performance and every single point on benchmarks.
Be the first to comment