Bring on the benchmarks!
We pass judgment on the patched FX-8150
and Core i7-3820
For our CPU showdown, the pretty much used
the exact same hardware that we used in our Holiday 2011 CPU showdown. The AMD
tests were run on an Asus Crosshair V Formula board, the Core i7-2600K on a
Gigabyte GA-Z68X-UD3H-B3, and the LGA2011 part on an Asus P8X70 Deluxe.
The only differences from the tests run
several months ago were BIOS updates to the AM3+ and LGA2011 platforms.
Otherwise, we used Windows 7 Professional, a 150GB WD Raptor hard drive, 8GB of
DDR3/1600, and a stock GeForce GTX 580, all running the same drivers. For the
AMD rig, we reran our tests with both of the hotfixes installed.
GeForce
GTX 580
The verdict? It’s a mixed bag. But let’s
face it, we shouldn’t expect miracles in the first place. AMD has already said
the patch adds maybe 1 or 2 percent in some tests and nothing in others. That’s
about what we saw: Some of the tests offered maybe a couple of ticks of
additional performance, while others were the same. Still others showed the
pre-patched state being faster. Does that mean the patch is hurting
performance? No. more likely we’re seeing the normal variances in performance
that happen in any benchmark. We are, after all, talking about 1 to 2 points –
that’s pretty much within the margin of error for most tests. That’s likely to
disappoint the already dejected AMD faithful who were hoping the patch would
give more serious benefits, such as the 10 percent boost predicted from Windows
8, with its fully updated thread scheduler. The bad news is that even with a 10
percent bump in performance from Windows 8’s improved scheduler, it’s not
enough to put the chip in real contention with the Core i7-2600K, 2700K, and
certainly not the new Core i7-3820. Even worse, by the time Windows 8 hits the
shelves, Intel will likely have moved on to its 22nm Ivy Bridge CPUs. Ivy
Bridge won’t be a big jump forward on the x86 side of the fence, but it’s going
to be significant in terms of integrated graphics and power consumption. Still,
for anyone running an AMD FX chip, there’s no real reason not to install the
patch, lest you leave any additional performance on the table.
The real battle here is between the 2600K
and 3820 parts. When you consider that both chips use what are essentially the
same execution cores, any performance difference can be attributed to cache,
clock, and memory bandwidth differences. Some of it may also come from
motherboard differences, but there’s nothing to be done there, as both chips
require different motherboards.
We expected this showdown to be close, but
our tests, at least with this configuration, showed the Core i7-3820 to have
some pretty healthy leads in the benchmarks. It didn’t win across the board,
but we saw everything from virtual ties to 14 percent performance boosts. For
example, in Sony Vegas the Core i7-3820 achieved a considerable bump of 12
percent, in HandBrake about 8 percent, and in Valve’s particle test a 9 percent
lead over the 2600K. Where does the extra performance come from? Certainly part
of it is from the i7-3820’s 4 percent clock-speed advantage, but the rest we
legitimately give to the memory bandwidth and cache differences between the
chips. It’s clear to us that the Core i7-3820 is the winning part.
The big question is whether the 3820’s
better performance is worth the price differences elsewhere in the system.
That’s really up to you. We continue to think that LGA1155 is a pretty awesome
platform for a lot of people. It offers everything from $37 Celerons to $332
Core i7-2700K parts, and within a few months you’ll probably have Ivy Bridge
parts to pick from, too. Where Core i7-3820 takes over is at the high-end where
you truly need a six-core part or a ton of RAM. Where would you need those
six-cores or that much RAM? Work. If your day job is pushing pixels in video,
3D rendering, or any other app where more cores means more free time for you,
building on the LGA2011 platform makes sense. If you can’t afford a hexa-core
processor, the Core i7-3820 lets you build a box with plenty of headroom for
the future. We’d also select the Core i7-3820 if we were running a Radeon HD
7970 card. As the sole PCIs 3.0 card out today, we’d want that 10 percent
performance bump you get from the faster PCIe 3.0 interface. We’d also go with
the Core i7-3820 if we wanted to build either a tri-SLI or tri-CrossFireX rig,
where additional PCIe lanes matter.
Making the comparison
Core i7-2600K
Core
i7-2600K
§ Clock:
3.46GHz
§ Cores/
Threads: 4/8
§ PCMark7
Score: 3,450
§ PCMark7
Lightweight: 2,612
§ PCMark7
Productivity: 2,269
§ Cinebench
10 Single-core: 6,011
§ Cinebench
10 Multi-core: 23,315
§ Cinebench
11.5: 6.84
§ POV
Ray 3.7 (sec): 218.93
§ Bibble
(sec): 137
§ Fritz
Chess Benchmark: 13,065
§ Intel
Burn Test (GFlops): 89.7
§ Sony
Vegas Pro 10 (sec): 2,752
§ ProShow
Product (sec): 1,063
§ MainConcept
(sec): 1,120
§ CyberLink
Espresso 6.5 using CPU (sec): 379
§ CyberLink
Espresso 6.5 using GPU (sec): 329
§ 7-Zip
64MB load, 12 threads (MIPS): 19,046
§ 7-Zip
64MB load, max core threads (MIPS): 19,288
§ wPrime
4-thread, 1024M: 349.27
§ wPrime
6-thread, 1024M: 293.16
§ wPrime
8-thread, 1024M: 248.3
§ wPrime
12-thread, 1024M: 271.2
§ HandBrake
(sec): 336
§ Sandrea
(GB/s): 17.6
§ Valve
particle Test (fps): 179
§ Dirt
2 (fps): 189
§ Far
cry (fps): 202.39
§ 3DMark2011
Score: P6469
§ 3DMark2011
GPU: 6,186
§ 3DMark2011
Physics: 8,184
§ 3DMark2011
Combined: 6,671
Core i7-3820
Core
i7-3820
§ Clock:
3.6GHz
§ Cores/
Threads: 4/8
§ PCMark7
Score: 3,443
§ PCMark7
Lightweight: 2,598
§ PCMark7
Productivity: 2,306
§ Cinebench
10 Single-core: 6,209
§ Cinebench
10 Multi-core: 24,456
§ Cinebench
11.5: 7.46
§ POV
Ray 3.7 (sec): 202.01
§ Bibble
(sec): 129
§ Fritz
Chess Benchmark: 14,190
§ Intel
Burn Test (GFlops): 87
§ Sony
Vegas Pro 10 (sec): 2,448
§ ProShow
Product (sec): 1,004
§ MainConcept
(sec): 730
§ CyberLink
Espresso 6.5 using CPU (sec): 309
§ CyberLink
Espresso 6.5 using GPU (sec): 306
§ 7-Zip
64MB load, 12 threads (MIPS): 21,485
§ 7-Zip
64MB load, max core threads (MIPS): 21,590
§ wPrime
4-thread, 1024M: 353.5
§ wPrime
6-thread, 1024M: 265.5
§ wPrime
8-thread, 1024M: 228.1
§ wPrime
12-thread, 1024M: 245
§ HandBrake
(sec): 311
§ Sandrea
(GB/s): 37.84
§ Valve
particle Test (fps): 196
§ Dirt
2 (fps): 188.8
§ Far
cry (fps): 220
§ 3DMark2011
Score: P6653
§ 3DMark2011
GPU: 6,267
§ 3DMark2011
Physics: 9,477
§ 3DMark2011
Combined: 6,759
FX-8150 Patched
AMD
FX-8150
§ Clock:
3.6GHz
§ Cores/
Threads: 8
§ PCMark7
Score: 2,912
§ PCMark7
Lightweight: 2,152
§ PCMark7
Productivity: 1,903
§ Cinebench
10 Single-core: 4,006
§ Cinebench
10 Multi-core: 20,414
§ Cinebench
11.5: 6.01
§ POV
Ray 3.7 (sec): 213.1
§ Bibble
(sec): 128
§ Fritz
Chess Benchmark: 11,567
§ Intel
Burn Test (GFlops): 29
§ Sony
Vegas Pro 10 (sec): 3,312
§ ProShow
Product (sec): 1,164
§ MainConcept
(sec): 946
§ CyberLink
Espresso 6.5 using CPU (sec): 430
§ CyberLink
Espresso 6.5 using GPU (sec): 426
§ 7-Zip
64MB load, 12 threads (MIPS): 19,904
§ 7-Zip
64MB load, max core threads (MIPS):20,361
§ wPrime
4-thread, 1024M: 448
§ wPrime
6-thread, 1024M: 353
§ wPrime
8-thread, 1024M: 293
§ wPrime
12-thread, 1024M: 322
§ HandBrake
(sec): 347
§ Sandrea
(GB/s): 17.6
§ Valve
particle Test (fps): 107
§ Dirt
2 (fps): 115.4
§ Far
cry (fps): 107
§ 3DMark2011
Score: P6146
§ 3DMark2011
GPU: 6,194
§ 3DMark2011
Physics: 6,335
§ 3DMark2011
Combined: 5,580
FX-8150 Pre-patch
§ Clock:
3.6GHz
§ Cores/
Threads: 8
§ PCMark7
Score: 2,838
§ PCMark7
Lightweight: 2,222
§ PCMark7
Productivity: 1,957
§ Cinebench
10 Single-core: 4,080
§ Cinebench
10 Multi-core: 20,277
§ Cinebench
11.5: 5.79
§ POV
Ray 3.7 (sec): 213.08
§ Bibble
(sec): 136
§ Fritz
Chess Benchmark: 11,704
§ Intel
Burn Test (GFlops): 29
§ Sony
Vegas Pro 10 (sec): 2,663
§ ProShow
Product (sec): 1,171
§ MainConcept
(sec): 902
§ CyberLink
Espresso 6.5 using CPU (sec): 429
§ CyberLink
Espresso 6.5 using GPU (sec): 412
§ 7-Zip
64MB load, 12 threads (MIPS): 20,400
§ 7-Zip
64MB load, max core threads (MIPS): 20,773
§ wPrime
4-thread, 1024M: 469
§ wPrime
6-thread, 1024M: 357.07
§ wPrime
8-thread, 1024M: 293.7
§ wPrime
12-thread, 1024M: 328.08
§ HandBrake
(sec): 345
§ Sandrea
(GB/s): 17.7
§ Valve
particle Test (fps): 108
§ Dirt
2 (fps):120
§ Far
cry (fps):111.23
§ 3DMark2011
Score: P6138
§ 3DMark2011
GPU: 6,167
§ 3DMark2011
Physics: 6,426
§ 3DMark2011
Combined: 5,569
