Piledriver doesn't have "weak" IPC at all as it's IPC is as good as Nehalem's.
Piledriver's IPC is on par with Bulldozer's. See here:
The performance data in that article was measured with Turbo Mode turned *off* for all processors in order to take a closer look at IPC and efficiency.
Piledriver's improved performance in general single-threaded benchmarks (when Turbo Mode is enabled) comes from two improvements:
1) The chip spends more time in Turbo mode than Bulldozer did. BD ran closer to its envelope and had a harder time actually utilizing Turbo mode.
2) Higher clock speeds.
Once you normalize against clock speed and deactivate Turbo Mode, Piledriver's improvements are on the order of 5-7%. It's not as strong a chip as Nehalem. The Thuban-based Phenom II hits a single-threaded score of 1.0x at 3.3GHz with its Turbo Mode off. The Nehalem-based Core i7-970 scored ~1.2 in the CB 11.5 test (at 3.46GHz).
The FX-8350, with Turbo Mode enabled, only scores a 1.1. That's a 4.2GHz clock speed.
If the Core i7-970 is pulling a 1.2 at 3.46GHz while the FX-8350 does 1.10 at 4.2GHz, it indicates that Nehalem has a significant IPC advantage over AMD's Piledriver.
The Anandtech benchmarks do indeed show a stock A8-5600K beating a Q6600. Your point is taken. While the A10-5600K would still be more expensive than $50-$70 for a new Q6600, it would be substantially less expensive than upgrading to an Intel Core i7 quad-core + new mobo + RAM.
Well, thanks for all that. Didn't really keep up with AMD tech to know about those as I only checked out one or two FX-8350 reviews, BUT, it still has slightly improved IPC over Bulldozer. I'd say it's somewhere in between Nehalem and Yorkfield, which is somewhat respectable.
Anyway, I personally could care less about all that TBH. This Z77 Extreme9 + 3770K is so good I'm treating it like my baby. The Extreme9 has a maximum OC button in it's UEFI which just OC's the *** out of the CPU to 4.8 with a single click... and the setup is so much more fluid, stable and efficient than that old, fat and hot ass QX9650 which barely did 3.6 after a good amount of setting and tweaking...
anyway, what I was, from the beginning, saying is that Q6600 really can't be the way to go when setups like that FM1 can be had. Hell, even second hand *** like Phenom 2 hardware from eBay would be easily better than a Q6600. The problem with Q6600 is mostly people on an as old as E6600 chips have old boards and platforms that can now be considered "shitty" that it'd, honestly, IMHO, just be better to upgrade the whole thing when you can do it for so cheap:
Also, that is NOT Nehalem. Your comparison is false since you're comparing a 6-core, 12 thread Gulftown to the PileDriver. It's easily a superior chip. The i7 920 is a Nehalem and has worse single-threaded performance, less cores and less OC'ability potential.
You really ought to deepen your understanding of these topics if you want to debate them cogently.
Westmere's benefits over Nehalem were:
Hardware AES instructions.
Improved virtualization performance.
A greater number of cores (6, instead of 4).
Higher clock speed.
Outside of these specific areas, Westmere's IPC was *not* significantly higher than Nehalem's. Benefits were in the 3-4% range, tops.
I didn't say Westmere's IPC was significantly higher than Nehalem's. I know it was just about that much better as it didn't offer way better single threaded performance when it came down to the real usage of it.
What I was saying is that, you were comparing Cinebench scores of Gulftown to Nehalem, which is somewhere the Gulftown, definately WOULD, at whichever percent, come on top of Nehalem.
And I know that from first hand experience since I had an E5620 on an X58 UD9 (which unfortunately shorted out and died the day I booted it up) and it, generally, was barely any faster than an i7 920. Was mostly a give or take.
It doesn't matter, in this instance. Let me show you why, using Anandtech's database.
Remember, we're looking for IPC, so clock speeds matter. The simplest way to get a comparable metric is to divide the score by the clock speed. The trick here is that we have to use Turbo Speeds. This figure gives us a steady comparison between clock speed and Cinebench 11.5 performance.
Core i7-3770K = 0.000425.
Core i7-2600K = 0.00040.
Core i3-3220 = 0.000415 (IVB part, slightly more efficient than the Core i7-2600K).
Core i7-990X = 0.000346 (Westmere)
An overclocked Core i7-920 at 3.8GHz turns in a Cinebench 11.5 score of 1.39. You can check that here:
Core i7-920 OC'D = 0.000365
Core i7-990X OC'D = 0.000345
Why is Nehalem slightly faster than Westmere's single-threaded performance when overclocked? Let's call that a test errata. But we've got a steady pattern here. Intel's single-threaded efficiency in CB11.5 has improved 23% between Westmere and Ivy Bridge.
Now let's compare AMD. We're back to Anandtech's numbers.
FX-8350 (Piledriver) = 0.000261
FX-8150 (Bulldozer) = 0.000261
A10-5800K = 0.000257
A8-3850 = 0.000306.,
X6-1100T = 0.000308
Phenom II 975 BE = 0.000307
And there you have it. Once we normalize clock speed and compare performance-per-MHz, you see that 1) the FX-8350 isn't any more efficient than the FX-8150 in single-threaded tests and 2) AMD is so far behind Intel as far as single-threaded efficiency, it obviates the advantage of overclocking.
And how it doesn't matter? Do you realize the 970 has a stock clock of 3.2 agaist the 2.66 of the 920, right? And that it can easily pull a 4.5 whereas the 920 barely does 4.2? And that it also has slightly better per Mhz performance?
That comparison you did there is absurd. There's no way for an OC'ed i7 920 to beat a 990X. The 990 can easily do 5 Ghz abouts that there's no *** way for a 920 to beat it out. The test is either errored out as you said, or maybe even MADE UP to shill that pre-built machine which got into the test.
It's more like a difference between test setups that made the difference. You're comparing things from different sites... mind you, the test benches have to be the EXACT SAME for them to be 100% accurate.
I admit that I didn't know PileDriver's IPC was about the same as Bulldozer's and that it was relying heavily on Turbo speeds. And that's because I personally didn't care about any of that.
Regardless, an OC'ed Westmere will come on top of an OC'ed Nehalem %90 of the time. The other %10 is very rare cases where the 920 does 4.5-5 Ghz. And yeah, AMD's in-game performance doesn't look too good either. Even when things are threaded, the FX-8150 still used to get spanked by a QX9770...
You don't seem to understand what a single-threaded test *is*.
Single-thread. One core.
If Westmere and Nehalem have the same architecture as far as Cinebench 11.5 is concerned (and they do), then Westmere and Nehalem will perform identically once clock speed variation is taken into account. It doesn't matter that Westmere has 6 cores and Nehalem has 4. We're testing ONE core.
Also? I wrote that review. Implying that I made up the benchmark results is insulting and inaccurate.
Finally, Cinebench 11.5 is widely quoted because it's a simple test. A Core i7-3770K will perform the same in Cinebench 11.5 regardless of RAM loadout, RAM speed, or QPI linkages.
The data I've just showed you compared performance in Cinebench as a function of clock speed. The stock clocks are irrelevant.
Look at the Core i7-3770K vs. the Core i3-3220. The Core i7-3770K scores 0.000425 points per MHz. The Core i3-3220 scores 0.000415 points per MHz. That difference (2.4%) is either margin of error or due to different cache sizes. The point is that the scaling is *consistent.*
And *because* scaling is consistent, we can get a sense for the relative performance of other architectures. That's how we see that IPC between AMD chips and Nehalem is still quite wide.
What I was saying is, Gulftown/Westmere is EASILY a better chip than Nehalem when you put it on the run in real World conditions. Gulftown will even beat Sandy Bridge when a game scales across all 6 threads very well. Nehalem isn't as good of a chip as Westreme is and that's a fact.
Now, what I was implying is, when you put the FX-8350 on real world gaming benches, it WILL be up there with an i7 920 as far as performance in concerned. The reason being turbo speeds, IPC whatever. I'm talking about the end result.
But yes, you're correct that AMD's single threaded, or even multi-threaded performance isn't anywhere as good as Intel's and in no line I despited that. I agree about that because even in some completely threaded apps the FX-8150 still loses to a Sandy Bridge chip. You don't need to repeat so basic things as if I'm a moron over and over again.
I wrote a crap ton, and decided to summarize.
I think we can all agree that once you enter a scenario such as the one in reference, the cost per performance ratio is the most important thing, or if you plan on never playing another game, getting the maximum level of performance in that particular game is the most important. Yes, you may hit a bottleneck, but any budget PC is going to have a bottleneck somewhere. What's important is gauging your level of satisfaction after installing your "new" hardware, and this article shows that you indeed could find a satisfactory level of gaming by simply upgrading to a new GPU.
Those are some solid numbers I can sit behind... though I'm still sitting on a 6-core, so I'm not too far behind.
This is valuable information for those us making the most of our systems, and do not upgrade so often. My Q6600 system is still running from 2008 and will be upgraded when Haswell arrives.
Overclocked to 3.52GHz, I too have noticed significant performance increases going from a GTX 460 to a GTX 570 and GTX 660. At 99% GPU load and 39% to 47% oc on CPU, the following were my results in The Witcher 2 on high settings at 1080p:
3.33GHz CPU -- GTX 460 (763MHz): 25fps
3.33GHz CPU -- GTX 570 (732MHz): 44fps
3.52GHz CPU -- GTX 570 (950MHz): 49fps
3.52GHz CPU -- GTX 660 (1124MHz): 44fps
Currently only getting 6060 in 3DMark11 on the GTX660, but once haswell is released, I expect up to another 33% increase in performance from better overall system throughput, based on other people's results with i7-3770k's. Also my ram is 2x4GB DDR3 running at 1174MHz with timings 5-6-6-14-2T, running on an ASRock P35 board. Frankly, I'm impressed at how game-able the system still is at 1080p on high settings in BF3, Crysis 3, and Borderlands 2.
I would like to add that I got the GTX 570 for $100 (friend), and the GTX 660 (ASUS DC2O) for $120 (online deals), so definitely was worth throwing them into an older system while I wait for Haswell.
NEWS TIPS |
This site is intended for informational and entertainment purposes only. The contents are the views and opinion of the author and/or hisassociates. All products and trademarks are the property of their respective owners. All content and graphical elements areCopyright © 1999 - 2013 David Altavilla and HotHardware.com, LLC. All rights reserved. Privacy and Terms