GTX 970@1.5Ghz and 2500k vs Q9550 in 70+ benchmarks (+bonus i7-860)

Thanks for these tests. I appreciate the considerable time you must've spent compiling all of this.

Which is why I'm half kidding when asking if you might have picked up any stock scores. They just would be more meaningful to me personally.

Anyway good job!
 
Using the point for decimal separation is terrible.
The point is already used for finishing sentences, at some (unfortunate) places it's even used as a multiply operator like the "x" or "*" and other places use it for separation between hundreds and thousands.

At this point, 10.100 can be read as:
a) ten times one hundred (one thousand)
Not really. In mathematics, the point would be centered like ·, not on the bottom line like a period. Ditto for ×, vs. 'x'.
ToTTenTranz said:
b) ten thousand and a hundred
c) ten point one.

Using the point for decimal separation is no more confusing than using it to separate groups of 10^3, as you note in (b). If one sees "10,000.00" or "10.000,00" there is no confusion, but if one writes "10,000" or "10.000" it's ambiguous. Perhaps the decimal [overstrike]point[/overstrike]character should always be provided so you would write "10,000." or "10.000," to avoid confusion. Personally, I never liked the separating the 10^3 groups for this very reason.
 
So do you have to buy a special cpu with a FCU ? (Floating Comma Unit)
Yes, we do. And it's better than your point unit. Everyone knows that a comma is way cooler than a point. :D
But actually, latin based languages call it Floating Comma Unit, not Floating Point.
"Unidade de Vírgula Flutuante" in Portuguese, "Unidad de Coma Flotante" in Spanish, "Unità in Virgola Mobile" in Italian, "Unité en Virgule Flottante" in French, similar thing in Romanian..
And Germans call it "Gleitkommaeinheit", which I'm pretty sure it involves a comma and not a point ;)


Using the point for decimal separation is no more confusing than using it to separate groups of 10^3, as you note in (b).
Agreed. In my opinion, we should avoid using the point for anything at all in math.
More often than not, I have trouble writing an article where one sentence finishes with a number and the next sentence starts with another.
"The resulting study included a number of samples up to 100.432 is the total number of subjects."
Yeah...
 
It's like the selfie stick for cameras: many years ago, it was invented in Japan (obviously) and wildly ridiculed. And now it can be seen at any slightly popular tourist destination: it was a solution to a problem only visionaries understood existed!
So it will go with the comma displacing the point as decimal separator. [emoji3]
 
Really great tests. I just got a GTX 970 (msi) which I don't intend to overclock and my system has a standard i7 3rd gen vpro edition and 16GB of RAM. Maybe I will try to replicate one of these tests.

I think it will also be really interesting to redo the tests with DirectX 12.
 
Not to mention yards, furlongs, miles, nautical miles (as if things weren't confusing enough already), leagues, acres; plus units of other kinds, such as ounces, pints, gallons, pounds, stones, short tons, etc.

Actually I find cups/pints/quarts/gallons to be a rather elegant method of measurement needing little to no explanation. It's basically a power of 2 numbering system.

1 cup.
1 pint - 2 cups
1 quart - 2 pints - 4 cups
1 half gallon - 2 quarts - 4 pints - 8 cups
etc.

Going the other way, each step is half the previous one. It makes it easy to measure just about any coarse volume of liquid as it is all binary. And makes sense as it was a good way of measuring things when the majority of people couldn't count. As it's easy to go from one step to another it's either half or double.

Regards,
SB
 
Hello guys.

Thank you very much for you comments.

I am really sorry for the commas. Indeed fraps produces results with points and my excel is setup for my locale, which uses commas. But I thought that it would be clear for everyone what the scores mean. :)

As for the stock results that are missing, my thinking was that I wanted to see what is the max performance I could extract from the systems, both by their manufacturing process and architecture.

Also I am recording cpu usage and gpu usage throughout the benchmarks and I show them at the end of the videos, so if a game runs at 60fps for the Q9550, but it has 95% load, I would expect the stock clock to produce about 3/4 of that or less. I'd like to think these tests as a tool, so people can extract more info than just a sterilized result.

Also as I said, uploading takes too long for me and if I included stock clocks as well, I would another couple of months to finish this test. What I could do however and I will certainly entertain the idea, is provide stock clock result only and overclocked result with the video. This should be doable.

Again thanks for the ideas.
 
Really excellent work, pslord! This will actually get much better once Windows 10 is officially out and your charts reflect new differences/changes based on OS.

Keep up the good work!
 
So back on topic, having a new cpu makes little difference in most games

Actually, for some games the difference is huge. Look at Dying Light, DA: Inquisition and Grey Goo.
Battlefield 4 also shows a big difference, but at least the Core Quad is still doing >60FPS.
FrostByte 3 seems to not like the Conroe cousins at all.

Though I doubt there are man people using a Q9550 pushed all the way up to 4GHz. Perhaps using default clocks or maybe just a mild overclock would be more representative of what to expect?
Just a suggestion for next time.
 
Recalling back to my 4.2Ghz P4 days with an AGP 7950GT 24-pipe card, the CPU bottleneck was typically "solved" for me by continuing to crank on the graphics settings until I could drag the GPU far enough into the dirt to match my CPU output :D

"Oh look, 4xSSAA for free!" heh.

As for a "new" CPU, even the i5-2500k is going to perform just fine at stock speeds, and it's four years old and three generations behind at this point. I wager that the i7-870 will have quite good performance also, since it was the first desktop CPU to include all the PCI-E lanes directly within the uncore, which drove measurable performance gains specific to GPU usage.
 
Okay now reproduce this test on a 4770k @ ~4.6ghz and a 5960X @ ~4.0ghz. I'll even be nice and say just limit it to current gen games (ps4/xb1 ports or something relatively recent). :D
 
So back on topic, having a new cpu makes little difference in most games

It depends entirely on how CPU dependent a game is. Something like an RTS (LOTS of pathing calculations and AI and possibly more draw calls) is generally going to be far more CPU dependent than an FPS shooter. Something that makes use of a lot of physics (Mafia II is a great example with high PhysX versus no PhysX, an apples to apples comparison) is going to be more CPU dependent then something with simplistic physics.

On top of those generalities, a game may try to do more or less on the GPU versus CPU.

Regards,
SB
 
Hello my friends. Thank you for you comments again. :)

Since I have had some requests to post some i7-860 results already, here is the sum up of the 2500k, Q9550 and Core i7-860, on the most cpu limited benchmarks of the original benchmark suite.

Here is the list sorted by the 2500k vs Q9550 delta, from highest to lowest





and here are the same results, sorted by the 2500k vs core i7-860 delta, from highest to lowest.




As you can see, the 2500k/Q9550 delta dropped from 63% to only 13% for the 2500k/i7-860 comparison.

The clock difference of the 2500k@4.8Ghz and the i7-860@4.0Ghz is 4.8/4=20%, so for 20% more clock we only have 13% more performance. Of course this is due to the fact that this whole benchmark suite, is primarily designed to be GPU limited, with real life, very high settings, hence we have many GPU limits in all runs.

At this point, I believe it would be interesting to redo every 2500k benchmark that has a delta of 20% or more compared to the 860, which would correspond to the operating frequency difference of the two cpus, but this time clocked at 4Ghz, as the i7-860 is.

To stretch things even more, I could also disable the HyperThreading of the i7-860 for another run of these above 20% deltas, so we can see what a Lynnfield can do without the help of hyperthreading, but this is a story for another time I guess!

Of course I will come back to post the complete version of this test, with graphs, bbcode with video links etc. :)
 
Thanks! Very interesting stuff. Watch Dogs in particular caught my eye: it scales spectacularly from Penryn to Sandy Bridge (214.73%) but not at all from Lynnfield to Sandy. I wonder what that's about.
 
Remember that Lynnfield is when Intel moved the PCI Express controller "onboard" versus a chipset implementation. This radically drove down PCI-E latency; the i7-800 series of CPU's showed marked improvements for graphics intensive workloads even when compared to the i7-900 series CPUs of the day. Also, this is when Turbo first hit the market on the desktop side.

A bit of memory refresher: http://anandtech.com/show/2832/8

This may be why Lynnfield shows "abnormally" higher performance when compared against Penryn.
 
I recently put my HD 7950 (a great card) in my Q6600 based system. It's a Dell XPS 420, so no overclocking. Benching it in Crysis 2 reminded me of the moment that I thought it would appropriate to start planning on an upgrade. ;) Playing Far Cry 3 seems to strain the CPU when I come out of the safe house and walk around the village. Two of the cores use 80%, or better, and the other two of the cores use over 60%, IIRC. I suspect that represents an internal bottleneck of the CPU, as fed by the game engine.

In Crysis 2, using the Times Square scene, I see my gpu usually going well below 100%. Cranking up the AA changes very little in my FPS. I have the comparison from the i7 2600 in my newer machine to confirm that the HD 7950 is cpu bound with the Q6600.
I could never get Far Cry 3 to work on my new system. Point being that I can't compare.

Edit: Huh, I did some back of the envelope, rough, calculating, and it looks like my Q6600 might be sitting at around half the processing power of the overclocked Q9550. That would explain what I was wondering about. Many thanks for the chart!
 
Last edited:
Thanks! Very interesting stuff. Watch Dogs in particular caught my eye: it scales spectacularly from Penryn to Sandy Bridge (214.73%) but not at all from Lynnfield to Sandy. I wonder what that's about.

You are welcome my friend. :)

My Watch Dogs 970+860 was the last video of the queue of the above benchmarks and it just finished. I post the final version of the benchmark soon. Of course, Witcher 3 is coming, so at this point I might just wait to add this as well.

Here are the three runs of Watch Dogs. (spicy wallpaper alert :p)

Watch Dogs 1920X1080 Ultra GTX 970 @1.5Ghz Core i5 2500k @4.8GHz

Watch Dogs 1920X1080 Ultra SMAA GTX 970 @1.5Ghz CORE i7-860 @4GHz

[URL='https://www.youtube.com/watch?v=l15vaWF0fbE']Watchdogs 1920X1080 Ultra SMAA GTX 970 @1.5Ghz Q9550 @4GHz[/URL]

Also here is an attempt for side by side comparison of the 2500k and the i7-860

http://youtubedoubler.com/fqnh

You may want to use the pause buttons to resync it, because it is quite dynamic and the speeds tend to change from video to video.
 
This thread is a gold mine, thanks @psolord this is exactly the sort of info I'd been hunting around the web for, and it was right here.
 
Back
Top