• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

SuperPi puts me off Skylake

Database stuff.

I tend to update the rig every couple of years in order to get a 10-20% performance boost from the new technology. It looks as though this move to DDR4 will not see this kind of performance increase for some of the tasks this rig performs.
 
Jono8 said:
Just for gaming, yes a 6700k would probably be best but it is hardly a huge leap from Haswell so with Haswell E you get the best of both worlds for the same price. That is why most suggest getting a 5820k over a 6700k.
Click to expand...

mitx pretty much rules out x99.

All my desktop CPUs are retired to my living room mitx box after two or three years. It means I get very good value out of a good k series part.
 
Frank_Rizzo said:
Database stuff.

I tend to update the rig every couple of years in order to get a 10-20% performance boost from the new technology. It looks as though this move to DDR4 will not see this kind of performance increase for some of the tasks this rig performs.
Click to expand...

Surely for 'database stuff' you'd be using a workstation with a xeon, or at the least HEDT?
 
It's custom designed, which works best with the highest OCs available and the lowest latencies. For that reason multi core is no use. Neither is stock speeds.

I will give it a go with a 6700K and try a few different mem sticks.
 
Intel have been focussing more on the iGPU with their mainstream platform, afterall Intel's mainstream platform is nothing but glorified mobile processors packaged into desktop form.

Have you conducted any database benchmarks Frank? or are you just assuming that there is a correlation between SuperPI and database work? I was under the impression that SuperPI was very IPC/cache speed centric, for database work I would've thought multiple cores, special instructions, memory and hard drive performance will come into play more.

EDIT: Also like Dave2150 says about using a Xeon, I'd have also thought that anyone doing anything remotely relating to databases would use a system with ECC memory and overclocking would be frowned upon. With databases the integrity of the data should be your main concern, speed is very much secondary.
 
Last edited:
Frank_Rizzo said:
It's custom designed, which works best with the highest OCs available and the lowest latencies.
Click to expand...

Obviously you don't value data integrity much if you are risking an OC and have no ECC memory. Can't think what single threaded x86 database you are using, unless it is MS Access/Jet based, or a legacy version of something like MySQL that doesn't scale with cores.

If at all possible spend your time/money solving the software issue first, so that you are not tied to clock speed - there is not going to be any huge per core gains - gains going forward are going to be from more cores or newer instruction sets.
 
This conversation has happened before and it will happen again :D

Look. The OCs are fine. The RAM is fine. Why can't some people accept that there are certain users out that that require rigs like this?

---

mmj_uk Yes to benchmark testing. As I said it is mostly custom stuff so no Access or SQL on this rig,

One particular process has been performed since 1997. I always come here for advice on the best / fastest platform for (my rig). In previous years I have used

Pentium 400?
Celeron 3.6Ghz?
5600X2
PhenomII ?
2500K
4770K

A few years ago (I think when I switched to / from AMD) I asked users of this forum to run a custom written benchmark file. I believe I give out a gift voucher to the person that had the best results.

I have also (10+ years ago) gone into PC stores with a CD and fired it up. Amazingly the sales staff would let me do that. I doubt they would now, but then again :D
 
What database, and what particular purpose? (What size company?)

I'm not a DB direct guy, but I work/interact/design with a lot of different databases/record stores/flat files architectures etc (DB2, Mongo, CAS, Endeca etc etc), across many instances (and across many countries) with real time needs. Which is why I'm curious.
 
Last edited:
Phil2008 said:
I get 8.7secs with my [email protected]. I got 12secs with my [email protected] if I remember rightly
Click to expand...

My E6600 did 13 seconds at 3.8GHz - I've had sub 11 seconds with my old Q9550 at around 4.5GHz. For a long time the WR was held by an E6600 at around 7-8 seconds (running like 7GHz on LN2 or something stupid). IIRC my 4820K manages about 8.1 at 4.6GHz but its awhile since I ran it.
 
Scougar. It's xBase language / format. Various programs are run for processing sports information. There is no need for the bloat of large database frameworks. It's all run in a DOS box.

Rroff and Phil2008s examples. This is the point I am trying to make here (of which some do not see the point). As each new technology appears there is a distinct change in SuperPi performance. I seem to remember the AMDs hitting 30 then 20 seconds. When Sandybridge came along it was around 10 seconds, Haswell around 8 (I get 7.846 @4.6Ghz). Skylake has now reversed that drop.

I am reminded of a thread I started 5 years, where I asked Will 1 sec 1 Million Super Pi ever be achieved? :D
 
Frank_Rizzo said:
. But on one site I see this:
SuperPi 1M
i7 6700K 8.643
i7 4790K 8.159

SuperPi 1M
i7 6700K @4.6Ghz 8.003
i7 4770K @4.6Ghz 7.846
Click to expand...

Results vary - especially when it is between two different platforms e.g. some more results from:
http://www.overclockers.com/intel-skylake-i7-6700k-cpu-review/
http://www.bjorn3d.com/2015/08/intel-core-i7-6700k-review-skylake-falling/3/

SuperPi 1M
i7 6700K @ 4.8Ghz 7.587
i7 4790k @ 4.8Ghz 7.504

SuperPi 1M
i7 4770k @ Stock 8.472
i7 6700K @ 4Ghz 8.33
i7 4790k @ Stock 8.283
i7 6700K @ 4.8Ghz 7.675

So different results to yours, due to different RAM speed/timings, different motherboards (and even different bios versions), and even different windows installations - all are factors.






Frank_Rizzo said:
Scougar. It's xBase language / format. Various programs are run for processing sports information. There is no need for the bloat of large database frameworks. It's all run in a DOS box.
Click to expand...

xBase is a catch all term - what does it actually run under? Dbase? Do you have the source, as there are ways of upsizing it, or using a more modern runtime.


Frank_Rizzo said:
Rroff and Phil2008s examples. This is the point I am trying to make here (of which some do not see the point). As each new technology appears there is a distinct change in SuperPi performance. I seem to remember the AMDs hitting 30 then 20 seconds. When Sandybridge came along it was around 10 seconds, Haswell around 8 (I get 7.846 @4.6Ghz). Skylake has now reversed that drop.
Click to expand...

I don't think Skylake has reversed that drop at all, most results seem to be within ~1% either way, which is certainly within the margin of error - would be better to say that Skylake has no gains in single threaded performance.

If it was easy to improve single threaded performance via IPC gains, or brute force ramping of clock speed, then we would still be going in that direction. The fact that we haven't and that the entire computing industry has moved to multi-core or multi-threaded solutions, shows that there we are basically coming towards the end of single thread performance gains, as it becomes increasing harder to find increasing small gains.

If your app is still slow, and you are counting on hardware gains to make it faster, then unfortunately it is not going to happen. In order to exploit the gains in recent hardware (not just multi-cores but newer instruction sets) then you need to use more recent software - whether this is possible with your current app (or worth it - if it's just for fun), or whether it needs to be rewritten, Only you can decide.


Frank_Rizzo said:
I am reminded of a thread I started 5 years, where I asked Will 1 sec 1 Million Super Pi ever be achieved? :D
Click to expand...

Short of a major breakthroughs, e.g. different conductors that scale faster, quantum computing, "reverse-hyperthreading" style solutions etc, then I'm not sure it ever will.
 
I use Harbour. Remember this is custom stuff, so the fact that MariaDB 10.1 can do 1 million queries per second does not mean I should switch this rig to MariaDB (incidently the server version of the rig does use MariaDB).

It's not so much that the code runs slow that is the issue. Some of the tasks are long processing loops producing summaries, some of the tasks just take in realtime data and produce suggestions on the fly.

It's just that at every tock stage it is nice to receive a performance boost purely from the hardware. It has always been a quick and easy way to get an instant ~15% increase in performance.

As suggested in the 1 second SuperPi thread it seems for certain tasks technology was going to hit the sound barrier and would need a totally different platform to get past it. That sound barrier is Skylake. The new supersonic equivalent platform is not even on the horizon as Kaby Lake and Cannonlake are still using the same design so nothing will change quite yet. SuperPi world records will not be broken.
 
I imagine that pure single-thread performance is not a particularly high development priority at Intel these days. Increasingly the main use cases for a fast processor (productivity software, encoding/rendering, games, etc) are relying on multi-threading and specialised instructions rather than single-core performance. And a lot of software development now targets platforms which are inherently parallel (GPUs, phones/tablets, games consoles etc.)

Essentially, achieving single-threaded performance gains is difficult and the benefits are not what they used to be. I see that the last stable release of Harbour was in 2011 - that's ages in computing terms! Is there an update on the horizon which could suit today's CPUs better?
 
mmj_uk said:
Nightly builds, overclocks, non-ECC memory... it sounds like a recipe for disaster tbh. :p
Click to expand...

Indeed... what if the database gets subtly corrupted due to an OC or a bug in a nightly build, and by the time you realise, your last month's backups contain the corruption as well? Bad news!
 
You must log in or register to reply here.
Back
Top Bottom