Jump to content
HWBOT Community Forums

The day PSC took over from Hyper (Flare beats GTX2) ...


Massman

Recommended Posts

Hey all,

 

As I was running some queries through the HWBOT database regarding the overclocking capabilities of the LGA1155 platform, I decided to re-run the query used for the HWBOT Memory Index v0.1 article posted a few months ago. To my big surprise, the GTX2 memory was no longer the best performing kit after overclocking, but was taken down by the PSC-based Flare kit.

 

The methodology used for the charts are the same as described in the original article (page 2). The variables a, b and c were estimated using the performance results published by Leegfhoofd at Madshrimps (Intel Sandy Bridge CPU In-Depth Look at Overclocking, Memory Timings and More), where:

 

- a = 25,49

- b = 4,69

- c = 1

 

SELECT manufacturer.name, memoryproductgroup.label, memClock, result.memTCas, result.memTRCD FROM result JOIN memoryproductgroup ON (memoryproductgroup.memproductgroup_id = result.mem_productgroup_id) JOIN cpu_model USING (cpu_id) JOIN manufacturer ON (memoryproductgroup.manufacturer_id = manufacturer.manufacturer_id) WHERE (result.mem_productgroup_id!=0 AND result.status_id<10 AND memClock>600 AND memClock<1350 AND memTcas>5 AND memTrcd>4 AND application_id!=11 AND application_id!=13 AND application_id!=18 AND application_id!=21);

 

I also applied the same limitations to the query as described in the original article. There are two charts: one based on the LGA1155 overclocking results and one based on all results.

 

mem_sb.png

 

long.png

 

As these results surprised me, I re-evaluated the LGA1366 chart posted in the original article. In other words, I applied the LGA1366 platform variables to the current dataset and this is what happened:

 

lga1366-2.png

 

Again, Flare is beating GTX2. It's a close call ... but still.

Edited by Massman
Link to comment
Share on other sites

I.... have no idea how you got these graphs.

 

s1155.... isn't Hyper really badly supported at BIOS level?

 

It's also out of production.

 

 

Do Flares run 6-7-6 on X58 or something? Are the RAM BIOS/SPD tweaked for better efficiency than Dom-GT, GTX etc?

Link to comment
Share on other sites

I.... have no idea how you got these graphs.

 

Might be good to read through this: http://hwbot.org/article/news/hwbot_memory_index_v0_1_page_2. It explains the methodology.

 

s1155.... isn't Hyper really badly supported at BIOS level?

 

That's an explanation for the first chart (oc perf based on LGA1155 results only), but not for the second chart (where oc capabilities are estimated on all hwbot results, but performance scaling based on lga1155 platform variables).

 

Do Flares run 6-7-6 on X58 or something? Are the RAM BIOS/SPD tweaked for better efficiency than Dom-GT, GTX etc?

 

Ehr ... what?

Link to comment
Share on other sites

Well..... when you're running 1050MHz+ on X58 @ 7-7-7, you're in the territory of IMC limits. Could the conclusion be swayed by a Golden combination of Flare RAM on the best ICs... or are the sticks tuned better? Can Flares run tighter timings?

 

The RAM SPD/BIOS DOES have an effect on performance efficiency. You can test and confirm that easily yourself.

Link to comment
Share on other sites

Might be good to read through this: http://hwbot.org/article/news/hwbot_memory_index_v0_1_page_2. It explains the methodology.

 

Does this mean you're comparing highend kits without taking the best results into consideration? (no >2700MHz scores allowed)

 

If you want to compare top end products you need to compare top end results, too. GTX2's aren't for average joe overclockers:D

Link to comment
Share on other sites

Most of the 1350+ data is skewed because people enter the DDR rating, not the actual frequency. Also, I don't think anyone is running >1350 MHz besides for the memory validation shots, which is data I don't want in these charts.

 

The motivation and methodology is all in the article :D

Link to comment
Share on other sites

  • Crew
nothing beats GTX2 :D

 

2200Mhz CL6-9-6 is hard to beat on air cooled dimms... with respectable voltages... :D Plus My GTX2 kit is not a screamer... think we did 2180CL7-7-7-21 on them (not on a P67A-UD7Dinolicious-OC ofc)...further mhz made pi32M so unstable...

 

Must say this Gskill Flare kit roxors on P67...

  • Like 1
Link to comment
Share on other sites

Sorry to burst your bubble, guys. Here's the chart where overclocking capabilities of each kit is based on all SuperPI-32M results (where memory finetuning required). In a last attempt to give you your 'GTX2 is tha best'-chart, I'll re-evaluate the SB platform variables based on SuperPI-32M scaling.

 

superpioc1155.png

Link to comment
Share on other sites

Define 'best' result.

 

Please understand the main principle of these charts: "The quality of a DRAM product is determined by how people configure the memory in a safe-clock situation. The easier it is to reach frequency X, the more likely people will use it as safe-clock. The more difficult it is to reach frequency X, the less likely people will use it as safe-clock". They are key for understanding.

Link to comment
Share on other sites

2200Mhz CL6-9-6.........GTX2: 2180 CL7-7-7

 

Is tRCD = 7 not enough to compensate for one cycle tighter on CAS? tRP... who cares, it just looks nice in CPU-Z

 

The whole point of Hypers was tight rRCD for the IMC wasn't it?

 

 

(I'm not a GTX2 user, I'm using Dom-GT R2.1)

Link to comment
Share on other sites

Okay, this has been an interesting afternoon.

 

Turns out, the original formula had a logical problem: when going from absolute to relative performance by comparing the obtained performance estimate to the estimated performance for a DDR3-2000 CL6-6 configuration, the weight variable for the Frequency was no longer having any impact. This means: when using the formula to calculate relative performance (score(memory kit) / score(ddr3-2000CL6-6), changing the weight of the frequency had no effect on the relative performance.

 

So, to overcome this problem, Elmor developped a new formula:

 

Score = [(a * Fk / Fr) + (b * TCLr / TCLk) + (c * TRCDr / TRCDk)] / (a+b+c) where,

 

a = weight of frequency

b = weight of tCL timing

c = weight of tRCD timing

 

Fk, TCLk and TRCDk = average frequency, tcl and trcd of a specific memory kit

Fr, TCLr and TRCDr = reference frequency, tcl and trcd (in this case: 1000, 6 and 6)

 

As most of you are probably interested in the SuperPI-32M performance of each kit, I've also re-evaluated the platform-benchmark scaling parameters a, b and c. For SuperPI-32M on Sandy Bridge, the relative performance scaling parameters are:

 

a = 3,28

b = 2,09

c = 1

 

To make a long story short ...

 

I extracted all the SuperPI-32M memory overclocking results from the database, which would be the most objective data on the overclocking capabilities of every memory kit (as benchmark requires good memory subsystem). Using Elmor's formula and the new SB-SPI32M parameters, we can now give a rough estimation of the SuperPI-32M performance when overclocking a specific memory kit. Here's the chart.

 

ff1.png

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...