Page 1 of 2 12 LastLast
Results 1 to 25 of 50

Thread: Madhsrimps artricle about tRD (performance level) on i7!

  1. #1
    Xtreme X.I.P.
    Join Date
    Nov 2002
    Location
    Shipai
    Posts
    31,147

    Madhsrimps artricle about tRD (performance level) on i7!

    http://www.madshrimps.be/?action=getarticle&articID=940

    interesting!
    the title of the article is a bit misleading imo, not exactly what the article is about. i was quite surprised when i read it, positively surprised...
    has anybody else played with this cas B2B timing? sounds a lot like tRD alright... hope we can get it unlocked soon! though i think both dfi and evga already play with it by adjusting the memory speed straps... though adjusting tRD directly would be a lot more efficient of course

  2. #2
    Xtreme Enthusiast
    Join Date
    Mar 2005
    Posts
    544
    Very interesting Article, many thanks Saaya.

    How were they able to determine the value of B2B on the ASUS board, I can't even see this timing in CPU Tweaker.
    Or maybe it's present in the BIOS, I think they started to show the actual values set in one of the latest BIOS versions.

    Is there any software which will show the value of this B2B setting ?

    PC: Intel Core i7 920 D0 | Kingston Hyperx T1 6GB 2000Mhz 8-8-8 RAM | Foxconn Bloodrage GTI | Sapphire 4870X2 | Perc 5/i - WD 640AAKSx4 RAID0 | Asus 2014L1T | Dell 2407WFP | Lian Li G70 Silver Watercooled
    WaterCooling: Q² Project by The-Fox
    CPU Loop: Swiftech Apogee GTZ | Feser X-Changer 480 | Swiftech MCP655 Vario w/ DetroitAC top | EK Multioption RES 250 Rev.2 | Tygon R3603 1/2" | YL D12SL@1350RPM x4
    GPU Loop: EK-FC4870 X2 Nickel | Swiftech MCR320 | Swiftech MCP655 Vario w/DetroitAC top | EK Multioption RES 150 Rev.2 | Tygon R3603 1/2" | YL D12SL@1350RPM x3



  3. #3
    Banned
    Join Date
    Nov 2005
    Location
    Acreageville, Alberta
    Posts
    1,411
    Very intereesting indeed Looks like I need to get my hands dirty on my R2E

  4. #4
    Xtreme Cruncher
    Join Date
    Oct 2006
    Location
    Boston, Massachusetts
    Posts
    2,224
    could be the answer to all the problems with memory the DFI UT X58 has... good find saaya!

  5. #5
    Xtreme Addict
    Join Date
    Jul 2007
    Location
    Alberta, Canada
    Posts
    1,264
    Very cool. One thing I like about the the Rampage Formula is that it allows tRD adjustment. It can make a noticeable difference for memory performance I'd be interested to find out how similar B2B is.
    Feedanator 7.0
    CASE:R5|PSU:850G2|CPU:i7 6850K|MB:x99 Ultra|RAM:8x4 2666|GPU:980TI|SSD:BPX256/Evo500|SOUND:2i4/HS8
    LCD:XB271HU|OS:Win10|INPUT:G900/K70 |HS/F:H115i

  6. #6
    I am Xtreme
    Join Date
    Dec 2008
    Location
    France
    Posts
    9,060
    Very interesting, thanks for the link!

  7. #7
    Registered User
    Join Date
    Sep 2008
    Location
    Portugal
    Posts
    41
    Amazing... What a difference.. Nice post ;P

  8. #8
    Xtreme Addict
    Join Date
    Aug 2006
    Location
    eu/hungary/budapest.tmp
    Posts
    1,591
    Dramatic effect on bandwidth, indeed.
    Great article!
    Usual suspects: i5-750 & H212+ | Biostar T5XE CFX-SLI | 4GB RAndoM | 4850 + AC S1 + 120@5V + modded stock for VRAM/VRM | Seasonic S12-600 | 7200.12 | P180 | U2311H & S2253BW | MX518
    mITX media & to-be-server machine: A330ION | Seasonic SFX | WD600BEVS boot & WD15EARS data
    Laptops: Lifebook T4215 tablet, Vaio TX3XP
    Bike: ZX6R

  9. #9
    Xtreme Guru
    Join Date
    Apr 2005
    Location
    Finland
    Posts
    4,308
    Would be nice to see more motherboards allowing to change this, but thinking of how this effectively would ruin DDR3-2000 kit sales which are significantly expensier than DDR3-1600 I don't think it'll happen...

    At least RAM manufacturers would be highly resistant about adding this setting to more motherboards.
    Last edited by RPGWiZaRD; 07-07-2009 at 08:06 AM.
    Intel? Core i5-4670K @ 4.3 GHz | ASRock Extreme6 Z87 | G.Skill Sniper 2x8GB @ DDR4-1866 CL9 | Gigabyte GTX 970 OC Windforce 3x | Super Flower Titanium 1000W | ViewSonic VX2268wm 120Hz LCD | Phanteks PH-TC14PE | Logitech MX-518 | Win 7 x64 Professional | Samsung 850 EVO & 840 Pro SSDs

    If all people would share opinions in an objective manner, the world would be a friendlier place

  10. #10
    I am Xtreme
    Join Date
    Jul 2007
    Location
    Austria
    Posts
    5,485
    Quote Originally Posted by Frank M View Post
    Dramatic effect on bandwidth, indeed.
    Thats the least concern on i7.

    Im more intrested in the scaling of bclk.

  11. #11
    I am Xtreme
    Join Date
    Jan 2005
    Posts
    4,714
    Almost no difference in BCLK, at least, not at first sight.
    Where courage, motivation and ignorance meet, a persistent idiot awakens.

  12. #12
    Xtreme Addict
    Join Date
    Aug 2006
    Location
    eu/hungary/budapest.tmp
    Posts
    1,591
    Quote Originally Posted by Hornet331 View Post
    Thats the least concern on i7.
    Oh yeah, really
    It might come in handy with dual-channel Lynnfield, though.
    Usual suspects: i5-750 & H212+ | Biostar T5XE CFX-SLI | 4GB RAndoM | 4850 + AC S1 + 120@5V + modded stock for VRAM/VRM | Seasonic S12-600 | 7200.12 | P180 | U2311H & S2253BW | MX518
    mITX media & to-be-server machine: A330ION | Seasonic SFX | WD600BEVS boot & WD15EARS data
    Laptops: Lifebook T4215 tablet, Vaio TX3XP
    Bike: ZX6R

  13. #13
    Xtreme X.I.P.
    Join Date
    Feb 2006
    Posts
    2,741
    RTL is more tRD than anything else on this platform, the RTL value is a result of the post Tx equation and sets the read back time from the DIMM's (the the IMC needs to be ready within that window for the transfer to take place). All downstream offsets between the Uncore an CPU are set according to RTL. CAS perforamance can be mirrored by adjusting the RTL value by upto 4ns past the sensed value, so long as a transistor pad is avaialble by the time of a readback request you'll get a post. In effect one can get CAS 8 performacne at CAS 7 if the value is moved out too far. Basicaly speaking keep RTL as tight as possible if you want the fastest performance at every point. The only thing I have not looked at much is the effect of higher Uncore frequencies upon the avaialable RTL window.
    Last edited by Raja@ASUS; 07-07-2009 at 08:36 AM.

  14. #14
    Banned
    Join Date
    Nov 2005
    Location
    Acreageville, Alberta
    Posts
    1,411
    Quote Originally Posted by raju View Post
    RTL is more tRD than anything else on this platform, the RTL value is a result of the post Tx equation and sets the read back time from the DIMM's (the the IMC needs to be ready within that window for the transfer to take place). All downstream offsets between the Uncore an CPU are set according to RTL. CAS perforamance can be mirrored by adjusting the RTL value by upto 4ns past the sensed value, so long as a transistor pad is avaialble by the time of a readback request you'll get a post. In effect one can get CAS 8 performacne at CAS 7 if the value is moved out too far. Basicaly speaking keep RTL as tight as possible if you want the fastest performance at every point. The only thing I have not looked at much is the effect of higher Uncore frequencies upon the avaialable RTL window.
    Great post raju.

  15. #15
    Xtreme X.I.P.
    Join Date
    Nov 2002
    Location
    Shipai
    Posts
    31,147
    Quote Originally Posted by RPGWiZaRD View Post
    Would be nice to see more motherboards allowing to change this, but thinking of how this effectively would ruin DDR3-2000 kit sales which are significantly expensier than DDR3-1600 I don't think it'll happen...

    At least RAM manufacturers would be highly resistant about adding this setting to more motherboards.
    not really, intel has obviously tweaked memory performance for 1333, thats what they focus on... since there is no official 1600 support they dont have a propper strap for that with really tweaked timings, so once bios engineers rip the intel memory code apart more and more, we might be able to tweak performance manually and get the same efficiency out of the imc as at 1333, which means with propper tweaking ddr3 2000 might actually make sense

    does anybody remember how higher fsb clocking actually made little or no sense initially on 965/p35 etc as trd and the fsb straps above 333 were really lax and performance barely scaled or got even worse? similar thing here... im not surprised theres barely a gain with more than ddr3 1333 atm cause thats the max intel tweaked the memory code for, its the highest propperly programmed/tweaked strap...

    its not like mem above 1333 will have huge boosts from this, 3 channels is really massive bandwidth even at low clocks already, and especially with apps only using 2, maybe 3 cores, there isnt a huge need for bandwidth... so far... but i do expect higher than 1333 to scale better if it would be tweaked propperly

    and raja, im not sure but there is no rtl setting on the asus boards is there?
    could it be that what they call b2b cas is a part of rtl or somehow adjusts rtl?
    Last edited by saaya; 07-07-2009 at 09:42 AM.

  16. #16
    I am Xtreme
    Join Date
    Sep 2006
    Posts
    10,374
    Quote Originally Posted by The-Fox View Post
    Very interesting Article, many thanks Saaya.

    How were they able to determine the value of B2B on the ASUS board, I can't even see this timing in CPU Tweaker.
    Or maybe it's present in the BIOS, I think they started to show the actual values set in one of the latest BIOS versions.

    Is there any software which will show the value of this B2B setting ?
    We compared the AUTO setting, which reads out as zero in the bios with manual settings. What we did was test it with Everest benchmark, superpi 1 and 32mb. Then did the manual settings on the Gene II and REX II mobo's... till we found the B2B setting that gave the +/- same results.

    It really has a dramatic influence on ram bandwith ( and not really on latency like with S775 ), luckily for us I7 users there's plenty of bandwith to mess with, but we want every Mb we paied for don't we ? When being around 12 ( for eg 2000mhz action ) and dropping to 10 gave nice boost, going down to 8, the boost was not as big and so on...

    I only tested the values that were 100% Hyperpi stable (8 threads) and will include them results in the upcoming Gene II review. This means that eg if we recommend a setting of 8 for daily usage (this to be bulletproof) that you can squeeze out some more performance for eg for Pi 1Mb benching purposes by dropping an increment or two.

    We passed Massmans and my findings to Asus...

    Quote Originally Posted by RPGWiZaRD View Post
    At least RAM manufacturers would be highly resistant about adding this setting to more motherboards.
    I think it will be better to include this bios settings as it will get more performance if adjusted properly (if set too loose by the mobo's bios) and if required ( when set too tight like eg with the Asus boards) to allow more stability.



    Some 32Mb and Everest screenies of the tests :

    C12 Test 1600Mhz C7



    C10 Test 1600Mhz C7



    C8 Test 1600Mhz C7

    Last edited by Leeghoofd; 07-07-2009 at 10:59 AM.
    Question : Why do some overclockers switch into d*ckmode when money is involved

    Remark : They call me Pro Asus Saaya yupp, I agree

  17. #17
    Xtreme Addict
    Join Date
    Feb 2007
    Posts
    1,674
    What a big difference!

  18. #18
    Xtreme Guru
    Join Date
    Apr 2005
    Location
    Finland
    Posts
    4,308
    My point was the higher frequency you go the higher you also have to have the B2B timing, handicapping the benefit of paying a load of extra money for very minor/no performance if at DDR3-2000 you'll have to use like 10-12 and at DDR3-1600 you can use like 6-8. That's what I'd think would hurt the DDR3-2000 sales unless there will be kits that handle such low B2B timings at higher frequencies too (but perhaps it's more of a mobo limitation? and then you're screwed eitherway).

    When you're not able to change this setting (it's set to say 12), the performance gain at higher frequencies, eg. going from DDR3-1600 to DDR3-2000 there's a bigger benefit for these RAM kits (altough only a fraction of DDR3 sales are those kinda kits anyway). But average joe won't even know of this setting in first place so it won't matter however it would turn out, but assuming every1 were aware of it then clearly more people would start looking into lower frequencies and tweaking tight timings and low B2B at slightly lower frequencies like say DDR3-1600, at least with todays offerings and price of the higher binned stuff remain unchanged.

    ^ That's what RAM manufacturers might not like about this. It might lead to having to lower prices on the higher binned kits and that's something every manufacturer likes to avoid if being able to or then they simply disappear if there's no reason to sell them if you can get roughly same performance at lower frequencies with tighter B2B setting.
    Last edited by RPGWiZaRD; 07-07-2009 at 12:43 PM.
    Intel? Core i5-4670K @ 4.3 GHz | ASRock Extreme6 Z87 | G.Skill Sniper 2x8GB @ DDR4-1866 CL9 | Gigabyte GTX 970 OC Windforce 3x | Super Flower Titanium 1000W | ViewSonic VX2268wm 120Hz LCD | Phanteks PH-TC14PE | Logitech MX-518 | Win 7 x64 Professional | Samsung 850 EVO & 840 Pro SSDs

    If all people would share opinions in an objective manner, the world would be a friendlier place

  19. #19
    I am Xtreme
    Join Date
    Sep 2006
    Posts
    10,374
    I see your point, good arguments there , but think that the high binned ram kits are also interesting to work with on other platforms, this is just purely I7 related (I5 is on the M testbench, will have to check with Massman how it goes there).

    Indeed with this setting you can regain some big ram bandwith (on certain mobo's with a too loose B2B setting) with a mere 1600mhz kit. Ram manufacturers might not like it, but we do :p So I still wish all manufacturers would add this feature to their biosses (under whatever name)
    Question : Why do some overclockers switch into d*ckmode when money is involved

    Remark : They call me Pro Asus Saaya yupp, I agree

  20. #20
    Xtreme Guru
    Join Date
    Apr 2005
    Location
    Finland
    Posts
    4,308
    Quote Originally Posted by Leeghoofd View Post
    Indeed with this setting you can regain some big ram bandwith (on certain mobo's with a too loose B2B setting) with a mere 1600mhz kit. Ram manufacturers might not like it, but we do :p So I still wish all manufacturers would add this feature to their biosses (under whatever name)
    Yes ofc as a tweaker that could spend several hrs just to find optimal RAM performance by changing one setting one-by-one, I personally would want this to be changable on all X58 (and P55) mobos ofc as I have my mind set on the DDR3-1600 kits anyway since those have by far the best price/performance/overclockability ratio IMO and this setting offers a little extra fun and seems very useful for these kits.

    It reminds me a lot of Performance level (PL) setting on DDR2 as I get roughly similiar results at DDR2-1000 4-4-4-8 PL7 as DDR2-1200 5-5-4-10 PL8. This setting wasn't changable on all mobos either but then most of the time MemSet would do it for you, I'm hoping some1 would manage to do this for B2B too.
    Last edited by RPGWiZaRD; 07-07-2009 at 03:37 PM.
    Intel? Core i5-4670K @ 4.3 GHz | ASRock Extreme6 Z87 | G.Skill Sniper 2x8GB @ DDR4-1866 CL9 | Gigabyte GTX 970 OC Windforce 3x | Super Flower Titanium 1000W | ViewSonic VX2268wm 120Hz LCD | Phanteks PH-TC14PE | Logitech MX-518 | Win 7 x64 Professional | Samsung 850 EVO & 840 Pro SSDs

    If all people would share opinions in an objective manner, the world would be a friendlier place

  21. #21
    Xtreme Legend
    Join Date
    Mar 2006
    Location
    France
    Posts
    599
    This timing is mentionned in the Intel datasheet at page 69.
    I made some test and it's possible to change it under Win; I'll add this feature in the 1.2 final version.
    WebSite: www.Tweakers.fr


  22. #22
    Xtreme Enthusiast
    Join Date
    Mar 2005
    Posts
    544
    Thanks for the info and support Felix, you are truly a Gem to our community, what would we do without you ?

    I am quoting from Intel's datasheet Felix has linked:
    B2B_CAS_DELAY.
    Controls the delay between CAS commands in DCLKS. The minimum spacing is
    4 DCLKS. Values below 3 have no effect. A value of 0 disables the logic. Setting
    the value between 3-31 also spaces the read data by 0-29 DCLKS. The value
    entered is one less than the spacing required, i.e. a spacing of 5 DCLKS
    between CAS commands (or 1 DCLK on the read data) requires a setting of 4.
    So, massman, do you want to confirm that 0-3 values has not effect ? :p
    Last edited by The-Fox; 07-08-2009 at 03:57 AM.

    PC: Intel Core i7 920 D0 | Kingston Hyperx T1 6GB 2000Mhz 8-8-8 RAM | Foxconn Bloodrage GTI | Sapphire 4870X2 | Perc 5/i - WD 640AAKSx4 RAID0 | Asus 2014L1T | Dell 2407WFP | Lian Li G70 Silver Watercooled
    WaterCooling: Q² Project by The-Fox
    CPU Loop: Swiftech Apogee GTZ | Feser X-Changer 480 | Swiftech MCP655 Vario w/ DetroitAC top | EK Multioption RES 250 Rev.2 | Tygon R3603 1/2" | YL D12SL@1350RPM x4
    GPU Loop: EK-FC4870 X2 Nickel | Swiftech MCR320 | Swiftech MCP655 Vario w/DetroitAC top | EK Multioption RES 150 Rev.2 | Tygon R3603 1/2" | YL D12SL@1350RPM x3



  23. #23
    xtreme energy
    Join Date
    Oct 2004
    Location
    Europe, Latvia
    Posts
    4,145
    I can confirm that on asus p6t dlx auto = 4 (bios 1606)
    ...

  24. #24
    I am Xtreme
    Join Date
    Jan 2005
    Posts
    4,714
    Quote Originally Posted by The-Fox View Post
    So, massman, do you want confirm that 0-3 values has not effect ? :p
    It makes sense since setting below 4 there's no gain in performance and, as Kiwi said, Asus defaults at 4 although the bios value indicates 0.
    Where courage, motivation and ignorance meet, a persistent idiot awakens.

  25. #25
    Xtreme Mentor
    Join Date
    May 2005
    Location
    Westlake Village, West Hills
    Posts
    3,046
    It's been here on the Gene II. Makes a huge difference in memory performance when you change values... 0 to 4 dont seem to make any difference, but as you increase the timing, bandwidth seems to suffer and scales with b2b
    PC Lab Qmicra V2 Case SFFi7 950 4.4GHz 200 x 22 1.36 volts
    Cooled by Swiftech GTZ - CPX-Pro - MCR420+MCR320+MCR220 | Completely Silent loads at 62c
    GTX 470 EVGA SuperClocked Plain stock
    12 Gigs OCZ Reaper DDR3 1600MHz) 8-8-8-24
    ASUS Rampage Gene II |Four OCZ Vertex 2 in RAID-0(60Gig x 4) | WD 2000Gig Storage


    Theater ::: Panasonic G20 50" Plasma | Onkyo SC5508 Processor | Emotiva XPA-5 and XPA-2 | CSi A6 Center| 2 x Polk RTi A9 Front Towers| 2 x Klipsch RW-12d
    Lian-LI HTPC | Panasonic Blu Ray 655k| APC AV J10BLK Conditioner |

Page 1 of 2 12 LastLast

Bookmarks

Bookmarks

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •