HyperTransport Link Testing

[BeepBeep2]

Hey guys, I'm not gonna post any screenshots but I just ran some quick tests in a real game called "TrackMania Nations"...it's got a built-in benchmark in the options.

1680x1050, settings maxed out, Forced Motion Blur, Forced HDR, Dynamic Shadows, 8xMSAA + 8xAF. Here's a resized screenshot of the benchmark: http://img710.imageshack.us/img710/9...onsforever.png

Phenom II 965BE C3 @ 4050 Mhz, NB 2.7 Ghz (18x225)
4GB DDR2-1200 5-5-5-15
Gigabyte GA-MA790X-UD4P
ASUS "TOP" ATI HD4850 @ 800/1125 (2250 effective GDDR3)
Two shortstroked WD 640GB Caviar Black's in a 320GB RAID 0 Array
Windows Vista Home Premium 32-bit

Total avg FPS:
52.0 FPS - 2025 Mhz HT Link (4050 Mhz effective)
52.1 FPS - 2250 Mhz HT Link (4500 Mhz effective)
52.3 FPS - 2475 Mhz HT Link (4950 Mhz effective)
54.1 FPS - 2700 Mhz HT Link (5400 Mhz effective)

I don't know whats up with the last result but I ran the same test 4 times, and re-ran the last one 3 times.

[NCspecV81]

BeepBeep2 - what is your NB frequency? It could have something to do with it being synced with the HT. So in essence, it may not be the case of a higher HT link but it being in sync.

[charged3800z24]

I am going to look at Min FPS on benches. That is what I care most about.

[BeepBeep2]

2.7 Ghz... I'm sort of wondering about this.

I saw (really) minor gains as I upped the HT Link but when the two were snyced I gained a whole frame.

[Mastiffman]

Nice Spec!!!

Numbers look good.... maybe I should Run this again and try and tighten up those timings...

[charged3800z24]

I did a little test with Crysis benchmark. Below you will see the detailed rsults. I gained 1FPS in Min and 3FPS on Average. Not exciting, but this is Crysis and it wasn't hard to up the HT link.

64bit
HD 5850 Single stock
965BE 3.4ghz / 2800NB
HT2000mhz vs 2600mhz

2000 HT

1/10/2010 4:25:35 PM - Vista 64
Beginning Run #1 on Map-island, Demo-benchmark_gpu
DX10 1400x960, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Demo Loops=3, Time Of Day= 9
Global Game Quality: Custom
Custom Quality Values:
VolumetricEffects=High
Texture=High
ObjectDetail=VeryHigh
Sound=High
Shadows=High
Water=High
Physics=High
Particles=High
Shading=High
PostProcessing=High
GameEffects=High
================================================== ============
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
Play Time: 30.51s, Average FPS: 65.55
Min FPS: 43.91 at frame 182, Max FPS: 95.98 at frame 1606
Average Tri/Sec: -67619304, Tri/Frame: -1031505
Recorded/Played Tris ratio: -0.89
!TimeDemo Run 1 Finished.
Play Time: 26.31s, Average FPS: 76.03
Min FPS: 43.91 at frame 182, Max FPS: 99.74 at frame 1760
Average Tri/Sec: -77447576, Tri/Frame: -1018687
Recorded/Played Tris ratio: -0.90
!TimeDemo Run 2 Finished.
Play Time: 26.26s, Average FPS: 76.17
Min FPS: 43.91 at frame 182, Max FPS: 102.71 at frame 131
Average Tri/Sec: -77495736, Tri/Frame: -1017396
Recorded/Played Tris ratio: -0.90
TimeDemo Play Ended, (3 Runs Performed)
================================================== ============

Completed All Tests

<><><><><><><><><><><><><>>--SUMMARY--<<><><><><><><><><><><><><>

1/10/2010 4:25:35 PM - Vista 64

Run #1- DX10 1400x960 AA=No AA, 32 bit test, Quality: Custom ~~ Overall Average FPS: 76.1

2600ht

1/10/2010 4:28:27 PM - Vista 64
Beginning Run #1 on Map-island, Demo-benchmark_gpu
DX10 1400x960, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Demo Loops=3, Time Of Day= 9
Global Game Quality: Custom
Custom Quality Values:
VolumetricEffects=High
Texture=High
ObjectDetail=VeryHigh
Sound=High
Shadows=High
Water=High
Physics=High
Particles=High
Shading=High
PostProcessing=High
GameEffects=High
================================================== ============
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
Play Time: 30.42s, Average FPS: 65.74
Min FPS: 44.99 at frame 183, Max FPS: 97.47 at frame 1621
Average Tri/Sec: -67774880, Tri/Frame: -1030965
Recorded/Played Tris ratio: -0.89
!TimeDemo Run 1 Finished.
Play Time: 25.98s, Average FPS: 76.99
Min FPS: 44.99 at frame 183, Max FPS: 104.86 at frame 1790
Average Tri/Sec: -80075168, Tri/Frame: -1040085
Recorded/Played Tris ratio: -0.88
!TimeDemo Run 2 Finished.
Play Time: 24.61s, Average FPS: 81.25
Min FPS: 44.99 at frame 183, Max FPS: 112.43 at frame 98
Average Tri/Sec: 82464824, Tri/Frame: 1014932
Recorded/Played Tris ratio: 0.90
TimeDemo Play Ended, (3 Runs Performed)
================================================== ============

Completed All Tests

<><><><><><><><><><><><><>>--SUMMARY--<<><><><><><><><><><><><><>

1/10/2010 4:28:27 PM - Vista 64

Run #1- DX10 1400x960 AA=No AA, 32 bit test, Quality: Custom ~~ Overall Average FPS: 79.12

[NCspecV81]

tried 2800 ht link? so it's 1:1 with the NB.

[charged3800z24]

I didn't want to reboot to get 2800 . AOD would only go to 2600HT.

[Daveburt714]

Man, I either has a sucky IMC or my water just isn't good enough...

Anything over 2650 on the IMC requires massive V's >1.35v and scaling above that is pretty much worthless.

I can push 2700 but it takes well over 1.4v, anything above that seems to be a lost cause....
There comes a point (at least in my book) where the voltage just doesn't justify the performance/heat involved...

Thanks for the numbers NCSpec !
I've always kept it matched or at least close to the IMC...
At least now I'll think twice before I blame the HTLink for OC problems.

[afireinside]

Only 3dmark I have installed/downloaded is 06 so I tried that. Got ~60 points out of 524mhz more HT link. Wasn't really expecting anything considering I'm only on a single 4850 though. Now to see if this HT link is stable. I recall having issues with higher HT link in the past but I may be wrong.

[NCspecV81]

I'd like to see some more results from you guys. Make sure it's not some anomaly with my setup.

[charged3800z24]

Man, I either has a sucky IMC or my water just isn't good enough...

Anything over 2650 on the IMC requires massive V's >1.35v and scaling above that is pretty much worthless.

I can push 2700 but it takes well over 1.4v, anything above that seems to be a lost cause....
There comes a point (at least in my book) where the voltage just doesn't justify the performance/heat involved...

Thanks for the numbers NCSpec !
I've always kept it matched or at least close to the IMC...
At least now I'll think twice before I blame the HTLink for OC problems.

I remember you saying you seen nice things keeping them in sync (NB/HT). I ran test a whil back on one of my Phenoms, I don't think I posted them. The gains were about what I saw to day. I think I migh do some experimenting this week with some other games. I think there might be a spot for HT with different CPU clocks.
On the NB, I had 1.35v on that for 2800, I only ran 4 instances of Memtest for an hour to test it. Not sure if it would pass Prime95. 3000mhz needs around 1.5+ just to do a bench.

[afireinside]

3Dmark 2003
single HD4850 @ 675/1050
PCIE 110mhz

GT3 seems to steadily increase with HT link on my setup. GT2 didn't skyrocket like yours nor did GT4 decrease. I'll try again in a few days when my 5850 arrives but I think the trifire you're running is part of the reason your gains are so massive. Still, there is performance to be had (both increased and decreased it seems...) with higher HT link







Tested near stock, linked, and one multiplier below linked to see if I had odd results like beepbeep. That didn't happen though. I don't think I can take my NB any higher

[Raiderman]

I didn't want to reboot to get 2800 . AOD would only go to 2600HT.

In order to get 1:1 with the northbridge you have to set the NB multi to 12, and use the bus speed to reach 2700+ on the NB clock. HT multi is locked at 12.
I am sure you already knew this, just thought I would post for other peoples benefit.

[chew*]

It's board dependant.

Not all boards gain from HT, also we have seen what appears to be a higher max fps but a lower min.

[SocketMan]

http://www.cpuid.com/perfmonitor.php


There is an option to monitor HT bandwidth.It doesn't get saturated with a single cpu,there will be a speed up with 3-4
cards plugged in.Thanks for testing,nice work

[Raiderman]

I mean 13x multi, my math was off

[BeepBeep2]

Man, I either has a sucky IMC or my water just isn't good enough...

Anything over 2650 on the IMC requires massive V's >1.35v and scaling above that is pretty much worthless.

I can push 2700 but it takes well over 1.4v, anything above that seems to be a lost cause....
There comes a point (at least in my book) where the voltage just doesn't justify the performance/heat involved...

Thanks for the numbers NCSpec !
I've always kept it matched or at least close to the IMC...
At least now I'll think twice before I blame the HTLink for OC problems.

I'm gonna guess sucky IMC...my air cooling isn't strong enough to run 4 Ghz 24/7, yet I can bench 3DMark06 @ 4.1 Ghz with a 2.8 Ghz NB.

[[XC] Oj101]

Now, would you like to try this with a 2D bench?

[gOtVoltage]

Also guys/gals..when your testing what works and dosnt for HT..

Make sure the PCI-E is Locked @ 100

Simply keep it it locked not on AUTO(set it manualy)... Some people use 100 or 101 and even up to 110...I always use 100(stability)


This will affect the Bandwidth of the Videocard/s if you do not keep this value the same for every run you do while changing HT.

Its something thats overlooked when comparing HT freqencies besides using the Same Ram and CPU and NB settings.

[chew*]

Not to mention I have seen upwards of a 500 point diff between bench runs at same settings in the same session just by a simple reboot.

200 points is well within bench variance. It's only the diff between a "bad" and "good" run.

[Don_Dan]

The first run after a reboot is always better than those run afterwards, keep this in mind when using 03 as a proof for your theory.

[ahmad]

Can you please rerun your game benchmarks at the lowest settings possible?

no aa, low low low, 1024x768

This will be a sign that it is an improvement for the CPU-GPU-memory transfers if it improves significantly.

[NCspecV81]

The first run after a reboot is always better than those run afterwards, keep this in mind when using 03 as a proof for your theory.

I ran them half a dozen times on the same boot for each. I'd say that factors that out. I'm also aware of this phenomenon.

[FireDragon]

TBH i have just kept my HT in sync with my NB speed through my overclocking of this chip and ram, i did try dropping the HT 2 see if it made a stability difference and there was none to b seen (at least for me) so for at least 95% of the time my NB is in sync with HT speed or vice versa depending on how ever you want to look at it

Dragon