Kingston HyperX 3000 Review

[ThugsRook]

Kingston HyperX3000
http://www.bleedinedge.com/images/thugs/kingston.jpg

i've had these pair of sticks in the TR lab for quite some time now. i personally always thought they ran very well in my systems but never realized just how fast this memory was until i started seeing reviews from of all the new fancy highspeed 500+ddr thats been coming out recently.

can 370ddr possibly perform as well as big money 500+ddr?
since im the resident budget overclocker around here, that's the question i'm going to try and answer today.

* please note that these are not test samples, nor are they even a matched set. these two sticks we're purchased seperately by yours truely from googlegear.com (http://www.googlegear.com/jsp/Home.jsp).



Specs from the Kingston website...

DESCRIPTION:
64M x 64-bit (512MB) DDR370 memory
sixteen 32M x 8-bit (8M x 8-bit x 4 Bank / 185MHz) in TSOP packages
gold contacts

FEATURES:
Power supply: Vdd: 2.5V ± 0.2V, Vddq 2.5V ± 0.2V
Programmable Read latency 2 (clock)
Programmable Burst length (2, 4, 8)
Programmable Burst type (sequential & interleave)
Timing Reference: 2-2-2-6-1 at +2.6V
Auto & Self refresh, 7.8us refresh interval (8K/64ms refresh)
PCB: double sided component

PERFORMANCE:
Clock Cycle Time (tCK) 5.4ns (min.) / 12ns (max.)
Row Cycle Time (tRC) 60ns (min.)
Refresh Row Cycle Time (tRFC) 72ns (min.)
Row Active Time (tRAS) 42ns (min.) / 120,000ns (max.)
Power (IDD0) 3.405 W (operating)
Operating Temperature 0o C to 70o C


>>>

[ThugsRook]

1:1 Bandwidth Testing
since no one benchmark can tell the whole story, im going to run 4 very different memory benchmarks to make sure we can see a well rounded picture of exactly how they perform from every angle.
starting at the memories rated speed, ill work my way up the fsb...



Test System
Kingston HyperX3000 2x 512mb
Pentium4 2.4B
Abit IC7 (canterwood)



Test Configuration
NBs800 | GAT Street Racer | cas2226-1 | 1:1 | 2.8v vdimm
was used for all of the following tests....



http://www.bleedinedge.com/images/thugs/ico-m86.gif Memtest86 v3.0
185fsb 370ddr = 2775 mb/s
190fsb 380ddr = 2847 mb/s
195fsb 390ddr = 2918 mb/s
200fsb 400ddr = 2998 mb/s



http://www.bleedinedge.com/images/thugs/ico-a32.gif Aida32 v3.70 read/write
185fsb 370ddr = 5118 / 2241 mb/s
190fsb 380ddr = 5254 / 2307 mb/s
195fsb 390ddr = 5386 / 2352 mb/s
200fsb 400ddr = 5545 / 2422 mb/s



http://www.bleedinedge.com/images/thugs/ico-pcm2k2.gif PCMark2002 memory
185fsb 370ddr = 9830
190fsb 380ddr = 10101
195fsb 390ddr = 10453
200fsb 400ddr = 10641



http://www.bleedinedge.com/images/thugs/ico-sss.gif Sandra v7973 unbuffered
185fsb 370ddr = 3277 / 3094 mb/s
190fsb 380ddr = 3362 / 3174 mb/s
195fsb 390ddr = 3448 / 3257 mb/s
200fsb 400ddr = 3545 / 3348 mb/s




400ddr with all the go fast goodies enabled!
but at this point i ran into a slight problem. no, it wasnt with the memory. its the cpu, it cant OC any further then 200fsb. (i know i know, poor me right?) in order for me to continue taking this memory higher ill need to enable the 4:5 memory ratio....

>>>

[ThugsRook]

4:5 Bandwidth Testing
since these sticks can obviously go further then 400ddr, i used my motherboards 4:5 memory ratio to boost the memory bus by 25%. this is an "async" ratio, which means i lose the full functionality of GAT. scores will be well below what i got while in 1:1 ratio, but this will show how far HyperX3000 can run... at least in async modes.



Test Configurations
NBs533 | GAT Auto (disabled) | 4:5 | 2.8v vdimm
was used in all of the following tests....

160fsb 400ddr = cas 2226-1 | RDA enabled | CPC enabled
165fsb 413ddr = cas 2226-1 | RDA enabled | CPC enabled
170fsb 425ddr = cas 2226-1 | RDA enabled | CPC enabled
175fsb 438ddr = cas 2226-1 | RDA enabled | CPC disabled
176fsb 440ddr = cas 2326-1 | RDA disabled | CPC disabled
177fsb 443ddr = cas 2448-1 | RDA disabled | CPC disabled



http://www.bleedinedge.com/images/thugs/ico-m86.gif Memtest86 v3.0
160fsb 400ddr = 2625 mb/s
165fsb 413ddr = 2703 mb/s
170fsb 425ddr = 2781 mb/s
175fsb 438ddr = 2869 mb/s
176fsb 440ddr = 2758 mb/s
177fsb 443ddr = 2491 mb/s



http://www.bleedinedge.com/images/thugs/ico-a32.gif Aida32 v3.70 read/write
160fsb 400ddr = 4565 / 1903 mb/s
165fsb 413ddr = 4695 / 1958 mb/s
170fsb 425ddr = 4837 / 2014 mb/s
175fsb 438ddr = 4999 / 2082 mb/s
176fsb 440ddr = 4777 / 2074 mb/s
177fsb 443ddr = 4762 / 2053 mb/s



http://www.bleedinedge.com/images/thugs/ico-pcm2k2.gif PCMark2002 memory
160fsb 400ddr = 8946
165fsb 413ddr = 9252
170fsb 425ddr = 9468
175fsb 438ddr = 9736
176fsb 440ddr = 9421
177fsb 443ddr = 9153



http://www.bleedinedge.com/images/thugs/ico-sss.gif Sandra v7973 unbuffered
160fsb 400ddr = 2881 / 2826 mb/s
165fsb 413ddr = 2997 / 2909 mb/s
170fsb 425ddr = 3068 / 2994 mb/s
175fsb 438ddr = 3164 / 3059 mb/s
176fsb 440ddr = 3140 / 2987 mb/s
177fsb 443ddr = 3111 / 2793 mb/s



while i was getting up to 438ddr with the 4:5 memory ratio, its actual performance isnt quite as fast as 1:1 was. past 438ddr these sticks fell off very quickly and so did their performance due to the loss of the remaining memory tweaks. not even cas 2.5-4-4-8 could help them, not that it mattered performance wise.

still, i was very impressed to see this ram get to 438ddr while maintaining a good performance curve. thats alot further then i was expecting.

speaking of GAT and CAS settings, lets run some tests and find out just how important that stuff really is...

>>>

[ThugsRook]

What is GAT worth?
this comparison will be very simple ~ fully enabled GAT vs disabled.



Test Configuration
195fsb 390ddr | NBs800 | cas2622 | 1:1 | 2.8v vdimm
nvidia geforce4 ti4600 @ 320x714mhz
was used for all of the following tests....



http://www.bleedinedge.com/images/thugs/ico-3dm2k1.gif 3DMark2001
streetracer = 16563
disabled = 16221



http://www.bleedinedge.com/images/thugs/ico-a32.gif Aida32 v3.70 read/write
streetracer = 5386 / 2352 mb/s
disabled = 4938 / 1725 mb/s



http://www.bleedinedge.com/images/thugs/ico-pcm2k2.gif PCMark2002 memory
streetracer = 10453
disabled = 9323



http://www.bleedinedge.com/images/thugs/ico-m86.gif Memtest86 v3.0
streetracer = 2918 mb/s
disabled = 2589 mb/s




the HyperX 3000 does a very good job of being able to run GAT well past its own rated speed and specs. obviously the loss of GAT imposses a rather large performance hit. its not something you would want to give up, givin a choice.

now lets see how the HyperX 3000's performance is effected by CAS settings...

>>>

[ThugsRook]

What are cas settings worth?
another simple comparison.
this time i disabled GAT and ran the motherboard with 3 different but typical CAS settings.
lets see how bad, bad can really get...



Test Configuration
195fsb 390ddr | NBs800 | GAT disabled | 1:1 | 2.8v vdimm
was used for all of the following tests....



http://www.bleedinedge.com/images/thugs/ico-3dm2k1.gif 3DMark2001
cas 2-2-2-6 = 16221
cas 2-3-3-7 = 15971
cas 2-4-4-8 = 15735



http://www.bleedinedge.com/images/thugs/ico-a32.gif Aida32 v3.70 read/write
cas 2-2-2-6 = 4938 / 1725 mb/s
cas 2-3-3-7 = 4879 / 1708 mb/s
cas 2-4-4-8 = 4851 / 1708 mb/s



http://www.bleedinedge.com/images/thugs/ico-pcm2k2.gif PCMark2002 memory
cas 2-2-2-6 = 9323
cas 2-3-3-7 = 9027
cas 2-4-4-8 = 9016



http://www.bleedinedge.com/images/thugs/ico-m86.gif Memtest86 v3.0
cas 2-2-2-6 = 2589 mb/s
cas 2-3-3-7 = 2406 mb/s
cas 2-4-4-8 = 2249 mb/s




again, slowing the CAS settings imposses another large performance hit. (actually large is an understatement at this point if you include the performance loss of no GAT too) losing a combined total of 823 3DMarks and 669 mb/s of raw bandwidth (memtest86) is not something you can ignore easily.

thankfully the HyperX3000 is more then capable of running aggressive cas settings even with GAT fully enabled and was able to take this CPU to its maximum possible speed of 200fsb quite easily.

so now that we know what the HyperX3000 is capable of, lets see how it compares to a Pentium4 "C" system running pc3700 and pc4000 speeds.

>>>

[ThugsRook]

PC3000 vs PC3700
at this point i was very curious as to how well this very aggressive HyperX3000 stands up to pc3700 and pc4000 with their more relaxed timings.
to provide the 466 and 500ddr results i asked forum member LastRide to complete this comparison. all results were using best possible settings and 2.8v vdimm.



Test Systems
Kingston HyperX3000 2x 512mb | Abit IC7 | 2.4B
185fsb 1:1 370ddr | cas 2.0-2-2-6 | nbs800 | GAT street racer
200fsb 1:1 400ddr | cas 2.0-2-2-6 | nbs800 | GAT street racer

OCZ Gold PC3700 2x 256mb | Abit IC7 | 2.4C
233fsb 1:1 466ddr | cas 2.0-3-3-5 | nbs800 | GAT disabled
250fsb 1:1 500ddr | cas 2.5-4-3-7 | nbs800 | GAT disabled



http://www.bleedinedge.com/images/thugs/ico-m86.gif Memtest86 v3.0
370ddr = 2775 mb/s
400ddr = 2998 mb/s

466ddr = 2955 mb/s
500ddr = 3012 mb/s



http://www.bleedinedge.com/images/thugs/ico-a32.gif Aida32 v3.70 read/write
370ddr = 5118 / 2241 mb/s
400ddr = 5545 / 2422 mb/s

466ddr = 5349 / 1992 mb/s
500ddr = 5798 / 2206 mb/s



http://www.bleedinedge.com/images/thugs/ico-pcm2k2.gif PCMark2002 memory
370ddr = 9830
400ddr = 10641

466ddr = 10279
500ddr = 11020



http://www.bleedinedge.com/images/thugs/ico-sss.gif Sandra v7973 unbuffered
370ddr = 3277 / 3094 mb/s
400ddr = 3545 / 3348 mb/s

466ddr = 3209 / 3227 mb/s
500ddr = 3437 / 3315 mb/s




You can draw your own conclusions as to whether PC3700 or PC4000 memory is right for you, but it's clear that Kingston HyperX3000 is a solid option for those who aren't ready to spend their money on a marginable upgrade. The ability to enable GAT when running KHX3000 is worth it's weight in gold, and the addition of the capability to run extremely tight timings is the icing on the cake.


:wave: -thugsrook

[Mikki]

:)

[eva2000]

nice review..

Thing is, even some HyperX users are saying the recent batches of Kingston HyperX range aren't doing as tight settings as the original HyperX .. so maybe it's harder these days to get the same HyperX memory to do what yours did ? Can all recent HyperX run GAT and tight timings as yours ?

[Hawk]

Excellent info . . . . . .

[Mikki]

Originally posted by eva2000
nice review..

Thing is, even some HyperX users are saying the recent batches of Kingston HyperX range aren't doing as tight settings as the original HyperX .. so maybe it's harder these days to get the same HyperX memory to do what yours did ? Can all recent HyperX run GAT and tight timings as yours ?
It would be interesting to see if other HyperX could perform this well, as well as seeing whether other memory based on the same chips would also...;)

Maybe we could talk Thugs into buying some of each? :rolleyes: :look:

[ThugsRook]

ive already had HyperX3500 and XMS3200LL, neither of those performed as well as the HyperX3000.

the only other memory ive ever tested that performs similar, is the original white labeled XMS3000.

:)

[Mikki]

Which you don't have anymore...;):D

[eva2000]

Originally posted by Mikki
Which you don't have anymore...;):D guess those are in your sig now ?

[Mikki]

:D

[toymaker]

Another 'sobering' review, Thugs. Thanks.:wave:

[roystarman]

Great review, I'm sticking with my 3200LL for now. I'm not going past 250FSB anytime soon and it looks like that the loose memory timings are a real performance hinderance unless you do run that fast. One really wonders if there is a difference in the chips at all. Some just get fancy gold packages others don't.

[Paragon]

Outstanding review Thugs:wave: I remember very well the review that you did on XMS3000 and the results being also impressive...Maybe some of us missed the boat when we jumped from 2700 straight to 3200 without giving 3000 a chance;) :cool:

[poMONKey]

this "doesnt perform as well as earlier batches" thing is MORE than annoying... why does it happen?

i happen to have the corsair xms batch with the holographic labels...:rotflmao: oops... missed again :stop:

*edit* oh yeah... sweet review :)

[jhites]

Interesting review Thugs.

My HyperX3500 from the same time period as your 3000 seems to run along the same lines as your 3000. Google bargains back then. ;)

[Silencer]

How did you disable GAT in a CW motherboard like the IC7?
Put it to auto disable GAT?

[ThugsRook]

thx guys! :-)


"Auto" is actually "Disabled" for GAT on an IC7.

:)

[kevvv]

great review....


what are the chips on khx3000? BH-5?

I have PC3500 and I think they are Winbond CH-5 in them, from what I hear they have problems with dividers and 875 boards.

[dbwillis]

Both my HyperX 3500 (512mb) sticks are BH5, I had the spreaders off the other day.
Unfortunately, they are running at 266mhz inmy Shuttle...:(

[RotorHead]

Well done Thugs, I like the final comparison. With the 3000 at DDR400 scores being faster then the 3700 at DDR466, it makes you stop and think before making another purchase. It’s good to see some of the some of the budget hardware is still on the Bleedin Edge.

Did you run any game benches to see how it effected overall performance?

RH

[ThugsRook]

you mean compared between my system @ 400ddr and Lastride's @ 466-500ddr?
no ~ we dont use the same vid card.


:wave:

[RotorHead]

Ya, that’s what I meant, my bad. Didn’t know you guys had different video cards. That would make it a little hard to do.

Do you think a measurable difference would be seen, all thing being equal that is?

RH

[ThugsRook]

all things being equal ~ one is just as fast as the other.
(the same can be said for a properly setup 5:4 ratio "C" chip too)

:wave:

[LastRide]

I ran the same tests with GAT turbo enabled and CPC disabled at 250FSB,DDR500 2.5/7/4/3.Had to put ras to cas at 4 to run turbo.Not much of an improvement other than Aida read.It actually hurt most of my scores.No difference in M86,same score :scratch:

M86-3012
Aida32-5998/2164
Sandra Unbuffered-3305/3364
PCMark2002 Memory-10920

[ThugsRook]

Lastride ~ id be interested in seeing a thread with your 5:4 vs 1:1 results ;)

[LastRide]

Right now I will try 5:4 and see what I can run turbo at.

[LastRide]

Complete waste of time..LOL.I couldn't even boot at 250 FSB 5:4 and 2.5/7/4/3 with turbo enabled and CPC disabled.Its obvious turbo doesn't like 5:4 with my OCZ Gold.Once I disable turbo all is fine,bah.

I'll be doing a 1:1 / 5:4 comparison later.

[toymaker]

Thugs,
On my P4C800 Dlx, I just finished testing a 512 MB of OCZ4000EL dual channel kit out of curiosity.

As you can see in my sig, the Corsair (Winbond BH-5)is back in. The Corsair responds to voltage without errors and is at 3.1 Vcore, 4:5/2,2,2,5/454.9 Mhz right now. Anything over 2.85 vcore on the OCZ would cause errors in Memtest 86 no matter what the settings were. Also, the tightest timings that I could get with the OCZ was 2.5, 3,3,7 @ 400 Mhz 2.85 vcore. The Corsair will do 2,2,2,6 @ 412 Mhz, 2.85 vcore error free.

The OCZ will clock much higher than the Corsair @ 540 Mhz/ 3,4,4,8 / 2.85 v. compared to the Corsair at 454.9 Mhz / 2,2,2,5/ 3.1v. But because of the tighter timings and higher voltage, the Corsair performs much better.

I have nothing but good things to say about OCZ because they're doing their best to meet customer demand. And for that alone I'm very impressed and not the least bit disappointed. I hope that OCZ will be the first to come out with some DDR that'll do tight timings at the DDR 550 - 600 level. But for now.........

The OCZ has been RMA'd.