PC3200(DDR400) vs overclocked PC2700(DDR333) performance?

[Goi]

Hi guys,
apparently a lot of recent reviews of the KT400 chipset and motherboards based on them have shown DDR400 to be slower than DDR333 in most benchmarks. However, aren't DDR400 simply "overclocked" DDR333 memory? I mean, there's no JEDEC standard for DDR400 yet is there? If that's the case, would an overclocked DDR333 memory, say running at 170MHz DDR, be slower than default DDR333 memory running at 166MHz DDR? This is assuming the other memory settings are kept the same.

Anyway, is there any reason why 400MHz DDR SDRAM would run slower than 333MHz DDR SDRAM? I suppose the memory settings are less aggressive on the 400MHz memory?

 

[Mikki]

a lot of recent reviews of the KT400 chipset and motherboards based on them have shown DDR400 to be slower than DDR333 in most benchmarks
If they're running them both at their rated speeds, I don't understand. Can you give us a link?

aren't DDR400 simply "overclocked" DDR333 memory
No, it's what the memory is rated, or has been tested, at. If anything, it's over-->rated. WE do the overclocking...

would an overclocked DDR333 memory, say running at 170MHz DDR, be slower than default DDR333 memory running at 166MHz DDR?
If you tested the same memory (say pc2700, or ddr333), at 170mhz, and then at 166mhz, the 170mhz test would yeild faster results. Is that what you're asking?

is there any reason why 400MHz DDR SDRAM would run slower than 333MHz DDR SDRAM?
I can't see any. I'd like to read what you're reading if you get a chance...HTH

 

[thefool909]

It seems that the DDR400 memory available right now cannot be timed (CAS Latency etc...) as fast as DDR333 thus the latency of the DDR333 is faster. The DDR400 would still have more bandwith, but which system is faster depends on what the test was stressing (bandwith vs. latency).

 

[Mikki]

My Corsair XMS3200 defaults are 2.5/3/3/6, but I can run it easily at 2/2/2/5, depending on what fsb speed I'm at.

 

[grdh20]

If the memory settigns and timings are set the same, then it is the MB, not the memory causing the performance differences. If memory runs at 200fsb at the same timings, it really doesn't matter what it is called. It will be the same as long as it is stable at that speed.

 

[Mikki]

Check out my Playing with memory Settings thread, it's got some interesting info about this...

 

[SupermanCK]

MIKKI is right...got search for all those memory timing thread in here...
corsair actually had 2 different pc3200 now...one rate at 2.5-3-3-6 or 7...and the other suppose to rate at 2-3-3-6 or 7...at 400ddr = 200MHZ

 

[Goi]

Well, here are a few of the reviews/previews I've read showing DDR400 to be slower than DDR333, and they make it seem as if its common knowledge, which is what baffles me. I guess the memory timings play a large part of it.

Anyway, I don't think there's any JEDEC standard for DDR400 yet, so even though the memory might be rated at 400MHz, it isn't really a "standard" per se, which is what I meant by the overclocking part.

MSI KT4 Ultra(KT400) Motherboard review at t-break.com

In that benchmark, specifically
Synthetic Benchmarks show DDR400 to be slightly slower than DDR333 in all cases.
Office Benchmarks show it to be exactly the same as DDR333 in most cases.
Gaming Benchmarks show it to be slightly slower than DDR333 in all cases.

ASUS A7V8X KT400 Review at t-break.com showed things to be a bit better

In that benchmark, specifically
Benchmarks shows DDR400 to be slightly faster than DDR333 in most cases.


ASUS A7V8X KT400 Review at OCworkbench

In that benchmark, specifically
WinBench99/CC Winstone 2000 shows DDR400 to be slightly slower than DDR333 in all cases.
Sandra memory benchmark shows DDR400 to be about the same as DDR333(faster in ALU, slower in FPU).
3DMark2001SE shows them to be nearly identical.
Q3A shows DDR400 to be noticably slower than DDR333.

 

[Mikki]

Thanks for the info (that musta taken a while ). My thoughts on this are this--->after reading thru the reviews, and since the performance was roughly the same on each site, I think that Via has put out a chipset with a crappy memory controller. Obviously, we've shown that (at least with the 845x chipset) the faster the memory bus on the same stick the higher the performance. They couldn't get XMS3000 to run at 2.5 CAS Latency (??? first indication something's amiss), but I don't think that would have that much effect when you're talking a mem bus speed difference between 166 (333) and 200 (400). I'm surprised that they didn't talk more about that in those reviews. Anyway, I'm gonna look more into this, but that's my opinion for now.

 

[Goi]

Yeah well, I guess either that, or the Athlon XPs just aren't able to take advantage of the extra bandwidth yet. Perhaps the memory timings also played a major part in the performance. I'm surprised Anand or Tom hasn't done an article on this yet.

 

[Mikki]

I'm surprised too.
I'm not sure it's the fault of the cpu, I think the spread might be bigger with a P4, but I think something's wacked out with the KT400 southbridge. Weird.

 

[mechBgon]

PC2700 didn't exactly blow the doors off of PC2100 on KT333 either. The non-synchronous timing between the AthlonXP's bus and the memory forces the motherboard to buffer data instead of having a nice clean 1-to-1 ratio. At least that's my understanding of the issue. Making the setup even more out-of-sync can't be helping.

 

[Mickey21]

I think the fault of the performance is due to the fact that the tests shown in the benchmarks above were using Corsair XMS3000 memory. I personally recently purchased a Corsair XMS3200C2 v2.1 module and the timings are 2/3/3/6 1T. I bet when they ran the 3000 at the higher 400 speed, they went to 2.5CAS and that caused it to be slower. It is no secret a 2700 at CAS2 will beat a 3200 at CAS2.5, but there are 3200 at CAS2. Given the same timings, it is of course logical that a 3200 would outperform a 2700. Think about it. If it doesnt, then you can attribute it to something else, like poor support or incompatibilities. That I know of, Corsair tests each memory chip at 400DDR now, and as the modules fail one speed setting, they reclass the memory down a step till they find the module's sweet spot. I have heard all over on forums, and review sites how people are saying dont bother with a 3200 and just get a 2700 and run it at 3200 instead. You cant make that assumption, because if Corsair had it pass at 3200, I am sure they would sell that module as such. Buying 2700 instead is just taking a risk you might not be able to get it to 400. And besides, the 400's are only 30 bucks more. Not exactly breaking the bank once you are already ready to spend 150 for a 2700. Also, take 400DDR tests with a grain of salt, JEDEC has no spec, and current boards with 400DDR support are just trickling out. We are trying to compare them to 333DDR which has a spec, and has been out for some time now.

 

[Big8oy]

hey i kinda had a question on the corsair ram in general. I was wondering, I was at Frys and i saw that there are two different 512 chips of PC3200 corsair ram. One had a part number that was CMX512 3200PT and the other was CMS512A 3200C2. I dont think these are official corsair part nums but i was wondering if anyone knew if these are actual part numbers and know if there was a difference? Also they are different prices so there must be some difference but i can't tell. note that the CMS one was more expensive. thanks

 

[Kowan]

Via and SiS Battle for Supremacy: 3-way P4 DDR Motherboard Roundup
"However, current DDR400 memory is essentially overclocked DDR333 relabeled as DDR400. DDR333 is now a JEDEC-approved specification whereas there is no spec for DDR400, hence the reason why there only exists overclocked DDR333. Generally speaking, with a JEDEC-approved specification comes lower memory latency and better timings than just overclocking a previous technology and relabeling it. Even though DDR333 has considerably less available theoretical bandwidth compared to DDR400, DDR333 is in fact competitive with DDR400 if you purchase the right kind. For example, CAS 2 DDR333 from Samsung will perform more or less equal to the best DDR400 currently available on the market (whose CAS Latency is still no better than 2.5).

If you remember, DDR333 was once in the same situation as DDR400 is in today. Back then, DDR333 was nothing more than just DDR266 overclocked to DDR333 speeds. Weeks later, real DDR333 started hitting the streets and with that finally came some noticeable performance gains over DDR266. If the DDR333 situation of several months ago has taught us anything, it?s to wait for the real thing. Therefore, we suggest that everyone wait for real DDR400 to hit the market.

Looking ahead, if JEDEC ever does finally approve a spec for DDR400 memory, that will likely be the last frequency we see out of DDR-I technology. DDR-I is clearly reaching its physical limits, and a new design needs to be implemented to reliably scale past DDR400 speeds. As you might already know, DDR-II is this new, more scalable design. The current timetable for DDR-II?s initial availability is mid-2003, but because this timetable is only for first shipments, DDR-II probably won?t have a significant impact until months later. However, we have reason to believe that as early this winter we will see DDR-II products in the form of high-end graphics cards. NVIDIA?s next generation NV30 GPU is rumored to include DDR-II technology, and we?re sure ATI won?t be snoozing at the wheel in implementing DDR-II for their recently introduced R300 architecture."