
Real Review of DDR Mobo 63
An anonymous reader sent us an overview of the AMD 760 chipset, and
benchmarks to give some real numbers to DDR RAM. (10-15% speed increase over comparable SDRAM systems)
If you want to put yourself on the map, publish your own map.
The hell with this... (Score:2)
NOW (maybe I'm stupid), but it seems that, not only are halfway decent (read: fully working/compatible) mobos not going to be available for some time, but in order to fully exploit the new chipset, I need a NEW Thunderbird with a 266mhz FSB? Someone please let me know if I'm being a complete moron, reading the article wrong, whatever, but I'm going to be extremely upset if AMD plans on releasing a new version (faster front-side bus) of a chip I JUST bought in the next couple months.
Sorry if I'm ranting...that's what I get for gut reaction posting, I guess.
Re:Good chipsets on bad boards (Score:2)
Re:Either this mb is POC or... (Score:1)
7. Memory bandwidth is a big bottleneck, but very fast caching technology built on chip masks the slowdown caused by the memory bus.
Seriously, I was astounded when I saw that a PIII/1000 benchmarks nearly twice as fast as a PIII/500.
See for yourself at: http://www.tomshardware.com/cpu/00q2/000511/pentiu miii-11.html
Re:Either this mb is POC or... (Score:1)
"Prepoduction motherboard" (Score:1)
#4 is correct (Score:2)
DDR does not have latency problem. It is basically the same as current SDR, but has double the bandwidth (hence Double Data Rate). So, computers with DDR will get a slight speed increase without incurring latency penalty. DDR machines will be faster than both SDR and Rambus, but still *nowhere near* twice as fast!
___
That's BS (Score:1)
Re:Wither PC100 SDRAM? (Score:1)
but SDR SDRAM and DDR SDRAM have different pinouts. so if you want to use your old SDRAM in the new DDR MB, you'll need to buy the MB that has both SDR and DDR slots... but there might not be a lot of those around.
exactly. (Score:1)
Only very few applications (like large databases) will get a significant speed boost from greater memory bandwidth. 99% of other applications (even games) will get only a marginal increase.
___
Dual, SMP anywhere? (Score:1)
Re:Wither PC100 SDRAM? (Score:1)
But the 100 MHz fsb doesn't matter. The athlon's memory bus is asynchronous to the cpu bus speed. That's why it's so easy to support either pc100 (cheap upgraders like me) or pc133 (getting a new system?) ram speeds.
The good old Intel BX chipset, on the other hand, uses the cpu's bus speed and some ratios to figure out how fast the agp, pci and ram clocks should go. That's a bit annoying for overclockers which cheap components, and is one of the reasons that the Athlon was well-received by the overclocking community.
Re:The hell with this... (Score:2)
Somebody correct me if I'm wrong here, but I think you might have your priorities a bit wrong here for a webserver. Firstly, you don't stuff the very latest and greatest technology into a webserver, as it's just as likely to not be stable. Secondly, if you're serving static content, just about anything will do. Thirdly, stuffing your machine with lots of RAM and a fast hard disk is probably more important than absolute memory bandwidth.
For your own machine, stuff it with the fastest CPU, RAM and graphics accelerator you can find to build a fast Quake machine, but for webserving the priorities are just a little different.
Re:Cache Hit Ratio (Score:1)
Re:Okay, I know I'm cruising for a TROLL moderatio (Score:1)
WinNT vs. Linux as a web/ftp/mail/samba server
SBLive! vs. Vortex2 as a soundcard
one of those matchups is a lot closer than the other.
chris
full disclosure: chris uses a sound blaster live! value under win98, win2k, and linux.
I'd rather buy a new CPU than change my RAM (Score:1)
Re:Good chipsets on bad boards (Score:1)
Look through the older stuff being sold on ebay; the pentium 200's that are going for more than modern machines are often actually worth the money.
The motherboard makers are screwed by the chipset makers; why shouldn't they throw marketing trash like built in audio on; they board's never going to be worthwhile 'cuz the chipset sucks and there's very little they can do about it.
A little bit of all of those... (Score:1)
Out of order execution & some other cpu tricks help as well... If you could give a P3 a mobo with an EV6 bus, you'd see a smaller increase than the athlon more from caching then anything else... athlons have very good cache structures...
There was an earlier review of DDR at Anandtech (Score:2)
And don't forget, these are preview boards, hopefully the real thing should give us an even bigger boost. Go DDR! :)
DDR? (Score:1)
--
Re:Huh? (Score:1)
Re:My only complaint about the board is.. (Score:2)
The OSS drivers in kernel 2.2.16 are poor - very poor and they're actually YMFPCI drivers - with modification. Translation: bad OSS drivers which supports only up to (almost) 22Khz sound 8 bit stereo..
But - grab ALSA and try their drivers - they are great! everything is supported (latest ALSA version added support for Bass/Treble)..
So ofcourse, if you can afford to buy to your PC at work a sound blaster Live - then go ahead and be my guest..
But these boards are for the main stream and for shops who sell white boxes. Without adding sound card - the seller can sell the maching cheaper..
If you really don't like the VIA sound chip, and SB Live is too much for your office - I strongly suggest to buy the Yamaha DS-XG based cards, specially that now they are fully supported under Linux (with ALSA drivers and OSS emulation)
Re:That's BS (Score:1)
DDR? (Score:2)
Is it just me, or does anyone else see DDR as "Dance Dance Revolution"?
Scott.Re:The hell with this... (Score:1)
Re:Whoa (Score:1)
Re:I really don't think this is a real test. (Score:1)
OTOH if this was supposed to be MB test both boards should have used SDRAM. I'm afraid that current test has no real information whatsoever because pre-production MB is slower - but DDR memory is faster.
I would have wanted to see at least test where AMD760 was running SDRAM to have real results about DDR being faster.
_________________________
Re:Yay. (Score:1)
1. I understood the original poster's intent by saying that SDRAM only gives a 10% system boost. The point I was trying to make was that a ten percent system boost by using DDR RAM (and a supporting motherboard), all else being the same, is quite respectable. What did he want, a system to run twice as fast by changing the RAM? It's still gonna run in the same general neighborhood of preformance (due to efficient caching was my second point).
2. You're right. I ambiguously claimed that you could get a 10% increase without buying a new processor. I still think it was clear enough. I didn't say you didn't have to buy a new mobo. And besides, buying a new mobo is much cheaper than buying a new processor. But I said nothing about cost-free.
3. I will be the last one to claim that a new PIII/1000 is twice as fast as a PIII/500. My point was that it is much closer to double the speed on Sysmark than one wold expect. If you adjust PIII/533 =126 to the expected PIII/500 = 118, then you get that a PIII/500 gets around 100 sysmark (rounding grossly) and a PIII/1000 gets around 200, rounding less grossly.
Your number crunching is much more precise than mine. In reflection it should really be more like 60% difference between 500 and 1000. You got me there.
At the top of the charts, you really do see it waiting for the RAM. But not as much as one (e.g. I) expected.
And I would think that DDR RAM would bring the stats more in line with double the SYSMARK speed.
the real benefit (Score:1)
Wow ! Tired ! (Score:1)
Re:why choose pc1600 ? (Score:2)
From the article...
"The Tyan Trinity A762 motherboard we used sported 128MB of Micron PC1600 DDR SDRAM in one of its four slots. This matches up on bandwidth with the 1600MBps of the 200MHz FSB of the Athlon Thunderbird 1.1GHz we used for testing. You will likely need to have a 266MHz FSB Athlon Thunderbird or Duron in order to use DDR2100 memory with the AMD 760 chipset."
--
Re:the real benefit (Score:2)
Yay. (Score:3)
Intel and AMD need to stop their Mhz/Ghz race and prod some chip maker into making decent, fast RAM. Otherwise, we're gonna be running 2Ghz machines bottlenecked to 133Mhz bus. And that will not be cool.
Re:Yay. (Score:5)
Unfortunately RAM doesn't work that way. People don't want to trade off latency for overall throughput. RAMBUS traded off latency for throughput. It has theoretically higher throughput than SDRAM but more latency, as a result in a certain class of performance measurements it does significantly worse than SDRAM.
A good preview... (Score:1)
Spooon!
Re:Yay. (Score:1)
DDR vs. SD (Score:2)
We already have some good examples to compare DDR RAM vs. the old stuff:
GeForce-based graphic cards.
They give us some idea of how much performance increase we can expect. 100% speed boost would be ridiculous. Expect something in the range of 20-30%. And that's still very good if you look at the comparatively small price tag.
Happy happy, joy joy!
Re:Wow ! Tired ! (Score:1)
<a href="http://nothing.nothing.com/why _oh_why_does_it_count_binary_as_caps/i_will_never
<a href="http://nothing.nothing.com/why _oh_why_does_it_count_binary_as_caps/i_will_never
--
Real PREview (Score:3)
----
Okay, I know I'm cruising for a TROLL moderation.. (Score:1)
(I know that the Diamond Monster MX300 caused a performance hit, but apparently that was caused by a poor implementation of the Vortex2 chipset by Diamond - apparently the later versions put out by Aureal actually offered a smaller performance hit than the Sound Bastard Dead! card did.)
Creative Labs is just about as bad as Microsoft in my book.
I really don't think this is a real test. (Score:2)
I don't know about you, but this doesn't seem to be even a remotely accurate test. And when you're talking about a 10-15% difference, you really do need accuracy.
So if anybody is planning on buying a DDR mb based on this test alone, they're making a big mistake, IMHO. I know I'm not (either planning or buying). I'll be sticking with SDRAM for a while...
Good chipsets on bad boards (Score:2)
I think it'll be about a year before we see some legitimate DDR motherboards; and not these turds that look like they got ganked out of a Gateway. I'm still waiting for something legit to replace my trusty 440BX. After all, that chipset has withstood the test of time, and even holds up to 133MHz. Mine's at 100MHz, so it will last much longer than these Athlons going up in flames left and right.
Re:Wither PC100 SDRAM? (Score:1)
Never ceases to amaze me.. (Score:1)
Re:Meaningless article (Score:1)
You Like Science?
Re:Yay. (Score:1)
Sysmark ratings use a small enough working set that the memory isn't overly taxed (L2 cache is a big help). For some things, raw integer or fp power overwhelms the memory speed, but in a lot of real world situations, that memory speed does hold you back. The more multi-tasking that you plan to do, larger working sets (for db apps, etc), and faster through-data that you work with (streaming, virus scans, etc) the more important the memory speed becomes... Most benchmarks are greatly aided by L2 cache (as well they should for proc benchmarks), and it would be nice to see some other numbers.
The 'stream' numbers would be interesting, to see if the memory/chipset can really deliver close to the rated throughput. Run the same tests with the L2 disabled (yes, yes, then it isn't a real world example, but it isn't anyway) for 'normal' and DDR SDRAM - it would be interesting, if not revealing. A nice kernel compile (which is assumed to be CPU-bound) might reveal some interesting things.
RAM speed may not make a difference to fast processors, but it makes a difference to a fast system...
--
Meaningless article (Score:4)
But these results make no sense. DDR has the same latency, higher bandwidth, but results in speed increases from -5% to +6%? It should be consistently faster, never slower
Re:Yay. (Score:1)
Lets look at this objectively: a 10% increase in speed over SDRAM, which is already way behind what modern processors need in terms of speed/bandwith.
Actually, it's a 100% increase in speed over SDRAM. It delivers data twice as fast. 100MHz clock speed * twice the data. Or another way, it yields 1600MB/s bandwidth.
OK, yes, in theory it is supposed to be 100% increase in speed. So, where is that increase?? I think that is his point, this great new memory isn't too efficient if you double its performance window but only see a 10% improvement when actually trying to use it. That definitely says to me the wrong problems are being addressed at the system level!
Looking at the leaps and bounds with which processor speed is growing, a 10% increase is a drop in the bucket.
Yes, it is a drop in the bucket. But you're comparing apples to commodores. It's a 10% system performance increase over systems that use PC100 RAM. That's quite respectable without changing the processor. Wouldn't you like to make your processor run 10% faster without buying a new one?
Whoa! Hang on a second. Yes, a 10% improvement without any cost would be nice...but you seem to have forgotten that in order to get this increase, you have to change your MB and your memory (or just build the new system with these parts). This isn't a costfree speed gain here. I don't know if you intended to imply that, but that was the way I read it. I will admit that if you are building a new system, going the DDR route is probably the wise choice in terms of cost/price, but if you already have a system, the cost could definitely be deemed prohibitive!
Also, I take issue with what you say about modern processors needing more speed/bandwidth in memory. I thought I agreed with you until I took a look at Tom's Hardware. A 533 PIII gets a 126 sysmark rating. A 1 Ghz PIII gets a 194 rating, nearly twice as fast when thechip is nearly twice as fast! If modern processors are really waiting for RAM so much, why is processor speed a linear progression up the performance chart? It would be tailing off, with performance gains of a 1GHz PIII only marginally faster than a 700 or 800Mhz at the top of the chart.
OK, here is where I really question some things. First off, 194 is only a 54% increase over 126, that is hardly "nearly double" in my book (of course, the author of my book may be different than yours). To get this 54% increase in this benchmark, we cranked the CPU speed up by 88%! So, if the CPU isn't waiting around...what is it doing with those extra cycles it is clicking through??? Again, we have a very poor return on our investment in speed. Any engineer should be cringing at these numbers, I know I am. As for why we have a "linear" progression, linear doesn't mean 1:1, so yes, the wait time is still there, but the CPU time is getting smaller and smaller and we see the overall score go up. Finally, let's look at the data from Tom's Hardware and see if I am right, that things are flattening out....Using the BX chipset at 133MHz (since it was on top of the charts):
CPU speed Score
1000 (+7.2%) 194 (+4.3%)
933 (+7.7%) 186 (+5.1%)
866 (+8.4%) 177 (+5.4%)
800 (+9.1%) 168 (+7.0%)
733 (+10%) 157 (+6.1%)
667 (+11.2%) 148 (+8.0%)
600 (+12.6%) 137 (+8.7%)
533 126
(My apologies to Tom for using his data without permission, I hope he does not mind too badly)
So, we see that we cut the performance gain by more than half, while cutting the CPU speed gain by less than half. Sure looks to me like we are losing the benefits of speeding up the CPU. Perhaps the original poster wasn't so far off base?
My apologies as well if this post seems a harshly worded, I was just a bit put off by statements that lack backing from the stats they claim back them!
Re:Okay, I know I'm cruising for a TROLL moderatio (Score:1)
Huh? (Score:1)
Re:Real PREview (Score:1)
Seriously, though, the x86 architecture (the bus structure most notably) isn't the best optimized for memory transfers, though the I/O is much worse (the whole x86 addressing scheme is due for a rewrite). </rant> Oh well, that's another topic for another day.
--
Re:DDR vs. SD (Score:1)
This is obvious if you look at benchmarks. Crank up the core clock on a Ge-force2 and you get little to no improvement. Crank up the memory clock and you get measureable improvement.
For CPU's dropping in a 1Gig compared to a 800 gives you a noticable improvement. So obviously CPU's aren't bandwidth bound nearly as badly as the Geforces are, so they won't see that high of a performance increase.
Re:Either this mb is POC or... (Score:1)
#4 is also pretty accurate, and can be further alleviated with better memory access management (think RISC) and pipelining (aka, P4). We are only RAM limted in certain situations, for much of what we do, a great deal is held in the L2 cache, so the accesses to main store are far less frequent.
--
Re:The hell with this... (Score:1)
It seems AMD has been planning to release a 266MHz FSB processor in 2000 for a while (Note the slide is from Comdex in the Fall of '99). Now, whether or not this will actually happen is up for discussion.
Oh, and by the way, this preview was on Anand's web page on the 14th of this month... how do things slip through the cracks so often here?
Cache Hit Ratio (Score:2)
Given a sequence of memory accesses (e.g. in a memory benchmark), and the sizes of L2 and L1 caches, it's straightforward to calculate which will require main-memory reads, and which will not. For instance, if your cache size is 512kb, and you're getting a memory location that isn't among the 512k most recently accessed addresses, then you're going to incur a RAM hit (with slight alterations for associative caches, but the idea is the same).
Big scan or copy operations, where each address is only accessed once, will reduce the cache hit ratio to zero (and fill up the cache with useless data).
I suspect that these benchmarks are getting a cache hit ratio in the 80-90% range, which would explain the lack of dramatic improvement. But since the testers don't state this number, it's difficult to see if DDR is giving the improvement expected.
The benchmarks that I've seen for DDR on video cards have usually been able to highlight where DDR gives a big improvement, eg Quake at 1600x1200, and where it doesn't, such as 640x480 with 16-bit color.
'bout time (Score:1)
Re:Wow ! Tired ! (Score:1)
--
Re:Yay. (Score:1)
For comparison, how much extra do people pay for a 10% increase in processor clock?
And when these DDR boards come out, AMD will release its 133/266MHz FSB processors, which should give performance another significant nudge upwards.
Re:Real PREview (Score:1)
My only complaint about the board is.. (Score:2)
Onboard audio? Does anyone actually use onboard audio in their high-end desktop? I HATE onboard audio and video. It'd be a different story if the motherboard manufacturers would support the friggin' thing after release, but most of the time they don't.
SB-Live Platinum, thank you. I'll keep it versus onboard audio.
-- Talonius
Either this mb is POC or... (Score:3)
So, I conclude that:
---
Every secretary using MSWord wastes enough resources
why choose pc1600 ? (Score:1)
I think this tests shows some bias against ddr by not featuring the fastest (and probably the most used in some months) type of ddr sdram.
Wither PC100 SDRAM? (Score:1)
Re:Yay. (Score:3)
Lets look at this objectively: a 10% increase in speed over SDRAM, which is already way behind what modern processors need in terms of speed/bandwith.
Actually, it's a 100% increase in speed over SDRAM. It delivers data twice as fast. 100MHz clock speed * twice the data. Or another way, it yields 1600MB/s bandwidth.
Looking at the leaps and bounds with which processor speed is growing, a 10% increase is a drop in the bucket.
Yes, it is a drop in the bucket. But you're comparing apples to commodores. It's a 10% system performance increase over systems that use PC100 RAM. That's quite respectable without changing the processor. Wouldn't you like to make your processor run 10% faster without buying a new one?
Intel and AMD need to stop their Mhz/Ghz race and prod some chip maker into making decent, fast RAM.
Considering that Intel is pushing RDRAM running at 800 MHz and that still runs worse than DDR RAM, don't you think that this statement is a little off target? I understand that your and my idea of decent RAM is not Rambus. But as far as numbers on paper goes, RDRAM is quite fast.
Also, I take issue with what you say about modern processors needing more speed/bandwidth in memory. I thought I agreed with you until I took a look at Tom's Hardware [tomshardware.com]. A 533 PIII gets a 126 sysmark rating. A 1 Ghz PIII gets a 194 rating, nearly twice as fast when thechip is nearly twice as fast! If modern processors are really waiting for RAM so much, why is processor speed a linear progression up the performance chart? It would be tailing off, with performance gains of a 1GHz PIII only marginally faster than a 700 or 800Mhz at the top of the chart.
It looks as if the fast cache on chip is the answer to slow RAM, and at least according to this chart, RAM speed does not make such a difference to fast processors.