First "real" Nocona vs. Opteron review?

[Mloot]

First saw it posted at the 2CPU forums, but hadn't yet seen it here at AT, so here you go:

Link

 

[n0cmonkey]

I still like the opteron better. But I'm kind of surprised. Didn't think Intel's chip would be doing that well in some of those tests. The Apache test just kills it though.

 

[Mloot]

Yeah, I have to agree with you on the Opteron luvin'. It'll be nice to see what the 3.6 Noconas can do, though.

 

[Vee]

The Nocona is a Prescott Xeon. It can thus be expected to perform in a manner similar to the Prescott on 32-bit benchmarks. And it does.
These were only 32-bit benchmarks, albeit in a 64-bit OS. The Flash mpeg import is even single threaded.
The interesting question is if it will scale as poorly as current Xeons. Yes, indeed, it does. No improvement at all there.

The important and interesting benchmarks are a complete disaster for the Nocona in this review.
The result on the other 32-bit application benchmarks, are exactly as expected, or slightly worse. If there is any remarkable observation at all to do, it is that the dual Opteron beats the Nocona, on some of these highly Intel 32-bit optimized media encodings.
With the sole exception that these tests ran under WindowsXP64, this review doesn't differ from previous Xeon vs Opteron comparisions. Nor does it provide any unexpected results.

In light of this, I find the positive tone of the review, like in "The Final Word", very remarkable?!
And frankly, I can't see that Intel have done anything at all, besides supporting 64-bit which is great of course, to readdress the lack of competitive edge in the Xeons. The selection of benchmarks used in the review... - Who the hell purchases a 2-way 64-bit system to do single threaded 32-bit media encoding?

 

[bobbyk]

you are right that they didn't use the greatest benchmarks, but the site does have a point. The nocona was a major upgrade for the xeon and it barely manages to beat an opteron. Now what if you did 4-p benchmarks. Then xeon is dead hands down. Even with ht (no one really needs 8-cpu's until you get to much heavier loads.).

Now if opteron got a major upgare. for speculation, amd could do 250 * 4 hypertransport. unofficially support ddr-500 memory. They could update their chipset with new features and call it the amd x-killer 9000. I think the nforce 4 will be here by the time technologies that are new start to take off. You can bet that mb makers are scrambling to get samples to start board production. Amd also could press 90 nm and do a 2.6 and a 2.8 opteron. Intel wouldn't stand a chance. ( i wouldn't be surprised if the next major core revision has support for ddr-2. Amd said early 05 for that. It might be that time by the time the next one is out so it is possible.) I just wonder what xeon will do when amd gets even more bandwidth with low latencies. I mean if you have pc 5300 ram in a dual channel numa aware os that is a theoretical peak of of more than 20 GB/s. Opteron might be 18 of that. intel might get 10. Ohh wait, they don't do numa. Maybe they will be 7.

I do not think that amd will do a paper launch though. If they say the nforce 4 is there the boards will be shipped and ready that day.

 

[ncage]

I sure hope since intel has produced a 64 bit chip it doesn't hurt AMD's momentum. They have got a lot of mementum because they were the first out of the gate with a 64 bit chip that would still run 32 bit software but since intel now has 64 bit i hope they don't start to move to intel chips. Sometimes benchmarks don't matter. Me and you know AMD's better but intel makes people of companies get the warm & fuzzy feeling because intel has been a standard for so long. Look at what happened when the P4's first came. The athlon xp's where clearly superior yet most big oem's still went with intel chips. To make AMD big amd needs to move from the enthusiast community to the big oem's (thats where the sales are). I have heard rumors that dell is currently looking at the opteron. This would be a MAJOR acomplishment for amd and i hope intels 64 bit doesn't hurt amd's chances.

 

[Nemesis2038]

Thankyou. I started a thread or rant yesterday about the lack of benchmarks for a new cpu release even if they could only provide 32 bit ones.

As I understand it the next opteron core is getting the following.
DDR2 Support
SSE3 Support
Improved branch Prediction

I also thought AMD was extending the Pipeline to 16 from 12 in the next core release? I read this will increase the existing streaming applications by keeping the pipes full without sacrificing any if at all on the existing applications because of the improved branch prediction. It also compensates for the increased speed of the CPU.

I also heard that there are some changes to power management which will also improve performance. Something about power management keeping the processor cooler was causing some performance decreases and that has been changed.

Still it would appear the Opteron still significantly outpaces the Intel chip in the important benchmarks for me. Which translates to Coporate applications like the Opteron more than Intel Still.

 

[bobbyk]

you are correct about outpacing the intel in corporate benchmarks. Is it possible to do a core revision that lengthens the pipeline 4 stages? It would seem to me that it is one of the first thigns that the team decides because then everything else is designed to make it as effective as possible.

One question comes to mind though, doesn't it use 16 now, or is that just the athlon 64? I think one of them forsure if not both use a 16 stage pipeline already. But there is another benefit of a longer pipeline and that is HT. The longer it is the better ht works. Granted amd will see no real benefit from ht, but the marketing impact could be huge. It would be one less thing in intel's bag of tricks and would also allow them to perfect it before they actually do need it.

ddr-2 is posbbile, but amd did say they weren't doing that till 05 or ddr-2 667. the latter is only due to release just before the end of the year. By then it is possible. Who knows. I do not think that the a64 or opteron will benefit a lot from it until the latencies are lower, because it is proven that they have more impact than actual bandwidth for the opteron. I think maybe though they might segment the lines. Opteron might do ddr-2 while they work on the athlon 64. That would make sense because the server products are much more profitable than the desktop cpu's.

 

[bobbyk]

i meant to add on the pipeline part, that would sound like an entire redo of the core. Wouldn't they call it the k9 then?

 

[NewBlackDak]

I would like to have seen a large database test in that aswell. Was it just me, or did HT actually hurt the Xeon performance in some of the real tests.

 

[bobbyk]

you are right, that ht did hurt the xeon is some real world tests. I think this might be odd, but i do trust the site. They are fair to both sides. I don't think the results were skewed. I am not sure why it worked that way, but it did.

 

[Markfw]

It looks to me as if in server tests (apache,etc...) where these are normally used, the Opteron killed the nocona, but in the workstation tests, its was 50/50 as to who won.

 

[acejj26]

why didn't the reviewer use a NUMA-enabled OS like Windows 2003 Server? He claimed to use two sticks of memory for each processor to get 6.4 GB/s bandwidth from each processor, but he couldn't do that because WindowsXP Pro can't take advantage of the memory controller on the second processor. So, this "review" of 64 bit dual processor platforms uses an older OS (remember, XP is now 3 years old and not designed for these technologies) which is not NUMA-enabled, and he runs single-threaded 32 bit applications. Hmm....what I'd like to see is Win2k3 Server running at the very least or run some benchmarks on 64 bit Linux (which would also allow for 12.8 GB/s on the opteron platform).

 

[bobbyk]

ohh but intel wouldn't like that. Then their processor would still get what, 6.4 gb/s. In theory amd is at about 90-95% of the max of 12.8 so it should get >11 GB/s. With low latency ram, when would a xeon ever win again?

Oh well, intel will have to change.

I think the opteron is one of the best if not the best processor in the world right now.

The only other problem for amd is the possible power of a desktop p-m. With a nice fsb and a longer pipeline and therefore, higher clockspeeds; it could put up more of the fight then prescott or nocona.

 

[BDSM]

bobbyk.. the A64/opteron has a 12 stage pipeline.

I also wonder why they don't use a numa aware os, It would be VERY nice to see what kind of performance benefits it could give.
Ofcourse the extra bandwidth will help but probably not more than a few % in most apps, but maybe latencies will drop considerable if the second processor doesn't have to "ask" the first for ram access.

 

[Vette73]

Well look at it this way. This is the last P4/Xeon upgrade for the most part (st least planned) and he Xeon is still behind. Let alone the .090 Opteron/Athlon64 is getting some new improvments. SSE3, better prefetch, etc... so the Opteron will be getting a update soon also.

 

[glugglug]

Still waiting for 64-BIT Nocona benches...

 

[rck01]

Opteron is really good at doing a few things at once. Saturate the CPU, however, and it tanks.

RCK

 

[LTC8K6]

Saturate any CPU and it tanks.

 

[rck01]

I meant vs. Xeon. Under complex workloads, Xeon - especially the new Nocona-based model - stomps all over Opteron.

RCK

 

[mamisano]

Rck, what data are you basing your conclusions on??

I would also like to see a 4-way shootout, especially in a DB intensive test load.

 

[rck01]

We're working on a comprehensive, 5-way platform review comparing 3 Nocona (2 HP, 1 Dell), 1 Opteron (IBM) and 1 Prestonia (MPC) configurations. So far, the Opteron box is the laggard.

Note: You can read the review in an upcoming issue of InfoWorld (Test Center section). Should stir things up a bit...

RCK

 

[mamisano]

"Upcoming", as in when?

I am guessing these were all 2-way boxes running the latest OSes like Win2K3, Linux in 32 and 64bit, and Win-64?

 

[rck01]

All were 2-way running Windows XP (32-bit) SP1. Tests consisted of a mixture of client/server database, workflow, multimedia playback and media encoding (components of our Clarity Studio suite). Though its fast on linear tasks (OfficeBench is 20-25% faster), Opteron folds as we scale the number of concurrent workloads...

RCK

 

[clarkey01]

who cares OPTERON TILL I DIE