Bit-tech AMD six core review

[Zucker2k]

If you'll notice, something is wrong with their setup. Look at the LightWave benchmark in particular. Supposedly the program is utilizing all 12 cores in the system, yet the performance is one second slower? Definitely not right.

Also, this CPU was never meant to win out in sheer performance. If it was, there would have been no reason to try to keep it within the same power envelope of an equivalently clocked quad-core setup.

I'm sure there could be some improvement in that showing, but I don't think it's hardware, it's mostly in the software. For real world day to day apps, I won't expect much of a different picture.

And let's be honest, we've heard so much about AMD 6-core chips and AM3 upgrade path in the same sentence. As has been pointed out, more cores may not be the best strategy for AMD at the moment since it's quite evident that thread count vs. core frequency is what is going to matter.

[Hornet331]

@not utilizing all cores

well wasn't it already shown that some multithreaded apps have problems with core numbers that are not by the power of two. (with the X3 and now with hexa cores)

Gulftown will suffer the same fate. (though a bit eased with the help of HT)

[xlink]

Since bit-tech got sold to Custom PC things aren't good anymore.

The review has alot of mistakes like that.

Specially if you even look at the amounts of ram used between systems :/

you can even limit the amount of RAM accessible to the system through OS tweaks so there really is no excuse

[saint-francis]

Looks like a dud. It's really too bad. Even in the shoddy encoding test the Nehalem still mopped the floor with this 6 core disappointment. And I believe that handbrake uses x264 which is well multi-threaded. And this CPU still blew.

[demonkevy666]

guys , these boards don't have HT3 yet . Don't jump to conclusions.

would be nice even 3dmark06 would give us a better idea sometimes

Ehhh, how many threads are applications in the test using?

Four threads or less, no difference between quads and hexes. More than four and the instanbull starts to shine. if there are heavy load on each thread

What they should do is to run all tests at once

Heavy loads is where this K10 design seems to shine just swamp the thing they're incredible stuff at like that

I would use UT3 instead of Crysis.

Cyrsis was proven to favor high clocks and dual cores. +1 on the UT3

AMD's official info on the 2435: http://products.amd.com/en-us/OpteronCPUResult.aspx (Yes it is 75W for the Hexa-Core)

But that wasn't the only error in the article: WHAT ON EARTH are they thinking comparing a $1000 processor with a $1,600 processor.

The AMD 2435 @ 2.6Ghz and $989 competes with the Xeon X5550 @ 2.66Ghz and $959, not with the W5580 @ 3.2Ghz and $1,600.

Perkam

3.2ghz vs 2.6ghz

If you'll notice, something is wrong with their setup. Look at the LightWave benchmark in particular. Supposedly the program is utilizing all 12 cores in the system, yet the performance is one second slower? Definitely not right.

Also, this CPU was never meant to win out in sheer performance. If it was, there would have been no reason to try to keep it within the same power envelope of an equivalently clocked quad-core setup.

it's 600mhz difference ? too

@not utilizing all cores

well wasn't it already shown that some multithreaded apps have problems with core numbers that are not by the power of two. (with the X3 and now with hexa cores)

Gulftown will suffer the same fate. (though a bit eased with the help of HT)

anything in the pipe line that will yet ???

you can even limit the amount of RAM accessible to the system through OS tweaks so there really is no excuse

that's interesting if they're playing by that book you should be going 1-2-4-16-64.


Ht3.0 is really needed now the cinebenchR10 didn't gain much for the other cpu, HT2.0 bottleneck maybe.

a 4.0ghz phenom II gets about 15,000.

[Hans de Vries]

Doesn't show Istanbul's real potential.

1) Not using the bandwidth
2) Not using the cores
3) Not using the power budget

800 MHz DDR2 is simply not enough for many throughput applications.This is
currently a (temporary) Server platform limitation. Frequency scaling is just
38% (out of 100%) for SpecFP_rate. Just look at the six core Dunnington
to see what a lack of bandwidth does:

4 Socket: SpecFP_Rate: 4 core Nehalem 3.33 GHz: Xeon - X5570: 372
4 Socket: SpecFP_Rate: 6 core Istanbul 2.8GHz: Opteron 2439SE: 283
4 Socket: SpecFP_Rate: 6 core Dunnington 2.67GHz: Xeon X7460: 156

4 Socket: SpecInt_Rate: 4 core Nehalem 3.33 GHz: Xeon - X5570: 499
4 Socket: SpecInt_Rate: 6 core Istanbul 2.8GHz: Opteron 2439SE: 416
4 Socket: SpecInt_Rate: 6 core Dunnington 2.67GHz: Xeon X7460: 294

SPEC_2006

Dunnington has a 20% higher Floating Point peak performance as
Nehalem (63.85 vs 53.28 GFlops). Dunnington has 9 times the L2
cache amount as Nehalem (9MB vs 1MB) and Dunnington has twice
the amount of L3 cache (16MB vs 8MB).

But Nehalem runs circles around Dunnington in Spec_FP_rate because
Nehalem has a 3.75 times higher bandwidth as Dunnington (32.0 GB/s
vs 8.53 GB/s)

Istanbul with DDR2 is simply not that much better off with 12.8 GB/s.
(DDR3 should be better with 21.3 GB/s)



Regards, Hans

[perkam]

Doesn't show Istanbul's real potential.

1) Not using the bandwidth
2) Not using the cores
3) Not using the power budget

800 MHz DDR2 is simply not enough for many throughput applications.This is
currently a (temporary) Server platform limitation. Frequency scaling is just
38% (out of 100%) for SpecFP_rate. Just look at the six core Dunnington
to see what a lack of bandwidth does:

4 Socket: SpecFP_Rate: 4 core Nehalem 3.33 GHz: Xeon - X5570: 372
4 Socket: SpecFP_Rate: 6 core Istanbul 2.8GHz: Opteron 2439SE: 283
4 Socket: SpecFP_Rate: 6 core Dunnington 2.67GHz: Xeon X7460: 156

4 Socket: SpecInt_Rate: 4 core Nehalem 3.33 GHz: Xeon - X5570: 499
4 Socket: SpecInt_Rate: 6 core Istanbul 2.8GHz: Opteron 2439SE: 416
4 Socket: SpecInt_Rate: 6 core Dunnington 2.67GHz: Xeon X7460: 294

SPEC_2006

Dunnington has a 17% higher Floating Point peak performance as
Nehalem (63.85 vs 53.28 GFlops). Dunnington has 9 times the L2
cache amount as Nehalem (9MB vs 1MB) and Dunnington has twice
the amount of L3 cache (16MB vs 8MB).

But Nehalem runs circles around Dunnington in Spec_FP_rate because
Nehalem has a 3.75 times higher bandwidth as Dunnington (32.0 MB/s
vs 8.53 MB/s)

Istanbul with DDR2 is simply not that much better off with 12.8 MB/s.
(DDR3 should be better with 21.3 MB/s)



Regards, Hans

True (Awesome facts btw), but it will take more than just a DDR3 upgrade in Q1 2010 to level the playing field

Perkam

[Donnie27]

Doesn't show Istanbul's real potential.

1) Not using the bandwidth
2) Not using the cores
3) Not using the power budget

Istanbul with DDR2 is simply not that much better off with 12.8 MB/s.
(DDR3 should be better with 21.3 MB/s)

Regards, Hans

With a great deal of respect and admiration; That still leaves it at an 11.5GB disadvantage.

If #1 were true X2 creams Conroe. #2. Many apps don't make use of more than 4 cores. 6 is not a power of two. #3 for HPC who cares?

I'm just not so sure how important that is. By that I mean merely Bandwidth while not talking about latency (Inter CPU and to RAM) and other factors. Didn't we forget X2 did better with this same bandwidth test than Conroe and we know that ended. Give Istanbul more bandwidth, it wins a few bandwidth tests and still gets spanked in most real world apps. AMD simply needs to Boost IPC or we'll just keep seeing excuses for irrelevant benchmarks. Hyper Threading does work on some/many apps.

http://arstechnica.com/hardware/revi...-review.ars/11

Conclusion

When I set out to write this article, I had three areas I wanted to explore: HT performance, performance scaling in Nehalem vs. the QX9650, and the performance difference, if any, between 32-bit and 64-bit mode. Having done so, I could almost write the shortest conclusion on record: Nehalem is great, Hyper-Threading = generally awesome, and 64-bit > 32-bit mode.

http://www.anandtech.com/cpuchipsets...spx?i=3448&p=5

[informal]

Donnie,Hans was talking about server workloads where BW means a lot,not client one where it means very little. Give more BW to Istanbul and it will surely get better scores in a lot of workloads.How much better?Nobody knows until DDR3 server platform launches(my guess is quite good speed-up in spec rate tests at least,since these are mostly BW bound).
In the end,customers mostly look for perf/watt and AMD is doing very good at the moment. New SE 6core Opty at 2.8Ghz is already listed at HP's site,so the launch is imminent.With DDR3 things can only get faster for Istanbul,and with Magny Cours ,AMD will use simple yet effective tactic of core pair matching in direct connected MCM in order to get 2x more cores,2x more cache and almost 4 times more mem. BW than current Istanbuls(DDR3 vs DDR2 and 4 vs 2 channels).

[Chad Boga]

Donnie,
I think you need to read Informal's sig.

[villa1n]

Donnie,
I think you need to read Informal's sig.

I think from the data Hans has posted one can pretty easily extrapolate that Bandwidth is a bottleneck and that the cores are not being utilized to the full extent. Thats not saying this is a realworld problem that hurts this chip at the moment, or that it will magically landslide anything, but the numbers will definitely go up from where they are, so I guess he s saying lets not jump to conclusions just yet, as to where performance is.. as it has a lot of room to grow, with bandwidth increases, as demonstrated by the data.

[Chrono Detector]

Looks alright, but they need to bump up the clock speed.

[Vozer]

I think you need to read Informal's sig.

Now that you talk about it, I have something that I've always wanted to offer him.

[Chad Boga]

I think from the data Hans has posted one can pretty easily extrapolate that Bandwidth is a bottleneck and that the cores are not being utilized to the full extent. Thats not saying this is a realworld problem that hurts this chip at the moment, or that it will magically landslide anything, but the numbers will definitely go up from where they are, so I guess he s saying lets not jump to conclusions just yet, as to where performance is.. as it has a lot of room to grow, with bandwidth increases, as demonstrated by the data.

I'm not disputing any of that and believe this chip will be handy in a server environment, but not for the desktop.

However I still believe Donnie should read Informal's sig for reasons unrelated to this topic.

[Particle]

Don't get too hung up on x264. In my own testing on my 6-core, no matter what the thread count I'd set, x264 would only use like 75% of the processor.

[Zucker2k]

Doesn't show Istanbul's real potential.

1) Not using the bandwidth
2) Not using the cores
3) Not using the power budget

800 MHz DDR2 is simply not enough for many throughput applications.This is
currently a (temporary) Server platform limitation. Frequency scaling is just
38% (out of 100%) for SpecFP_rate. Just look at the six core Dunnington
to see what a lack of bandwidth does:

4 Socket: SpecFP_Rate: 4 core Nehalem 3.33 GHz: Xeon - X5570: 372
4 Socket: SpecFP_Rate: 6 core Istanbul 2.8GHz: Opteron 2439SE: 283
4 Socket: SpecFP_Rate: 6 core Dunnington 2.67GHz: Xeon X7460: 156

4 Socket: SpecInt_Rate: 4 core Nehalem 3.33 GHz: Xeon - X5570: 499
4 Socket: SpecInt_Rate: 6 core Istanbul 2.8GHz: Opteron 2439SE: 416
4 Socket: SpecInt_Rate: 6 core Dunnington 2.67GHz: Xeon X7460: 294

SPEC_2006

Dunnington has a 20% higher Floating Point peak performance as
Nehalem (63.85 vs 53.28 GFlops). Dunnington has 9 times the L2
cache amount as Nehalem (9MB vs 1MB) and Dunnington has twice
the amount of L3 cache (16MB vs 8MB).

But Nehalem runs circles around Dunnington in Spec_FP_rate because
Nehalem has a 3.75 times higher bandwidth as Dunnington (32.0 MB/s
vs 8.53 MB/s)

Istanbul with DDR2 is simply not that much better off with 12.8 MB/s.
(DDR3 should be better with 21.3 MB/s)



Regards, Hans

Check these out:

http://www.spec.org/cpu2006/results/...505-07308.html

http://www.spec.org/cpu2006/results/...413-07023.html

Look at Memory configuration and copy count and resultant scores. Interesting isn't it? Dunnington is running 24 copies for both tests you listed. Yes, bandwidth does matter, but it's not the entire story.

[kl0012]

Doesn't show Istanbul's real potential.

4 Socket: SpecFP_Rate: 4 core Nehalem 3.33 GHz: Xeon - X5570: 372

Regards, Hans

Since when X5570 has became 3.33 GHZ? Last time I checked it was 2.93. Turbo Bust is very unlikely since all cores are active&busy in this bench.

[Smartidiot89]

What on earth are they thinking comparing 12 cores to 8 cores...... nevermind! The x5550 still wins by miles - hell, even the x5482 held it's own.

This performance was really horrible; this coming from someone with low expectations of these AMD chips. Hopefully HT 3.0 and other tweaks and improvements would radically change the picture or Intel would start raising prices on the highend again.

It's not about Cores vs. Cores on the server/workstation market, its performance/powerconsumption/price.

Yes what you say may be correct but this review has so many flaws We already knew even Nehalem is overkill with 8 threads on desktop and that it gains abit by disabling HyperThreading in games etc. 2.6GHz Istanbul vs. 2.66GHz Nehalem is a fair comparision, not due to clocks but to price.

[blaH]

Don't get too hung up on x264. In my own testing on my 6-core, no matter what the thread count I'd set, x264 would only use like 75% of the processor.

care to post x264 command line? My guess would be --b-adapt 2 with big b frames is used?

[Zucker2k]

It's not about Cores vs. Cores on the server/workstation market, its performance/powerconsumption/price.

Yes what you say may be correct but this review has so many flaws We already knew even Nehalem is overkill with 8 threads on desktop and that it gains abit by disabling HyperThreading in games etc. 2.6GHz Istanbul vs. 2.66GHz Nehalem is a fair comparision, not due to clocks or performance but to price.

Fixed!

Don't worry, maybe things would get better with DDR3, and HT 3.0, and whatever else you guys are looking forward to. I like what Hans termed "(temporary) Server platform limitation;" to me it reads, a poorly thought-out strategy/design which does nothing but make AMD look bad when old gen intel is beating it in benches. In many ways, that is why some of you have no choice but always look to the future for hope and solace. Thing is though, if you want high-end AMD, you can only seriously consider Istanbul.

[PatRaceTin]

AMD Six Core Desktop will be out soon with no surprise because Intel Core i9 will release

[saint-francis]

Don't get too hung up on x264. In my own testing on my 6-core, no matter what the thread count I'd set, x264 would only use like 75% of the processor.

Using avisynth? If so the bottleneck is there, before x264. People get 100% usage from 16 core with x264. There is something wrong with your process that causes this.

[Chad Boga]

AMD Six Core Desktop will be out soon with no surprise because Intel Core i9 will release

I doubt AMD will bring Six core to the Desktop until they implement their own version of Turbo Boost(probably in Bulldozer).

Intel will be able to get away with it because Turbo Boost will hide how useless 6 core is over Quad for most desktop users.

[Clairvoyant129]

Doesn't show Istanbul's real potential.

1) Not using the bandwidth
2) Not using the cores
3) Not using the power budget

800 MHz DDR2 is simply not enough for many throughput applications.This is
currently a (temporary) Server platform limitation. Frequency scaling is just
38% (out of 100%) for SpecFP_rate. Just look at the six core Dunnington
to see what a lack of bandwidth does:

4 Socket: SpecFP_Rate: 4 core Nehalem 3.33 GHz: Xeon - X5570: 372
4 Socket: SpecFP_Rate: 6 core Istanbul 2.8GHz: Opteron 2439SE: 283
4 Socket: SpecFP_Rate: 6 core Dunnington 2.67GHz: Xeon X7460: 156

4 Socket: SpecInt_Rate: 4 core Nehalem 3.33 GHz: Xeon - X5570: 499
4 Socket: SpecInt_Rate: 6 core Istanbul 2.8GHz: Opteron 2439SE: 416
4 Socket: SpecInt_Rate: 6 core Dunnington 2.67GHz: Xeon X7460: 294

SPEC_2006

Dunnington has a 20% higher Floating Point peak performance as
Nehalem (63.85 vs 53.28 GFlops). Dunnington has 9 times the L2
cache amount as Nehalem (9MB vs 1MB) and Dunnington has twice
the amount of L3 cache (16MB vs 8MB).

But Nehalem runs circles around Dunnington in Spec_FP_rate because
Nehalem has a 3.75 times higher bandwidth as Dunnington (32.0 MB/s
vs 8.53 MB/s)

Istanbul with DDR2 is simply not that much better off with 12.8 MB/s.
(DDR3 should be better with 21.3 MB/s)



Regards, Hans

Since when was the X5570 3.33GHz?

[Shintai]

Not to mention the MB/sec that should be GB/sec just as another obvious error...