Is Tyan about to launch its seventh AMD64 motherboard?

WE REPORTED last week that Tyan had a quad Opteron board in the works. The Korean Darkcrow Web site has published the first picture of Tyan's latest AMD64 offering.

That board was displayed at an AMD Reseller Seminar, and it seems evident that it's ready to be launched. So it appears that we're just waiting on Tyan to formally announce the product. The picture confirms the 4+2+2+2 DIMM layout, and this particular board supports Ultra 320 SCSI. So it appears that S-ATA support will be a standard feature, and U320 SCSI will be a cost option. The 2CPU.com forum is showing a schematic which gives more detail on the hardware components used.

When you compare this motherboard to Tyan's Xeon counterpart, which one is the better deal? The Xeon board does not have on-board support for Gb Ethernet and S-ATA, and the Ultra SCSI is the U160 variety. It's not a 64-bit platform either. So on the features front, it's already looking like yesterday's technology.

If we do some bandwidth comparisons as well, it's clear to see that the current four-way Xeon is a generation behind Opteron in the throughput stakes:

How much less than a quad Opteron?
FSB bandwidth - Up to 5.5x
Inter processor bandwidth - 2x
Memory bandwidth - Up to 3.3x
I/O bandwidth - Up to 5.3x

Xeon is able to sport higher frequency and copious amounts of cache, which helps it from being a totally defeated loser. But as I've said before, and as can be seen in the benchmarks below, this will not be enough to save the platform's bacon from the frying pan.

SPECint_rate2000 4P Servers
Opteron 846 56.7 (117%)
Opteron 844 48.5 (100%)
Xeon MP 2.8 GHz (2 MB L3) 47.6 (98%)
Xeon MP 2.0 GHz (2 MB L3) 34.7 (72%)

Opteron is showing a 19% performance lead in the first benchmark discipline, and it's doing so without using PC3200 memory.

SPECfp_rate2000 4P Servers
Opteron 846 - 52.5 (107%)
Opteron 844 - 49.2 (100%)
Xeon MP 2.8 GHz (2 MB L3) - 25.4 (52%)
Xeon MP 2.0 GHz (2 MB L3) - 20.2 (41%)

Xeon is again found wanting, as Opteron punches out a mammoth 55% lead. Opteron is still using slower PC2700 memory.

SPECweb99_ssl 4P Servers
Opteron 848 - 3800 (112%)
Opteron 846 - 3399 (100%)
Xeon MP 2.8 GHz (2 MB L3) - 2177 (64%)

Opteron is hitting a 48% lead in this category, which happens to be another double digit embarrassment. In this configuration, Opteron could make do with PC2100 memory.

SPECweb99 (Red Hat CA2) 4P Servers
Opteron 846 - 10502 (104%)
Opteron 844 - 10135 (100%)
Xeon MP 2.0 GHz (2 MB L3) - 6700 (66%)
Xeon MP 1.6 GHz (1 MB L3) - 5750 (57%)

Opteron's home run number this time around is 38%. Again with this setup, Opteron could get buy with PC2100 memory.

SPECjbb2000 Performance 4P Servers
Opteron 848 - 113473 (105%)
Opteron 846 - 107786 (100%)
Xeon MP 2.8 GHz (2 MB L3) - 94405 (88%)

Opteron's lead is held back to a still very respectable 17%.

MMB2 (Windows) 4P Servers
Xeon MP 2.8 GHz (2 MB L3) - 16528 (106%)
Opteron 844 - 15520 (100%)
Xeon MP 2.0 GHz (2 MB L3) - 13200 (85%)

Xeon gains its first win with a single digit lead. Opteron will no doubt win back this benchmark before the end of the year is out.

The above results were compiled from AMD's competitive server benchmarks page. The Opteron 848 results are still under SPEC submission. Where newer results have been submitted, they've been used as well.

If you haven't been living on the moon these past few months, you'll now know that Xeon is clearly second best in the majority of server benchmarks. However, Xeon currently rules the roost in Microsoft's Exchange MMB2 benchmark. It has a 6% lead over the second placed Opteron. Since the publication of that result, Opteron's processor speed has increased by 22%. With official PC3200 memory support as well, it would be interesting to see how much better the Opteron MMB2 benchmark number would improve. We might not have to wait that long to find out.

Opteron's slaughter of Xeon in the SPEC benchmarks is only part of the story. One has to also take into account how Opteron scales with workload. I gave two examples of this in an earlier piece:

"ExtremeTech's Nile application [Now courtesy of PC Magazine] server benchmark emulates a book ordering transaction processing environment. At 500 users, Xeon took almost twice as long as Opteron to complete the same test. So to be competitive with Opteron, you'd need to be using "two" Xeon servers.

"AnandTech had its own database benchmark, which also highlighted the scaling prowess of Opteron. When the data base setting was increased to 20x, Opteron was able to complete 450 more transactions per second. Don't do what I did and calculate the 38% better performance figure. Over one minute, Opteron will process 27,000 more transactions. Over one hour that increases to 1,620,000. Again you'd have to use "two" Xeon servers to better the Opteron result.

It should be noted that these benchmark results are for two-way systems. Knowing how Opteron scales from two-way to four-way, the four-way numbers should surely get better when Xeon MP is in the frame."

So for those that have a choice, the decision to go with Opteron would appear to be a simple one. For those that have to buy from Dell or HP, you can dream about what might have been, or you can show the backbone that the EDA users displayed when they demanded Opteron support and got it.

When you compare Opteron's SPEC performance lead over Xeon - 17%, 19%, 38%, 48%, and 55% in five different quad disciplines, and let's not forget Opteron's single/dual SPEC leadership as well - it shows breathtaking arrogance from Dell and HP when it is not giving its customers the option to buy the Opteron platform, especially when IBM is doing so and Sun will do the same. To further illustrate my point, imagine the following scenario:

If Opteron's performance lead was transposed into the car world, how do you think the Opteron brand would be received? Imagine two independent car dealerships that compete against each other. They face each other across the high street, so customers play them off to garner the best possible deal. Both companies have been successfully selling the Xeon brand, but one of them decides it's time to give their customers the option to buy an Opteron. When customers are told that for the same price Opteron betters Xeon in the crucial areas that matter - 17% lower emissions, 19% faster top speed, 38% better fuel economy, 48% quicker 0-60, 55% shorter braking distance, and then find out that it can also pull twice the load of Xeon, which car do you think is going prove to be the better seller?

Also, if the car dealership that wasn't selling the Opteron brand refused to do so because of the relationship it has with the Xeon manufacturer, how long do you think it would remain a viable business? Maybe a month? A week before the cash flow dried up, and three weeks before the cash reserves were drained dry.

So if you had to decide between the Opteron and Xeon as your next family car, would its performance be important to you? Or would you be happy buy a Xeon car from a Dell or HP dealership? µ

See Also
Tyan has a quad Opteron board in the works
New Opterons extend AMD benchmark leadership
EDA users demand Opteron support and get it