Cisco’s server power engineering team recently compared the Cisco UCS 5108 Blade Server Chassis with B200 M3 blade servers against HP’s BladeSystem c7000 Enclosure and HP ProLiant BL460c Gen8 blade servers.
The results weren’t surprising internally, but they might surprise you. The main findings were:
- Efficiency
- The Cisco UCS 5108 enclosure configured with Cisco UCS B200 M3 blades achieved a 6.3% higher performance-to-power ratio than the HP BladeSystem c7000 with HP ProLiant BL460c blades, using with similarly configured hardware, BIOS, and OS settings.
- At a 70% target load, the Cisco UCS blade solution consumed 6.5% less power than the HP ProLiant blade solution.
- Power
- At the maximum target load, the Cisco UCS blade solution consumed 192 watts (W), or 7.2% less power, than the equivalently configured HP BladeSystem solution.
- The Cisco UCS blade solution consumed 167 watts, or 17.4% less power, while idle than the equivalently configured HP BladeSystem solution: a difference equivalent to the power consumed by three idle HP ProLiant BL460c Gen8 blades.
- Including the power consumed by a pair of redundant Fabric Interconnects, the Cisco UCS blade solution consumes less power than an equivalent HP BladeSystem solution.
The complete results are posted on Cisoc.com (link to download the entire white paper [updated to 2014 version of the whitepaper]) so that you can replicate the results independently.
What are our customers saying about their power and cooling savings? See my previous blog post “Cisco UCS Servers – Watts driving your power and cooling costs?”
How much could you save by switching to UCS?
Would you like to learn more about how Cisco UCS can help you? There are more than 250 published datacenter case studies on Cisco.com. Additionally, there is a TCO/ROI tool that will allow you to compare your existing environment to a new UCS Solution. For a more in-depth TCO/ROI analysis, contact your Cisco partner.
This is great information! I recently found myself in a debate with a VAR about this.. will hang on to these details for the next conversation!!
What details? “equivalently configured” does not give details. Also it seems it was the older c7000 and not the platinum model. Also as the c7000 holds 16 blades compared to UCS chassis holding 8 I would expect it to consume more power. More details please otherwise it is useless FUD.
Lee,
I’m confused. Doesn’t Appendix A & B, Tables 2, 3, & 4 give all the configuration information as to the solutions tested?
What details were missing from the whitepaper?
Bill
Hi Bill, I’m on my mobile so i haven’t read the whitepaper yet. Will save that for tomorrow when back at work on laptop. Got directed to the page from twitter and was referring to this page only. Hopefully that clears the confusion. Looking forward to reading the whitepaper in full.
Lee
First full disclosure. I work for HP. after reading the actual report, I found it interesting that you didn’t include the power for the 6248 interconnects. I’m not a UCS specialist, but I understand that without the 6248 interconnects, the UCS would not be able to function at all, or at the very least, not pass any traffic.
Since the HP C7000’s parts required for the whole system are in the chassis, I would find it interesting to know what the actual comparison would be when you have do apples to apples between two fully functional systems.
Can you please describe what the power comparisons would be when the 2 x 6248 interconnect models are included in the power measurements?
@netmanchris
Chris,
Thanks for taking the time to read the blog, white paper and post a comment.
To your question, I would point you back to hardware configuration #2 (page 7 in the PDF) and the results in Figures 5/6 (pages 8/9) where the Fabric Interconnects are included in the measurements which show a Cisco advantage. This is a conservative savings as no ToR switching was included in the HP configuration.
Thanks again.
Bill
I think that’s a really important point, Bill. The UCS power numbers in Figure 5 include power for switching between all 80 blades whereas the HP numbers only include local chassis switching.
Furthermore, there is the problem of having to power duplicate devices in each chassis. I thought this quote from the paper summed it up nicely: “The unified fabric power savings advantage becomes more apparent as the number of blade enclosures increases, because for every HP blade enclosure added to the system, duplicate onboard administrators and virtual connect switches are needed. As the number of blades increase, the power savings decisively favors the Cisco UCS blade solution.”
It is interesting how the power savings follow the same trend as cost savings… UCS scales better.
I don’t think it can be said better than Doron just said it. UCS just scales better. If we’re wrong about anything in the paper, please point us to any data showing different.
As an end user of both Cisco UCS blades and HP C7000, all of my purchases have shifted towards UCS. While certainly not perfect, the ease of management and configuration blows away HP. Firmware and server profile management are godsends and a nightmare with Flex*. HP is not a bad product by any means, but the management concepts in UCS are light years ahead.
Hey there!
Full Disclosure: I work for HP
I am thinking of setting up your test in my lab, can you post your OA (Onboard Administrator) power settings and if you were using single or three phase power? Also in looking over the paper, It looks like you used the FI’s in one test and not in another. Is that a typo?
-Neil
Neil,
The OA power settings are on page 4 of the PDF (“The blade enclosure power management policy was set to grid redundant for each solution.”).
As to the Fabric Interconnects, two different configurations were tested. Hardware configuration 1 was without FIs to show the incremental power of scaling of additional UCS into a solution. Hardware configuration 2 includes FI power and clearly shows the power efficiency of Cisco UCS solution as the solution scales under a single pair of redundant fabric interconnects.
Bill
Good clarity on the Fabric Interconnects (FI) Bill. It is important to understand your point that both power and cost for the FI’s are spread across all the attached chassis/blades and rack servers. As you stated, this is the reason that UCS scales so well and so easily.
Great post.
Thanks for the post, nice power figures on the UCS blade.