GRC & Minimus Servers Blog

GRC’s Oil Immersion Cooling System at PIC Supports Data Processing for The Large Hadron Collider at CERN

Posted by Dhruv Varma on April 19, 2017

PIC in Barcelona, Spain shares key performance and reliability results from its immersion cooled data-intensive cluster.

[April 19, 2017 – Austin, TX] Green Revolution Cooling (GRC), a leader in immersion cooling, today announced key performance and reliability results from its installation at Port d’Informació Científica (PIC) in Barcelona, Spain. The ultra-efficient cluster installed in October 2015 has since been used to process dozens of Petabytes of data from CERN's Large Hadron Collider, and leading-edge astrophysics projects.

Read More

Topics: data center efficiency, hardware, cost, cost optimization, efficiency, power, data center, News & Events, sustainability, reliability, press release

Bursting the Over-Provisioning Bubble

Posted by Peter Poulin on April 18, 2017

What’s a CIO to do? Business Unit leaders are notoriously unpredictable in forecasting their demands for new applications and IT support. Yet, it is critical that you respond quickly to these demands so the company can capture the incremental revenue, cost reduction, and/or customer service improvement that justifies the IT investment. In the past, one way of dealing with this challenge was the over-provisioning of datacenters – essentially assuring that there was always capacity on demand.

Read More

Topics: cost, cost optimization, data center, TCO

Whitebox Servers: Is It Time to Switch? Hyperscalers Seem to Think So

Posted by Christiaan Best on March 14, 2017

A 2015 research report from IDC1 indicates that the market is increasingly abandoning big brand, OEM servers in favor of “whitebox” (unbranded) servers made using parts from ODM suppliers.

  • HP’s server sales went down almost 30% between 2010 and 2015.
  • Dell’s sever sales were pretty much flat with a 4.5% annual (server) revenue growth. Dell’s sales seem to be increasingly focused on whitebox style servers for hyperscale companies.
  • IBM’s server sales are down almost 70% during the same 5 year period (2010-2015).

A separate study released by Lawrence Berkley National Labs2 shows that while the major players are in decline, the market overall is getting bigger with white box servers taking over the market.

Read More

Topics: Servers, components, hardware, cost, cost optimization, TCO

Picking the Right Processor: $/GFLOP with Different Intel Models

Posted by Christiaan Best on August 16, 2016

Most HPC buyers look to get the most performance for the lowest budget. As mentioned in an earlier post 'Server Cost Breakdown By Components', processors are normally the most expensive component in a cluster, this is especially true for non-GPU / Co-processor accelerated clusters. Picking the right CPU is critical.

Read More

Topics: HPC Cluster, cost, Processors, HPC, data center, TCO

The Incredible Shrinking Price of Memory

Posted by Christiaan Best on June 30, 2016
As covered in our previous blog reviewing server component costs, memory is often a dominant cost when buying a server. Surprisingly, memory prices are also one of the most volatile. Understanding trends are helpful to keeping your server costs low.

There are a couple of important points for those managing a data center:
  • Pricing can change fast, especially when a new line of processors chooses to support only certain standards. For example, when Intel E5-26xx v4 called out DDR4-2400 vs the DDR4-2133 for Intel E5-26xx v3.
  • While a specific model may go up or down, the overall trend in memory prices are downward. In some cases this is as much as 30% in 4 months.
  • Keeping on top of current memory is a constant battle. For example, we just updated minimusservers.com with significantly lower pricing despite being online for just over 2 months.
  • The current price for DDR4-2400 ECC memory is around $4.70 per GB.
    Note: this is Minimus pricing, pricing form other sources may be higher.
The website www.pcpartpicker.com has a great graphic for tracking general trends. The below graph shows pricing for many different brand and models of 4GB DDR4-2133 over time.


Figure 1: Price trends of 4GB DDR4-2400 from Jan 2015 to June 2016
Image credit: pcpartpicker.com



Comparing the pricing trends for two specific models of 16 GB DDR4 2133 from the same brand (Kingston), the pricing for one model is a lot more stable than the other.

Figure 2: Volatile price of the Kingston KVR21R15D4/16 (16 GB DDR4-2133)
Image credit: pcpartpicker.com



Figure 3: Relatively stable price trend of the Kingston KTH-PL421/16G (16 GB DDR4-2133)
Image credit: pcpartpicker.com



There are several conflicting factors that affect the daily price variance of memory:
  1. The cost of producing memory, like nearly all silicon based computer components, is largely from enormous upfront capital costs amortized over time. The marginal cost is actually quite low.
  2. When a new specification is released, pricing starts out high because of low demand, little competition, and low manufacturing yield. As a product matures and gains adoption, prices drop.
  3. Supply and demand does play a big role. Prices can change daily if demand is high or market conditions change.
  4. Manufacturers often increase their prices on specific models to incentivize customers to buy different models that cost less to produce. Searching against a specific configuration, you can see that memory costs will vary as much as 300% despite the same performance specification.

Check out the Minimus formula for cost savings and see a price comparison:
Read More

Topics: Servers, components, hardware, cost, cost optimization, memory

Experiment: Temperature Set Point vs Server Fan Power

Posted by Alex McManis on June 17, 2016
There is a prevalent industry focus on increasing data center temperature to improve PUE and lower power usage, but when does it go too far? While it may be common knowledge that raising data center air temperature to above 70°F saves power, finding the sweet spot and exploring higher temperatures with free cooling takes some experimentation.

Even if you aren’t considering raising your data center set point, consider that inlet air temperature varies significantly from top to bottom of the rack. Airflow engineering attempts to even out airflow across the room, but a variation of 10°F is still not uncommon.

To find the effect of higher temperatures, GRC decided to measure the power of a server with varying input air – and you should, too. Using a power meter and thermocouple, we gradually increased the amount of recirculation to increase the input temperature into the server.

*Note: Raising the air set point is to save power on the chiller. We can’t test that affect because GRC data centers are chiller-less.

The following data is taken from a 2009 Dell R410, populated with a single, low power processor. Since the chassis is lightly loaded, the fans aren’t working strenuously. Fan power in other chassis is shown in Table 3 below.


Table 1: Dell R410 (2009) lightly loaded with a single processor, representing relatively low fan power
Ambient Temperature (ºC) Ambient Temperature (ºF) Power (Watts) Increase (∆ 27ºC)
27ºC 80.6ºF 137.7 -
29ºC 84.2ºF 139.4 +1%
31ºC 87.8ºF 141.5 +3%
33ºC 91.4ºF 144.0 +5%
33ºC 91.4ºF 144.1 +5%
35ºC 95.0ºF 148.4 +8%
37ºC 98.6ºF 154 +12%


Table 2: Distributed Computing Load and Air Temperature


We found the server power increased logarithmically with input temperature. We also measured the difference between servers in air versus servers submerged in GRC cooling equipment.

Table 3: Power Savings from Submerging in Oil
Brand Model Size Power Savings in Oil*
OpenCompute Winterfell 2U 9%
Dell 1950, gen III 1U 30%
Dell R410 1U 8%
Dell C6100c 2U, 8 CPU 17%
Supermicro 1026 1U 20%
IBM X3550 1U 8%
HP 585 G5 4U 18%
CoinTerra Miner 4U 29%
*Compared to 75-80ºF inlet air temperature

With your servers, how much power could you save by running the fans slower? How much if you submerged the servers and removed the fans altogether?

To learn more about Green Revolution Cooling's solutions call us at +1 512-692-8003 or contact us here.


Read More

Topics: fans, temperature, Servers, components, hardware, cost, cost optimization

Transformer Efficiency

Posted by Christiaan Best on June 8, 2016

Transformer Efficiency: Is it worth it?

PUE often encompasses all of the losses from the utility down to the server. This includes transformers that convert power down to 400 or 208, 3 phase power. To get an understanding of your "total" efficiency, transformer efficiency and costs have to first be understood.

Here are few things of interest to data center operators:
  • There are several factors that affect transformer inefficiency (electrical losses):
    • Hysteresis due to charging the field in the iron core (which is NOT proportional to load).
    • Line losses through the coil (which are proportional to load).
  • There are several different type of transformers:
    • "Standard temperature rise" – More efficient at partial loads, where data centers actually operate.
    • "Low temperature rise" – This means there is a low surface temperature (less heat coming off the transformer). These transformers have larger iron cores, which make them more efficient at transferring energy at high loads, but less efficient at partial loads, where data centers actually operate.
    • DOE 2016” – In the US, the Department of Energy released a new efficiency standard for all transformers sold. These transformers are more efficient across the board.
Below, Figure 1 shows a typical efficiency curve for a 225 kVA transformer from Square D. This is juxtaposed against older, standard temperature rise, and low temperature rise transformers.

Figure 1: Transformer efficiency curves for standard, low temperature, and DOE 2016 type transformers
Read More

Topics: data center efficiency, transformers, cost, efficiency, power

Server Cost Breakdown By Components

Posted by Christiaan Best on June 1, 2016
A lot of people are curious to know how server costs breakdown, component by component. The costs can vary hugely depending on the configuration. Discounts also apply when you buy components in larger volumes. We put together an example below for a typical air-cooled server bought in quantities of a few hundred servers. There are a couple of takeaways:
  • Different buyers get different pricing* based on volume, type of organization, application, relationship with sales rep, etc. Big companies can end up spending as much as 40% less than small businesses.
  • CPUs are a dominant cost and are rarely negotiable
  • Memory and storage pricing is relatively constant between vendors after discounts
*GRC does not practice price discrimination.


Server Cost Breakdown

Note: the prices in the below table have been sourced from popular electronics vendor websites, and are not necessarily indicative of Minimus pricing.

Component Quantity Price Estimate Extended Total
CPU: E5-2650v4 (12 CORE) 2 $1,116.00 $2,232.00
Memory: 16GB, DDR4-2400, ECC 8 $95.00 $760.00
Storage: Boot SSD, 120GB 1 $119.00 $119.00
Storage: 480GB, Medium Endurance SSD 2 $391.97 $783.94
Network Card: None 0 $0.00 $0.00
Chassis Costs      
Motherboard: Dual Socket E5-26xx v4, 8 Memory DIMMS, On-board Network: 2x RJ45 Gbe LAN Ports, 1x RJ45 IPMI LAN Port 1 $299.00 $299.00
CPU Heat Sink 2 $14.00 $28.00
Power Supply 1 $95.00 $95.00
Storage Backplane 1 $75.00 $75.00
Drive Caddies 4 $13.00 $52.00
Fans 5 $10.00 $50.00
Internal Cables 1 $20.00 $20.00
Riser Cards 1 $19.30 $19.30
Sheet Metal Case 1 $100.00 $100.00
Assembly Labor and Test 1 $150.00 $150.00
10% Markup 1 $478.32 $478.32
    Total Cost $5,261.56



About GRC and Minimus Servers

Green Revolution Cooling has introduced Minimus Servers in partnership with hardware giants Supermicro and Gigabyte. The Minimus platform offers reliable, purpose-built servers for a fraction of the cost of OEM servers. The core of Minimus’s price-effectiveness is that customers get exactly the configuration they need, but without the added costs of extraneous features and brand-related premiums that are normally tacked onto OEM offerings.

To learn more about Minimus Servers contact us at [email protected] or call +1 512-692-8003.


Read More

Topics: Servers, components, hardware, cost, cost optimization