I assure you that $5500 is a substantial addition to the TCO of buying and running the server.
The reason these parts exists is for optimising for expensive software that's charged per core. For the same performance bump, you can add two cores and pay Intel $2000 and Oracle $10000, or you can buy a chip clocked higher and pay Intel $5000.
I assure you that $5500 is a substantial addition to the TCO of buying and running the server.
I assure you it's not. This thing is going to go into a 2U server, which is going to run ~$400 month for the rackspace, power and bandwidth. It's also not going to be purchased by folks building their own stuff from Newegg based on supermicro chassis and misc parts, it's going to be bought from Dell and HP as one part of the expense of a $30k+ server.
Could I build something cheaper and find cheap local colo to make it a 'significant' portion of TCO? Yes, doesn't change anything about the reality that this is not the target market so it makes no sense to be concerned about the cost in such a market. TBH if this segment of the market is your concern, you should buy a threadripper and hope the ECC ram you bought works with it. But, it's not the same market.
The numbers you're quoting don't match your claims. You said "the cost of the hardware is pretty much a blip in comparison to rackspace, power and networking costs."
Then once you start listing numbers, the server costs $30k+ and the data center $400/mo. I'm pretty sure that $30k for a server is not a "blip" on ($30k + $15k) assuming a three year lifetime.
Trying to make out like I'm coming it from some kind of hobbyist market segment point of view is a puzzling straw man. These products, which optimize for performance over performance/TCO$, are built with very valid niches in mind, and where the math works out it saves you a ton of money in software licensing or software development.
They are not, as you seem to be claiming, positioned as sensible choices for wide deployment in datacenters. If that were the case, why bother with the rest of the SKUs!?
Comments
I assure you that $5500 is a substantial addition to the TCO of buying and running the server.
The reason these parts exists is for optimising for expensive software that's charged per core. For the same performance bump, you can add two cores and pay Intel $2000 and Oracle $10000, or you can buy a chip clocked higher and pay Intel $5000.
I assure you it's not. This thing is going to go into a 2U server, which is going to run ~$400 month for the rackspace, power and bandwidth. It's also not going to be purchased by folks building their own stuff from Newegg based on supermicro chassis and misc parts, it's going to be bought from Dell and HP as one part of the expense of a $30k+ server.
Could I build something cheaper and find cheap local colo to make it a 'significant' portion of TCO? Yes, doesn't change anything about the reality that this is not the target market so it makes no sense to be concerned about the cost in such a market. TBH if this segment of the market is your concern, you should buy a threadripper and hope the ECC ram you bought works with it. But, it's not the same market.
The numbers you're quoting don't match your claims. You said "the cost of the hardware is pretty much a blip in comparison to rackspace, power and networking costs."
Then once you start listing numbers, the server costs $30k+ and the data center $400/mo. I'm pretty sure that $30k for a server is not a "blip" on ($30k + $15k) assuming a three year lifetime.
Trying to make out like I'm coming it from some kind of hobbyist market segment point of view is a puzzling straw man. These products, which optimize for performance over performance/TCO$, are built with very valid niches in mind, and where the math works out it saves you a ton of money in software licensing or software development.
They are not, as you seem to be claiming, positioned as sensible choices for wide deployment in datacenters. If that were the case, why bother with the rest of the SKUs!?