Intel Readies a 5.1 GHz Xeon Chip Based on the "Broadwell" Architecture

[TheoneandonlyMrK]

165 Tdp on a quad.

Now that's funny from intel, I'd say for them that's the GHz race done and dusted my fx pushes 8 cores to that speed at a 95Tdp,, a design that's well mocked from 4 years ago.

Yes its Ipc would be better ,obviously

 

[FreedomEclipse]

 

[HumanSmoke]

165 Tdp on a quad.

Now that's funny from intel, I'd say for them that's the GHz race done and dusted my fx pushes 8 cores to that speed at a 95Tdp,, a design that's well mocked from 4 years ago.

Yes its Ipc would be better ,obviously

Bask in the glory, because it is certainty that once AMD transitions to 14nm, they'll run into the same wall that Intel is at. I'm guessing that if Zen is ~ 4Ghz (or less) with limited OC because of the demands of transistor switching power envelopes, people will be falling over themselves to blame the process node, transistor size and density.

 

[cyneater]

Looks like intel has gone full retard on this one...

165w TDP wow....

And 2 of this bad boys would be good for CPU insensitive tasks...

I wonder what the price tag is ..

 

[NC37]

And only the 5 richest kings on the planet can afford it

 

[Frick]

2000-series E5s indicates it's a CPU that can be put in 2p boards. Dual 5Ghz quads anyone? That premium clock speed will be associated with a premium price though, this is Intel after all.

EVGA Classified XR or something.

 

[bonehead123]

Just be sure to have 4 things with you when you go to buy one of these:

A. Your banker or HIS gold card
B. Some strong booze...
C. Some REALLY good chemicals
D. A Huge box of kleenex.....

 

[GhostRyder]

Now this is a processor, no matter what I want to see one in action!

I am all for seeing higher clocks again, nice change of pace!

 

[TheoneandonlyMrK]

Bask in the glory, because it is certainty that once AMD transitions to 14nm, they'll run into the same wall that Intel is at. I'm guessing that if Zen is ~ 4Ghz (or less) with limited OC because of the demands of transistor switching power envelopes, people will be falling over themselves to blame the process node, transistor size and density.

Not me I'm educated;p , i see Intel's ipc gain as it is, using all resources in a Ht enabled core via macro ops in 1 thread if required,something bulldozer never could do, now Zens a different beast ,but yeah well see ,be interesting to see Amd(perhaps should have been And)leverage its power gateing and optimizing tech too ,they didn't do too bad for the node size.

Anyone think this could just be a special for some super computer somewhere and not going to be widely available.

 

[buildzoid]

Not me I'm educated;p , i see Intel's ipc gain as it is, using all resources in a Ht enabled core via macro ops in 1 thread if required,something bulldozer never could do, now Zens a different beast ,but yeah well see ,be interesting to see Amd(perhaps should have been And)leverage its power gateing and optimizing tech too ,they didn't do too bad for the node size.

Anyone think this could just be a special for some super computer somewhere and not going to be widely available.

I think this chip is probably meant to compete with IBMs 5Ghz+ CPUs and AFAIK those chips are used for high frequency trading.

Personally I wouldn't mind having 8 cores clocked at 5.1Ghz with enough PCI-e lanes to run 4 way at x16 on all slots.

 

[Scrizz]

Fact that all Xeon processors do not have opened multipliers is the ground for Xeon brand.

No, not really. Xeons use to be very overclockable.
I think the last ones were the skt1366 ones

 

[FordGT90Concept]

The only way this makes sense is if someone (or some industry) demanded low core count/high clockspeed/redundancy for specific applications from Intel. I can't think of what that application is but it must exist for Intel to bother with this launch.

Maybe deep learning AI?

 

[Aquinus]

I can't think of what that application is but it must exist for Intel to bother with this launch.

Databases in general? Tasks like indexing and reindexing tend to be much more CPU heavy than I/O heavy and like most database related tasks, it's single-threaded for the single request that made it.

 

[FordGT90Concept]

It'd be interesting to see how these processors stand up to a similarly priced two-way SPARC system running a heavily trafficked database. That certainly would explain the redundancy.

 

[Aquinus]

It'd be interesting to see how these processors stand up to a similarly priced two-way SPARC system running a heavily trafficked database. That certainly would explain the redundancy.

Heavy database use with multiple users is inherently multi-threaded because of the individual queries that come in on separate connections. Almost always (if I/O is there,) can databases scale linearly to cores if no locking is occurring. I was very careful to say indexing and reindexing because those are two cases where it can't run on multiple cores and almost is always a CPU bottleneck. More often than not, if you want a database to scale for users then more cores is what you want. If you want the workload to scale better with only a handful of users (heavy weight queries, fewer users as opposed to a ton of users and lightweight queries,) then more single-threaded performance is what you'll be wanting.

I just did database maintenance at work, removed millions of duplicate historical records re-parented them based on the unique data kept. A task like that tends to be single-threaded (certainly the way I did it for safety reasons,) so instead of taking 150 minutes to run, a CPU clocked twice as high would complete is probably a little over half of that. That means under 1 1/2 hours of down time instead of 3. That alone can make a big difference if you don't consider everything else.

My point is that even within a group of server tasks like running a database, even then only in some unique situations will more clocks outpace more cores or vise versa.

 

[geon2k2]

Hmm, something is fishy.

Xeons are supposed to be ultra reliable and they typically work in servers in 24x7 mode for years.

I really don't see a 5 Ghz part doing that in a server, and even if such a part will exist if I'd be the decision maker when buying such a thing I'd for sure recommend something more conservative.

 

[Aquinus]

Hmm, something is fishy.

Xeons are supposed to be ultra reliable and they typically work in servers in 24x7 mode for years.

I really don't see a 5 Ghz part doing that in a server, and even if such a part will exist if I'd be the decision maker when buying such a thing I'd for sure recommend something more conservative.

Nothing fishy about it. It wouldn't be the first time Intel has released a 150 watt TDP Xeon part. The Xeon X5365, X5482, X5492, and the E5-2687W are all examples of 150-watt TDP Xeon chips.

 

[GC_PaNzerFIN]

If you think releasing a Xeon part with over a GHz higher clock speed than any other model including overclocking K series consumer parts is too good to be true, it generally is.

 

[TheGuruStud]

If you think releasing a Xeon part with over a GHz higher clock peed than any other model including overclocking K series consumer parts is too good to be true, it generally is.

But those are a lot of grains. More grains means it true!

 

[HumanSmoke]

If you think releasing a Xeon part with over a GHz higher clock speed than any other model including overclocking K series consumer parts is too good to be true, it generally is.

As I pointed out earlier in this thread, Intel's fastest chip is a Xeon, the X5698

 

[TheGuruStud]

But dual core, which is the only way they kept consumption in check.

We all know how these things eat power when you start overvolting and OCing even just a little.

 

[cadaveca]

IF intel can push 165W through their silicon, this is actually quite the triumph. I could care less about anything else other than that power consumption on such a small piece of glass. I know the chips can handle it, since many of us are pushing similar power through our overclocked chips, and have run them that way for years, too. So it's not very surprising, to be honest, be to see Intel a bit more focused on performance rather than power-savings is something new and interesting.

 

[HumanSmoke]

But dual core, which is the only way they kept consumption in check.
We all know how these things eat power when you start overvolting and OCing even just a little.

Not in dispute, nor does it have any relevance to the post I quoted or my reply.

IF intel can push 165W through their silicon, this is actually quite the triumph. I could care less about anything else other than that power consumption on such a small piece of glass. I know the chips can handle it, since many of us are pushing similar power through our overclocked chips, and have run them that way for years, too. So it's not very surprising, to be honest, be to see Intel a bit more focused on performance rather than power-savings is something new and interesting.

Intel's installed base and market share have allowed them to target virtually every possible sector. Their previous v3 (Haswell) ran to over a hundred seperate SKU's (including some models specifically of individual customers), and I'd expect the company to continue the trend of exploring every parameter ( power, core count, clock, clock turbo bins) if there's money to be made from it. Haswell's Xeon managed a range of 2 cores @ 13W to 18 cores @ 175W, so I wouldn't think it beyond the realms of possibility that Intel will continue the practice.

 

[qubit]

As the stock speed is so high I wonder if it will be a good overclocker?

Percentage-wise to its stock speed, 6GHz+ should be possible on air without much effort, but I somehow doubt it.

 

[cadaveca]

As the stock speed is so high I wonder if it will be a good overclocker?

Percentage-wise to its stock speed, 6GHz+ should be possible on air without much effort, but I somehow doubt it.

As do I, given the power consumption. I don't imagine that simply adding a few "metal layers" on top was enough. Remember the HBM vias thing? Like if they can push what would be 200-250W on decent cooling through a broadwell core, then there has been significant changes to the design here. If we are talking Boardwell-E though, then this is pretty basic, and entirely believable.