Live data from Hacker News

Intel 28-core fantasy vs. AMD 32-core reality

techspot.com

121–130 of 134 posts

Re: Intel 28-core fantasy vs. AMD 32-core reality

#121

As an outsider to 'enterprise-grade' computing, I'm curious about situations where a high number of cores in a single processor would be superior to multiple processors with the same total energy draw sitting on a single motherboard? I can understand HPC applications where the high-speed interconnect on the chip would make a big difference. But in business applications where the cores are dedicated to running indepen…

> As an outsider to 'enterprise-grade' computing, I'm curious about situations where a high number of cores in a single processor would be superior to multiple processors with the same total energy draw sitting on a single motherboard?

Databases are the big one I'm aware of.

Intel's L3 cache is truly unified. Intel's 28-core Skylake means that the L3 of a Database is TRULY 38.5MB. When any core requests data, it goes into the giant distributed L3 cache that all cores can access efficiently.

AMD's L3 cache however is a network of 8MB chunks. Sure, there's 32MB of cache in its 32-core system, but any one core can only use 8MB of it effectively.

In fact, pulling memory off of a "remote L3 cache" is slower (higher latency) than pulling it from RAM on the Threadripper / EPYC platform. (A remote L3 pull has to coordinate over infinity fabric and remain cohesive! So that means "invalidating" and waiting to become the "exclusive owner" before a core can start writing to a L3 cache line, well according to MESI cc-protocol. I know AMD uses something more complex and efficient... but my point is that cache-coherence has a cost that becomes clear in this case. ) Which doesn't bode well for any HPC application... but also for Databases (which will effectively be locked to 8MB per thread, with "poor sharing", at least compared to Xeon).

Of course, "Databases" might be just the most common HPC application in the enterprise, that needs communication and coordination between threads.

Re: Intel 28-core fantasy vs. AMD 32-core reality

#122
post #23

Earlier quoted context omitted.

I recall reading that the thread ripper 2 is supposed to use more power so early motherboards with that socket my not be able to handle it. Just something to keep in mind. Regardless damn good on AMD.

2nd Gen TR will draw ~250W. X399 boards have at least 1x 4-Pin and 1x 8-Pin ATX-12V connector, so at least 225W. The rest is supplied by the ATX 24 pin connection. Some X399 boards use 2x 8-Pin ATX-12V connectors, that works out to 300W, which should be more than enough for Threadripper 2.

I'm just relaying what I read when they announced the TR2. Which seems to be copy-pasted to amongst the tech news sites.

Unfortunately they just state "... some first-generation X399 motherboards may not be able to deliver enough power..."

And not specifically which.

https://arstechnica.com/gadgets/2018/06/amd-unveils-threadri...

If their statement is false then that's on them.

Re: Intel 28-core fantasy vs. AMD 32-core reality

#123
post #98
post #91

Earlier quoted context omitted.

That would be akin to i7 8086k on 7.2GHz... just add LN2. What they presented was an extreme overclocking area with insulation (due to subambient condensation), a one HP chiller that runs on a banned gas and a 4 second benchmark (seriously aside CB being free is not a true testament of overclocking prowess). Such a demo is as pointless as it is almost as practical as daily LN2 use. I can imagine few cases where first…

> I can imagine few cases where first-to-right-the-bell performance on a single core determines if you get a specific quote in HFT but that's that. That is actually the case from what I've heard, A lot of them buy consumer chips then disable all but one core and overclock it to the max. Here is a guy from optiver talking about their process at CppCon: https://www.youtube.com/watch?v=NH1Tta7purM

What a great talk, thanks for sharing!

Re: Intel 28-core fantasy vs. AMD 32-core reality

#125

As an outsider to 'enterprise-grade' computing, I'm curious about situations where a high number of cores in a single processor would be superior to multiple processors with the same total energy draw sitting on a single motherboard? I can understand HPC applications where the high-speed interconnect on the chip would make a big difference. But in business applications where the cores are dedicated to running indepen…

> As an outsider to 'enterprise-grade' computing, I'm curious about situations where a high number of cores in a single processor would be superior to multiple processors with the same total energy draw sitting on a single motherboard? Databases are the big one I'm aware of. Intel's L3 cache is truly unified. Intel's 28-core Skylake means that the L3 of a Database is TRULY 38.5MB. When any core requests data, it goes…

>Intel's L3 cache is truly unified. Intel's 28-core Skylake means that the L3 of a Database is TRULY 38.5MB. When any core requests data, it goes into the giant distributed L3 cache that all cores can access efficiently.

This is less true now. Intel's L3 cache is still all on one piece of monolithic silicon, unlike the 4 separate caches of the 4 separate dies on a 32-core TR. But the L3 slice for each core is now physically placed right next to the core and other slices are accessed through the ringbus or in Skylake and later, the mesh. Still faster than leaving the die and using AMD's Infinity Fabric, and a lot less complicated than wiring up all the cores for direct L3 access.

https://www.anandtech.com/show/3922/intels-sandy-bridge-arch...

Re: Intel 28-core fantasy vs. AMD 32-core reality

#126

AMD did a great job with Threadripper, making high end CPUs much more affordable. It's interesting that Intel doesn't lower their prices. What's the logic behind it?

I'm of the mind that AMD's smaller cores working together is the secret sauce to their price advantage.

Intel has done an amazing done stuffing 28 cores into one piece of silicon and extracting as much performance as possible all for the low price of $10k.

AMD took their 8 core part that they are selling essentially up and down their product line... and slapped 4 of them together.

Re: Intel 28-core fantasy vs. AMD 32-core reality

#127
post #52

Earlier quoted context omitted.

Keep in mind that AMD has an inherent advantage: the Zen architecture and Glue. When AMD fabs a TR2 they have to find 4 good Ryzen dies which are fairly small and they make a lot of them since they are part of the entire lineup. Once they have 4 good dies, they get glued together. If they wanted 64 cores they'd just have to look for 8 good Ryzen dies, halving their yield compared to 32. On the Intel side they have to…

Err, that's only partially true at best. If you have many dies, the cost (in money, in chip real estate and in performance drop) of the interconnect skyrockets.

The interconnect cost also skyrockets on the cores themselves. Intel's moved to a mesh network on their newest cores: https://www.anandtech.com/show/11550/the-intel-skylakex-revi...

AMD's advantage of picking and choosing smaller parts still reigns supreme. If a Infinity Fabric component on the AMD die is defective enough to necessitate it is turned off the die is no longer able to participate in some of the more complex multi-die couplings. If the mesh component on an Intel die is defective the core(two actually because SKUs) has to be fused off and the part automatically bins as a lower SKU. As the mesh increases in complexity more and more of the design can be compromised.

Re: Intel 28-core fantasy vs. AMD 32-core reality

#128
post #31

Earlier quoted context omitted.

It is already the case with Thread Ripper processors. They have multiple NUMA nodes inside one socket.

It actually presents itself to the system as a single node: On a TR 1920x system: $ numactl --hardware available: 1 nodes (0) node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 node 0 size: 32107 MB node 0 free: 20738 MB node distances: node 0 0: 10

This is from a 4 socket Xeon E7-4860 with 64 ram slots(16 in use)

  e7-4860:~ Mon Jun 11
  03:06 PM william$ numactl --hardware
  available: 4 nodes (0-3)
  node 0 cpus: 0 1 2 3 4 5 6 7 8 9 40 41 42 43 44 45 46 47 48 49
  node 0 size: 16035 MB
  node 0 free: 1306 MB
  node 1 cpus: 10 11 12 13 14 15 16 17 18 19 50 51 52 53 54 55 56 57 58 59
  node 1 size: 16125 MB
  node 1 free: 3237 MB
  node 2 cpus: 20 21 22 23 24 25 26 27 28 29 60 61 62 63 64 65 66 67 68 69
  node 2 size: 16125 MB
  node 2 free: 11004 MB
  node 3 cpus: 30 31 32 33 34 35 36 37 38 39 70 71 72 73 74 75 76 77 78 79
  node 3 size: 16123 MB
  node 3 free: 12044 MB
  node distances:
  node   0   1   2   3 
    0:  10  20  20  20 
    1:  20  10  20  20 
    2:  20  20  10  20 
    3:  20  20  20  10
The chart at the bottom of the output is the weight for accessing a memory pool from a CPU socket. This is the most important part of the output.

On this server, CPU socket 0 is hardwired to ram slots 0-15

CPU 1 to ram slots 16-31

CPU 2 to ram slots 32-47

CPU 3 to ram slots 48-63

If CPU 0 wanted to read something outside of its local ram slots, it would have execute something on CPU n, then copy that segment to its local ram group.

Re: Intel 28-core fantasy vs. AMD 32-core reality

#129
post #113

This is a short term loss for Intel, but could end up being a long term win as an attack on AMD. Making this announcement forced AMD to advance their plans for the 32-core, possibly faster than they really wanted to right now. That depletes their product pipeline faster, making it more difficult to keep pace with future advances. Edit: initial reports said that AMD was only planning to announce the 24-core CPU, and m…

They maxed out the number of cores they can ship in a single CPU for now but that doesn't seem like a problem.

AMD will already launch their 7nm EPYC processor based on Zen 2 in 2019 (skipping Zen+ used by the new Threadripper and Ryzen 2xxx) which is expected to have 48 cores (some rumors even suggest 64 cores but that seems more likely for 7nm EUV instead of the first 7nm processes). So they will have no problem releasing more cores with Threadripper 3 next year (if they keep the yearly releases).

On top of that, in my layman eyes AMDs aproach of using infinity fabric to connect dies seems better suited to react to changes compared to Intels monolithic design.

Re: Intel 28-core fantasy vs. AMD 32-core reality

#130

I'm excited for what this will do to the cost of dedicated servers in ~1 year. Also, as a person who used to work at Intel, I don't know whose idea this was, but that person should probably have a long hard look at themselves -- hardware people are exactly the people that this kind of shit wouldn't fly with, because they'll almost always ask for details and can spot a hack from a mile away. On the one hand I can symp…

> I'm excited for what this will do to the cost of dedicated servers in ~1 year. This is the opposite though? The dedicated servers are turning into HEDTs. AMD 32-core EPYC has been available since last year, and Intel's 28-core Skylake (although $10,000) has been also available for a year. So dedicated servers got this tech first, then HEDT got it a bit later. I guess Threadripper is Zen+ so its technically HEDT get…

The problem IMO is that Intel HEDTs don't support ECC (as far as I know), so not very good idea when you are working with workloads that need 64GB - 128GB of RAM (video editing, etc)
Post reply on HN