AMD has Intel in full panic mode. :D

TennesseeTony

Elite Member
Aug 2, 2003
4,209
3,634
136
www.google.com
A quick look at the top stories on Anandtech.com today, a trio of articles regarding Intel's newest and best chipsets and CPU's, looks to me like panic from Intel. But greedy panic.

http://www.anandtech.com/show/11461...new-highend-desktop-platform-and-x299-chipset

http://www.anandtech.com/show/11463...-desktop-getting-the-latest-microarchitecture

http://www.anandtech.com/show/11464...ng-18core-hcc-silicon-to-consumers-for-1999/2

Intel still thinks it can price it's best consumer chips however it wants, and still thinks it can "differentiate product lines" with artificial limits, such as PCIe lane limitations. (seriously, some are limited to 16 lanes!?!?!)


On the other hand:
Email from AMD: ".....AMD’s EPYC CPU will also redefine expectations for how features are supported across the price/performance range. At AMD, we believe all our customers deserve the full range of capabilities of their systems. Thus, all EPYC processors will be unrestrained by artificial limits on features or capabilities. Every EPYC processor will support eight channels of DRAM, a full 128 lanes of I/O, and all our security and reliability features...."

Intel sure seems to have the clockspeed advantage though, to be fair.

Exciting times! :D Will this (ThreadRipper and EPYC) be as monumental as the core series being introduced by Intel?

Disclaimer: "all natural liquid muscle relaxer (made from potatoes) has been consumed by the author. ;)" It's for my back. Really.
 

StefanR5R

Elite Member
Dec 10, 2016
5,516
7,823
136
A quick look at the top stories on Anandtech.com today, a trio of articles regarding Intel's newest and best chipsets and CPU's, looks to me like panic from Intel.
Especially the announcement of additional SKUs with 14, 16, and 18 cores, but without base and turbo clocks nor release schedule known for the 12-18 core SKUs, very much look like an emergency plan which they came up with at the last minute.

But greedy panic. [...]
Intel still thinks it can price it's best consumer chips however it wants, and still thinks it can "differentiate product lines" with artificial limits, such as PCIe lane limitations. (seriously, some are limited to 16 lanes!?!?!)
All SKUs below the $999 price point being cut down to 28 PCIe lanes surprised me actually.

Of course for Intel there is the risk of losing some of their Xeon sales to HEDT sales, which is why they must be creative inventing all sorts of limitations for the HEDT chips and boards, to make them unattractive to more general use outside oversized gaming rigs.

The two Kaby Lake-X SKUs are totally weird, I agree. Laptop chips packaged for socket 2066? One of them even without hyperthreading? This must have been the bean counters' idea of the entry level offers on the socket 2066 platform. It could also be early signs of a long-term plan to remove overclocking from the mainstream socket. (Such a plan might not exist; or it might have existed but may no longer be viable since AMD is back with overclocking options for the mainstream.)

Intel sure seems to have the clockspeed advantage though, to be fair.
Apparently they left some of that advantage on the table by changing the thermal interface between die and heat spreader from a thin Indium solder layer to a thick polymer TIM layer. After desktop Kaby Lake brought back Sandy Bridge like clocks, I was looking forward to how Skylake-X with supposedly the same transistor generation would overclock. But current rumors of bad temperatures without delidding do not sound promising.

In my main work PC, I want a reasonable core count for the rare occasions when the software is able to utilize them, I want high IPC + high clock speed for when the software does not scale with core count, and I want a comfortable amount of RAM with ECC protection. At the times when I built my Ivy Bridge-E PC and then the Broadwell-E PC, AMD's processors lacked IPC, Intel's Xeons lacked clock speed, so I crossed fingers and gave up on ECC in the work PC. The home PC with less CPU demand (before I was hooked to Distributed Computing) has got a Xeon E3 + ECC RAM, and earlier a Phenom II + ECC RAM.

Hypothetically, if I wanted to replace my work PC or home PC later this year, I already have an idea with which processor vendor I would go.

Exciting times! :D Will this (ThreadRipper and EPYC) be as monumental as the core series being introduced by Intel?
I am fascinated by the thought that a single socket Epyc box could offer the capabilities of current dual socket boxes. And a dual socket Epyc would get you what had been a quad socket server.
 

Orange Kid

Elite Member
Oct 9, 1999
4,328
2,112
146
It's going to be interesting to say the least. I'll be waiting in the wings for the dust to settle a bit before making an upgrade decision. :)
 
  • Like
Reactions: Smoke

crashtech

Lifer
Jan 4, 2013
10,524
2,111
146
I'm happy to have an 8-core Ryzen on my desktop, although it doesn't overclock well, and doesn't seem stable for DC when it is, so I've been forced to run it at stock (for now, I think a ~200MHz OC is probably fine). My hope is that a later iteration will hit higher clockspeeds and I will upgrade to that. For now, this octacore 1700X at stock has only a bit more execution throughput than the hexacore 5820K that I sold, since the 5820K ran at 4.2GHz 24/7.
 
Last edited:

StefanR5R

Elite Member
Dec 10, 2016
5,516
7,823
136
From what I read, Ryzen is very similar to Broadwell-E with respect to overclocking potential (low frequency ceiling; Ryzen apparently a bit lower even than BDW-E), and very similar to Broadwell-E/EP with respect to potential performance-per-Watt (very energy-efficient at lower clock speeds).
 

crashtech

Lifer
Jan 4, 2013
10,524
2,111
146
@StefanR5R , I surmise this to be intentional on AMD's part, efficiency is paramount in server space, more cores (denser) with better or even comparable per-core efficiency combined with better initial value would make for a lower TCO, we'll see.
 

StefanR5R

Elite Member
Dec 10, 2016
5,516
7,823
136
https://videocardz.com/70266/amd-epyc-7000-series-specs-and-performance-leaked

AMD EPYC 7000 launches June 20th
The NDA on this material ends on June 20th. Here are some buzzwords from the slides:
  • up to 32 High-Performance “Zen” Cores
  • 8 DDR4 Channels per CPU
  • Up to 2TB Memory per CPU
  • 128 PCIe Lanes
  • Dedicated Security Subsystem
  • Integrated Chipset
  • Socket-Compatible with Next Gen EPYC Processors
No compromise 1-Socket

  • Right-size underutilized servers
  • Optimize storage and heterogeneous compute
  • Low power consumption
  • Up to 20% lower cap-ex

DBH2s.jpg


  • tZ8J3.jpg

Some guesstimation:

If this holds true, then the 24-core EPYC 7401 is about as fast as the 22-core E5-2699 v4 and E5-2696 v4 (slower at some vectorized workloads perhaps, but surely faster at memory-intensive workloads) at same or slightly higher cTDP, and costs a bit more than a second-hand E5-2696 v4 from e-Bay (customs duty for Asia import of the 2696 neglected).

Hopefully we learn more about price and performance soon...
 

StefanR5R

Elite Member
Dec 10, 2016
5,516
7,823
136
A German news item from AMD's Epyc launch event states that folks from HPE and Dell indicate 10-20 % performance-per-price advantage of AMD servers over Intel servers. Meaning that Epyc won't be a bargain for DC'ers. (But maybe it brings some movement into the second-hand server market...)
 
  • Like
Reactions: TennesseeTony

TennesseeTony

Elite Member
Aug 2, 2003
4,209
3,634
136
www.google.com
Anandtech's own articles indicate the TDP is much higher than Intel, 155 watts with light loads, 180w 'normal' and up to 200 watts in Turbo. Ouch.

Looking at the 16P/32T EPYC pricing, it doesn't look like we are going to get ThreadRipper for only $1000 as hoped. Boo.
 

StefanR5R

Elite Member
Dec 10, 2016
5,516
7,823
136
Anandtech's own articles indicate the TDP is much higher than Intel, 155 watts with light loads, 180w 'normal' and up to 200 watts in Turbo.
But this is for more cores, memory channels, PCIe and cross-socket interconnects --- compared to both Broadwell-EP and Skylake-SP. And TDPs of higher-clocked Skylake-SP will apparently be similarly to Epyc's TDPs.

Actual power consumption of Epyc looks promising though:
https://www.servethehome.com/amd-epyc-7601-dual-socket-early-power-consumption-observations/

Looking at the 16P/32T EPYC pricing, it doesn't look like we are going to get ThreadRipper for only $1000 as hoped.
Why not? Threadripper's clocks will probably be higher than those of the 16C/32T Epyc SKUs, but on the other hand, Threadripper consists of only two dies per package and offers only half the RAM channels and I/O lanes. Aready the single-socket "P" Epyc SKUs will apparently be priced a notch below equivalent dual-socket capable SKUs, and Threadripper could be another notch down in price.

BTW, Epyc's socket SP3 and Threadripper's socket TR4 are not compatible despite the same form factor. Meaning that these will remain separate platforms.
--------
Edit:
From AnandTech's launch article: "AMD claimed that for some parts of the market, only one AMD processor will be needed to replace two Intel processors". Will e-bay be flooded with socket 2011-3 chips and boards by the end of the year? :grin:
 
Last edited: