AMD's Future Chips & SoC's: News, Info & Rumours.

[juanrga]

1)
5989.9 / 24000 = 0.25 points per dollar
974.33 / 8000 = 0.12 points per dollar

The gap is bigger when one adds the acquisition cost of the rest of the platform: mobo, memory, disks, cooler...

2)
The system is not overclocked. It is clearly reported working at its stock 2.5GHz clock. The 3.8GHz is the one-core turbo, not an overclock.

http://ranker.sisoftware.net/show_run.php?q=c2ffcfe988e9d4ecddeedce5d7f183be8ea8cda895a583f0cdfd&l=en

3)
That is measuring total throughput. I.e. we are comparing 28C vs 32C. Which implies the performance gap is much higher when we compare one-core to one-core. This means in latency sensitive workloads or in latency phases of throughput workloads the Xeon chip will be much faster.

4)
Acquisition cost is only one aspect of the purchase. The real costs are the working costs. Xeon is less expensive to maintain because it is more efficient.

For all those reasons customers are choosing Skylake Xeon. No one want EPYC. Google for instance is using Xeon

https://www.overclock3d.net/news/cpu_mainboard/google_confirms_that_they_are_using_skylake_xeon_cpus_which_support_avx-512/1

 

[goldstone77]

1)
5989.9 / 24000 = 0.25 points per dollar
974.33 / 8000 = 0.12 points per dollar

The gap is bigger when one adds the acquisition cost of the rest of the platform: mobo, memory, disks, cooler...

2)
The system is not overclocked. It is clearly reported working at its stock 2.5GHz clock. The 3.8GHz is the one-core turbo, not an overclock.

http://ranker.sisoftware.net/show_run.php?q=c2ffcfe988e9d4ecddeedce5d7f183be8ea8cda895a583f0cdfd&l=en

3)
That is measuring total throughput. I.e. we are comparing 28C vs 32C. Which implies the performance gap is much higher when we compare one-core to one-core. This means in latency sensitive workloads or in latency phases of throughput workloads the Xeon chip will be much faster.

4)
Acquisition cost is only one aspect of the purchase. The real costs are the working costs. Xeon is less expensive to maintain because it is more efficient.

For all those reasons customers are choosing Skylake Xeon. No one want EPYC. Google for instance is using Xeon

https://www.overclock3d.net/news/cpu_mainboard/google_confirms_that_they_are_using_skylake_xeon_cpus_which_support_avx-512/1

You can apply that same math and logic to Intel broadwell chips that Epyc out performs. Epyc wasn't design to compete with intels top end server market. And still it's a dual $24,000 dollar setup vs an $8k setup! Why are trying to make that comparison? It's just silly...

"Given EPYC's advertised performance, one area where AMD hopes to see initial success is in the entry-level server market. Aylor said that single-socket and dual-socket servers now account for over 90 percent of the servers sold, with the vast majority being dual-socket models.

"Dual-socket servers are often the only option for performance and features," he said. "With EPYC, we have enough performance and I/O lanes to corner the bottom half of the server market with single-socket solutions. That's a TCO (total cost of ownership) advantage of over 20 percent, with no compromise in performance or features. And in this market, a 20-percent TOC advantage is very significant.""

http://www.crn.com/news/data-center/300085486/amd-looks-for-server-comeback-with-new-epyc-data-center-processor.htm/pgno/0/1

 

[juanrga]

1)
5989.9 / 24000 = 0.25 points per dollar
974.33 / 8000 = 0.12 points per dollar

The gap is bigger when one adds the acquisition cost of the rest of the platform: mobo, memory, disks, cooler...

2)
The system is not overclocked. It is clearly reported working at its stock 2.5GHz clock. The 3.8GHz is the one-core turbo, not an overclock.

http://ranker.sisoftware.net/show_run.php?q=c2ffcfe988e9d4ecddeedce5d7f183be8ea8cda895a583f0cdfd&l=en

3)
That is measuring total throughput. I.e. we are comparing 28C vs 32C. Which implies the performance gap is much higher when we compare one-core to one-core. This means in latency sensitive workloads or in latency phases of throughput workloads the Xeon chip will be much faster.

4)
Acquisition cost is only one aspect of the purchase. The real costs are the working costs. Xeon is less expensive to maintain because it is more efficient.

For all those reasons customers are choosing Skylake Xeon. No one want EPYC. Google for instance is using Xeon

https://www.overclock3d.net/news/cpu_mainboard/google_confirms_that_they_are_using_skylake_xeon_cpus_which_support_avx-512/1

You can apply that same math and logic to Intel broadwell chips that Epyc out performs. Epyc wasn't design to compete with intels top end server market. And still it's a dual $24,000 dollar setup vs an $8k setup! Why are trying to make that comparison? It's just silly...

Server market doesn't work that way. We don't compare $4000 CPUs only to $4000 CPUs and ignore the rest. This is not desktop. We can spend a given amount of money to purchase a server with 12000 CPUs (each costing $4000) or spend the same money to purchase a server with 4000 CPUs (each costing $12000). The key is how that 4000 CPUs server compares to this other 12000 CPUs server that costs the same? The metrics that matter are (i) single thread performance, (ii) total throughput, (iii) efficiency, (iv) performance/cost,... and the server with the 4000 CPU wins overall. Moreover, the server with the 4000 CPUs would have the advantage of being smaller and occupying less space, saving extra costs.

I don't follow marketing stuff posted on sites as CRN.com. I follow technical specs and what the market does; what are customers purchasing and what are rejecting? Server people doesn't purchase servers in base to marketing claims in CRN.com site. Server people test stuff before taking a buy decision. And one can easily check that the market is massively rejecting EPYC and either choosing Xeon or are moving to ARM servers (lots of companies have announced the acquisition of Cavium or Qualcomm servers).

It happens I know someone that tested EPYC and he doesn't have anything good to say about it: "Oh I can tell you from real world experience its bad, really bad."

I wonder until when all this hype about EPYC will continue and when the bubble will finally explode on contact with reality.

 

[goldstone77]

While I believe in the 4 points you made hold value I believe there are 2 more points that have a larger influence. Reliability and Trust! Those more than anything I feel are AMD's biggest hurdles. I think there is definitely portions of the market that AMD can compete with Intel, but the willingness for companies to take a risk on adopting AMD will be the biggest concern.

 

[8350rocks]

While I believe in the 4 points you made hold value I believe there are 2 more points that have a larger influence. Reliability and Trust! Those more than anything I feel are AMD's biggest hurdles. I think there is definitely portions of the market that AMD can compete with Intel, but the willingness for companies to take a risk on adopting AMD will be the biggest concern.

IT guys tend to mostly be agnostic about that sort of thing. If Perf/Watt and Perf/$ is there, they will buy it.

Consumers are the ones with tilted mindshare.

 

[-Fran-]

While I believe in the 4 points you made hold value I believe there are 2 more points that have a larger influence. Reliability and Trust! Those more than anything I feel are AMD's biggest hurdles. I think there is definitely portions of the market that AMD can compete with Intel, but the willingness for companies to take a risk on adopting AMD will be the biggest concern.

IT guys tend to mostly be agnostic about that sort of thing. If Perf/Watt and Perf/$ is there, they will buy it.

Consumers are the ones with tilted mindshare.

Plus, big companies don't "buy" nowadays. So price is a moot point at times and all it counts is: support model for hardware by OEM and particular software/capabilities bundled in it.

That being said, the leasing/renting model is still expensive as hell for Intel. If AMD can do something there, they'll still win marketshare.

Cheers!

 

[8350rocks]

While I believe in the 4 points you made hold value I believe there are 2 more points that have a larger influence. Reliability and Trust! Those more than anything I feel are AMD's biggest hurdles. I think there is definitely portions of the market that AMD can compete with Intel, but the willingness for companies to take a risk on adopting AMD will be the biggest concern.

IT guys tend to mostly be agnostic about that sort of thing. If Perf/Watt and Perf/$ is there, they will buy it.

Consumers are the ones with tilted mindshare.

Plus, big companies don't "buy" nowadays. So price is a moot point at times and all it counts is: support model for hardware by OEM and particular software/capabilities bundled in it.

That being said, the leasing/renting model is still expensive as hell for Intel. If AMD can do something there, they'll still win marketshare.

Cheers!

Precisely...I expect large gains in the very near future.

 

[jaymc]

Here's an article detailing the advantages of a single socket Epyc and Vega solution for the Data Center..

AMD Winds Up One-Two Compute Punch For Servers

Infinity Fabric, Heterogeneous compute and more PCI Express slots available for GPU's an SSD's.
Plus everything is directly linked into the CPU with the Infinity Fabric...

He draws a comparision to an Intel Xeon and Nvidia solution check it out:

"In the DGX-1, there are eight GPUs, not six shown in the diagram above, and with Volta they can all be cross-connected with NVLink; with Pascal Teslas, which has one fewer NVLink port, you have to use a mix of PCI-Express and NVLink to hook them all together. You also need a storage controller if you want to link flash SSDs to the compute complex, or burn PCI-Express slots using NVM-Express (which is not shown in the chart above). Because of the limited PCI-Express 3.0 lanes on each Xeon processor socket and the main memory cap per socket, you need a two-socket server to get the right mix of I/O and memory bandwidth and memory capacity."

"Contrast this with a single-socket hybrid Epyc-Radeon Instinct system. The single socket machine has enough PCI-Express (Infinity Fabric) bandwidth to hang 16 flash drives and 16 memory sticks off a single socket plus six Radeon Instinct GPUs. That is the same memory capacity as an HPC-configured, two-socket Xeon server. The 32 cores on the Epyc die are probably enough to drive six GPUs, just as 32 cores would be on a two-socket Xeon server using a pair of 16-core chips. But there are only enough lanes in the Epyc system to drive eight GPUs total – and that is if there is no other networking or peripherals attached. It is not clear why each GPU does not have its own directly attached flash SSD as is possible with the Radeon Instinct accelerators, and in a balanced setup, some flash drives might be dedicated to the CPU and others to the GPUs.

With the lower cost AMD CPU and GPUs, plus the integrated chipsets and the fact that there are fewer sockets, it is hard to imagine the AMD box not having the same or better bang for the buck. (We think that it will be a lot lower, especially if Intel jacks up prices on Skylake Xeons as we expect it to.) Even if hybrid AMD systems only deliver 75 percent of the performance, but do so at half the cost, the company can very likely sell the components in such machines at reasonably high margins and drive both its top line and bottom line. This, we think, is the AMD strategy, and it is one that is designed to get AMD market share over the longer haul.

The trick will be to sustain the gap and shoot through it, and that depends on what system makers, who don’t want to annoy Intel, do and when and how they do it."

https://www.nextplatform.com/2017/06/19/amd-winds-one-two-compute-punch-servers/

 

[-Fran-]

So far, according to reviews and presented line ups alike, the only thing that could make AMD fail this time around, would be Intel behaving like jerks once again and put pressure on OEMs instead of putting better products.

AMD is competitive in the layout and structure they have presented. Hell, they're even *very* attractive to any big system builder out there. So, I hope they recover marketshare in server and consumer markets.

Cheers!

 

[jaymc]

AMD 32-Core “Naples” EPYC CPUs Launching Tomorrow; Here Are The Details:

http://www.mobipicker.com/amd-32-core-naples-epyc-cpus-launching-tomorrow-details/

 

[-Fran-]

AnandTech's live blog: http://www.anandtech.com/show/11562/amd-epyc-launch-event-live-blog-starts-4pm-et-

Cheers!

EDIT: Interesting bits! (taken literally)

- 32-core single socket at $2100
- Single socket pricing: 16-core at $750
- 24-core from $1850, 16-core from $650, 8-core from $475.

So, ThreadRipper (16C/32T) should be circa USD $750?

 

[goldstone77]

Jefferies reiterates on AMD Buy after Epyc event; shares up over 2%

Jun. 21, 2017 7:18 AM ET|By: Brandy Betz, SA News Editor
Jefferies reiterates Advanced Micro Devices (NASDAQ:AMD) at a Buy rating with $16 price target following yesterday afternoon’s launch event for new Zen MPU Epyc.

Analyst Mark Lipacis discusses the key reveals during the event including the fact that Baidu, Microsoft Azure, Dropbox, and Bloomberg all plan to put Epyc in their data centers.

Dell EMC, SuperMicro, and HP all announced Epyc-based SKUs.

Lipacis expects Epyc to capture 7% of the total server market by the end of next year.

Source: StreetInsider

Latest AMD analyst standings: 5 Buy, 5 Outperform, 17 Hold, 2 Underperform, and 2 Sell.

Median price target: $12.50.

AMD shares are up 2.06% premarket to $12.90 after closing yesterday up nearly 6%.

https://seekingalpha.com/news/3274700-jefferies-reiterates-amd-buy-epyc-event-shares-2-percent?dr=1#email_link

 

[gamerk316]

I'll say it again: I wouldn't touch AMD stock at this point in time. They had their run up in price, it's time to get out. On a long-term basis, the stock isn't going to move much, and thus isn't worth holding on to.

 

[goldstone77]

I'll say it again: I wouldn't touch AMD stock at this point in time. They had their run up in price, it's time to get out. On a long-term basis, the stock isn't going to move much, and thus isn't worth holding on to.

I never hold onto a stock. The longer you hold a stock the more you increase your risk. And trading stocks is all about risk management. I buy and sell same day. Every once in a while I will hold for a day or 2 depending on the stock.

 

[jaymc]

Canaccord Genuity analyst Matthew Ramsay raised its price target on AMD (NASDAQ: AMD) to $20.00 (from $17.00) while maintaining a Buy rating.

https://www.streetinsider.com/Analyst+Comments/UPDATE%3A+AMD+%28AMD%29+PT+Raised+to+%2420+at+Canaccord+Genuity/13030903.html

13.62 @ the mo... up 7.75 percent just this morning.. it's on a tear !!

 

[gamerk316]

I'll say it again: I wouldn't touch AMD stock at this point in time. They had their run up in price, it's time to get out. On a long-term basis, the stock isn't going to move much, and thus isn't worth holding on to.

I never hold onto a stock. The longer you hold a stock the more you increase your risk. And trading stocks is all about risk management. I buy and sell same day. Every once in a while I will hold for a day or 2 depending on the stock.

I buy companies that are undervalued, then wait for the stock to recover, then sell. Make five figures off AMD, for example.

 

[goldstone77]

I'll say it again: I wouldn't touch AMD stock at this point in time. They had their run up in price, it's time to get out. On a long-term basis, the stock isn't going to move much, and thus isn't worth holding on to.

I never hold onto a stock. The longer you hold a stock the more you increase your risk. And trading stocks is all about risk management. I buy and sell same day. Every once in a while I will hold for a day or 2 depending on the stock.

I buy companies that are undervalued, then wait for the stock to recover, then sell. Make five figures off AMD, for example.

I do the same. Last one was a stock that was worth over $100 per share went bankrupt, and I bought it for penny's. 3.6 million float and it took off like a comet!

 

[-Fran-]

http://www.anandtech.com/show/11567/tyan-announces-amd-epyc-tn70ab8026-server-1p-16-dimms-26-ssds-oculink

No idea if "big news" or not, but it's another Server at least 😛

Cheers!

 

[goldstone77]

AMD EPYC 7000 Series Data Center Processor Launch - Gunning for Xeon
Author: Ryan Shrout
Date: June 20, 2017
Subject: Processors
Manufacturer: AMD
Tagged: amd, data center, EPYC, epyc 7000, Zen

"AMD is offering immediate availability of the top five CPUs in this stack, with the bottom four due before the end of July."

"AMD claims to have significant performance advantages in a 2P (two socket) configuration compared to the equitably priced Xeon product line. At the flagship level, comparing the >$4000 EPYC 7601 to the Intel Xeon E5-2699 v4, AMD expects to have 47% faster SPECint performance. Keep in mind that the E5-2699 v4 is a 22-core/44-thread processor, so that performance advantage AMD holds is a result of having 45% more cores at the same price point. Clock speeds of the Xeon part are comparable, with a 2.2 GHz base and 3.6 GHz Turbo clock."

"These solutions are useful for storage systems or GPU-centric builds that don’t need a significant amount of CPU compute power. Customers will still get access to 8-channels of DDR4 memory and 128 lanes of PCI Express, something that comparable Intel single-socket solutions will not be able to match."

"AMD is running the comparison of a single EPYC processor against a pair of Intel Xeon processors, with the obvious performance advantage going to AMD in SPECint once again."

"The software ecosystem is of critical import for the early success of AMD EPYC, and though there is clearly some additional work and partnerships to come, AMD feels confident that the collection of software partners it has on-board for the launch window is solid.
All the key operating system and hypervisor companies appear to be working with AMD to prepare the software infrastructure for use. Microsoft, Red Hat, Ubuntu, VMware, Citrix, etc. and development tools like Visual Studio and GCC have libraries and compilers in place to help give AMD the steady footing it needs to push upwards with EPYC."

"In a similar way, AMD has been working with the hardware ecosystem to prepare for today’s release, bringing many of the industry’s top-level competitors into the fold. HPE, Dell EMC, Tyan, Supermicro, and even Lenovo are building platforms for EPYC."

https://www.pcper.com/reviews/Processors/AMD-EPYC-7000-Series-Data-Center-Processor-Launch-Gunning-Xeon

Start watching at 48:36 EPYC launch discussion. PC Perspective Podcast #455 - 06/22/17 https://www.youtube.com/watch?v=ZmWDsB0B514

Great launch for EPYC!!!

 

[goldstone77]

AMD Unveils EPYC Server Processor Models And Pricing Guidelines
by Paul Alcorn June 20, 2017 at 1:00 PM
"AMD's re-entrance into the server market brings about understandable concerns about the ecosystem. At the end of the day, most will purchase systems from OEM providers, and administrators expect rock-solid support with enterprise-class applications. AMD's been hard at work on the enablement front and has amassed a solid set of launch partners for both hardware and software.
AMD has also developed a robust set of features that should further its objectives in the data center. The company noted that it designed the architecture from the ground up for data center workloads, which isn't a surprising admission. AMD's reentrance into the server market will be a long process, which company representatives have repeatedly acknowledged, but it does look promising.
Four of the high-end SKUs, along with several OEM systems, are available today. The remainder of the product stack, and further expansion of OEM's servers, comes in July."
http://www.tomshardware.com/news/amd-epyc-processor-models-pricing,34833.html

 

[juanrga]

AMD caught again providing fake scores. This time for EPYC:

AMD is making some claims about the performance of various Epyc products' performance today, and the initial outlook is good. However, we do have to take issue with a couple of the choices AMD made on the way to its numbers. After compiling SPECint_rate_base2006 with the -O2 flag in GCC, AMD says observed a 43% delta between its internal numbers for the Xeon E5-2699A v4 and public SPEC numbers for similar systems produced using binaries generated by the Intel C++ compiler. In turn, AMD applied that 43% (or 0.575x) across the board to some publicly-available SPECint_rate_base scores for several two-socket Xeon systems.

It's certainly fair to say that SPEC results produced with Intel's compiler might deserve adjustment, but my conversations with other analysts present at the Epyc event suggests that a 43% reduction is optimistic. The -O2 flag for GCC isn't the most aggressive set of optimizations available from that compiler, and SPEC binaries generated accordingly may not be fully representative of binaries compiled in the real world.

https://techreport.com/review/32125/amd-epyc-7000-series-cpus-revealed/2

And that further ignoring that AMD is comparing EPYC to old Broadwell Xeons, when fasts Skylake Xeons are available...

 

[-Fran-]

AMD caught again providing fake scores. This time for EPYC:

AMD is making some claims about the performance of various Epyc products' performance today, and the initial outlook is good. However, we do have to take issue with a couple of the choices AMD made on the way to its numbers. After compiling SPECint_rate_base2006 with the -O2 flag in GCC, AMD says observed a 43% delta between its internal numbers for the Xeon E5-2699A v4 and public SPEC numbers for similar systems produced using binaries generated by the Intel C++ compiler. In turn, AMD applied that 43% (or 0.575x) across the board to some publicly-available SPECint_rate_base scores for several two-socket Xeon systems.

It's certainly fair to say that SPEC results produced with Intel's compiler might deserve adjustment, but my conversations with other analysts present at the Epyc event suggests that a 43% reduction is optimistic. The -O2 flag for GCC isn't the most aggressive set of optimizations available from that compiler, and SPEC binaries generated accordingly may not be fully representative of binaries compiled in the real world.

https://techreport.com/review/32125/amd-epyc-7000-series-cpus-revealed/2

And that further ignoring that AMD is comparing EPYC to old Broadwell Xeons, when fasts Skylake Xeons are available...

You do realize they benchmark their targets way way before Intel released their new stuff, right?

Cheers!

 

[8350rocks]

AMD caught again providing fake scores. This time for EPYC:

AMD is making some claims about the performance of various Epyc products' performance today, and the initial outlook is good. However, we do have to take issue with a couple of the choices AMD made on the way to its numbers. After compiling SPECint_rate_base2006 with the -O2 flag in GCC, AMD says observed a 43% delta between its internal numbers for the Xeon E5-2699A v4 and public SPEC numbers for similar systems produced using binaries generated by the Intel C++ compiler. In turn, AMD applied that 43% (or 0.575x) across the board to some publicly-available SPECint_rate_base scores for several two-socket Xeon systems.

It's certainly fair to say that SPEC results produced with Intel's compiler might deserve adjustment, but my conversations with other analysts present at the Epyc event suggests that a 43% reduction is optimistic. The -O2 flag for GCC isn't the most aggressive set of optimizations available from that compiler, and SPEC binaries generated accordingly may not be fully representative of binaries compiled in the real world.

https://techreport.com/review/32125/amd-epyc-7000-series-cpus-revealed/2

And that further ignoring that AMD is comparing EPYC to old Broadwell Xeons, when fasts Skylake Xeons are available...

You do realize that AMD used GCC for their benchmarks, and SPEC uses ICC, right?

There is not a dispute that ICC *still* favors Intel, and SPEC will not allow AMD to publish GCC figures, or figures without libquantum, or they will sue.

So, in this case, AMD has to adjust for that somehow. It is not disputed that an adjustment is warranted, the only dispute is if the % is accurate.

 

[juanrga]

Spoiler

AMD caught again providing fake scores. This time for EPYC:

AMD is making some claims about the performance of various Epyc products' performance today, and the initial outlook is good. However, we do have to take issue with a couple of the choices AMD made on the way to its numbers. After compiling SPECint_rate_base2006 with the -O2 flag in GCC, AMD says observed a 43% delta between its internal numbers for the Xeon E5-2699A v4 and public SPEC numbers for similar systems produced using binaries generated by the Intel C++ compiler. In turn, AMD applied that 43% (or 0.575x) across the board to some publicly-available SPECint_rate_base scores for several two-socket Xeon systems.

It's certainly fair to say that SPEC results produced with Intel's compiler might deserve adjustment, but my conversations with other analysts present at the Epyc event suggests that a 43% reduction is optimistic. The -O2 flag for GCC isn't the most aggressive set of optimizations available from that compiler, and SPEC binaries generated accordingly may not be fully representative of binaries compiled in the real world.

https://techreport.com/review/32125/amd-epyc-7000-series-cpus-revealed/2

And that further ignoring that AMD is comparing EPYC to old Broadwell Xeons, when fasts Skylake Xeons are available...

You do realize they benchmark their targets way way before Intel released their new stuff, right?

Broadwell Xeons have been available in the market during years. AMD could have provided real-world scores for Xeon systems instead providing fake scores in marketing slides. Now you can find people in forums confused because the 47% claimed by AMD doesn't correspond to real-world measurements

http://semiaccurate.com/forums/showpost.php?p=291608&postcount=1

Spoiler

AMD caught again providing fake scores. This time for EPYC:

AMD is making some claims about the performance of various Epyc products' performance today, and the initial outlook is good. However, we do have to take issue with a couple of the choices AMD made on the way to its numbers. After compiling SPECint_rate_base2006 with the -O2 flag in GCC, AMD says observed a 43% delta between its internal numbers for the Xeon E5-2699A v4 and public SPEC numbers for similar systems produced using binaries generated by the Intel C++ compiler. In turn, AMD applied that 43% (or 0.575x) across the board to some publicly-available SPECint_rate_base scores for several two-socket Xeon systems.

It's certainly fair to say that SPEC results produced with Intel's compiler might deserve adjustment, but my conversations with other analysts present at the Epyc event suggests that a 43% reduction is optimistic. The -O2 flag for GCC isn't the most aggressive set of optimizations available from that compiler, and SPEC binaries generated accordingly may not be fully representative of binaries compiled in the real world.

https://techreport.com/review/32125/amd-epyc-7000-series-cpus-revealed/2

And that further ignoring that AMD is comparing EPYC to old Broadwell Xeons, when fasts Skylake Xeons are available...

You do realize that AMD used GCC for their benchmarks, and SPEC uses ICC, right?

There is not a dispute that ICC *still* favors Intel, and SPEC will not allow AMD to publish GCC figures, or figures without libquantum, or they will sue.

So, in this case, AMD has to adjust for that somehow. It is not disputed that an adjustment is warranted, the only dispute is if the % is accurate.

It is not true that "SPEC uses ICC". Any compiler can be used, and one can find GCC or any other compiler used in submissions in SPEC database. Next a Xeon submissions using GCC as compiler

https://www.spec.org/cpu2006/results/res2009q2/cpu2006-20090607-07605.html

The reason why most submissions use ICC is because it is almost guaranteed to provide the highest scores due to extracting more performance from the hardware.

The issue is not on what compiler was used for the scores. The problem is that the GCC scores that AMD is assigning to Xeon chips are incorrect and don't correspond to real-world measurements on Xeon chips using GCC. A real-world Xeon system using GCC scores higher than what AMD pretends in marketing slides.

 

[-Fran-]

Broadwell Xeons have been available in the market during years. AMD could have provided real-world scores for Xeon systems instead providing fake scores in marketing slides. Now you can find people in forums confused because the 47% claimed by AMD doesn't correspond to real-world measurements

http://semiaccurate.com/forums/showpost.php?p=291608&postcount=1

You missed the point. It's fine.

It is not true that "SPEC uses ICC". Any compiler can be used, and one can find GCC or any other compiler used in submissions in SPEC database. Next a Xeon submissions using GCC as compiler

https://www.spec.org/cpu2006/results/res2009q2/cpu2006-20090607-07605.html

The reason why most submissions use ICC is because it is almost guaranteed to provide the highest scores due to extracting more performance from the hardware.

The issue is not on what compiler was used for the scores. The problem is that the GCC scores that AMD is assigning to Xeon chips are incorrect and don't correspond to real-world measurements on Xeon chips using GCC. A real-world Xeon system using GCC scores higher than what AMD pretends in marketing slides.

I'm going to use your same quoted statement:

AMD is making some claims about the performance of various Epyc products' performance today, and the initial outlook is good. However, we do have to take issue with a couple of the choices AMD made on the way to its numbers. After compiling SPECint_rate_base2006 with the -O2 flag in GCC, AMD says observed a 43% delta between its internal numbers for the Xeon E5-2699A v4 and public SPEC numbers for similar systems produced using binaries generated by the Intel C++ compiler. In turn, AMD applied that 43% (or 0.575x) across the board to some publicly-available SPECint_rate_base scores for several two-socket Xeon systems.

It's certainly fair to say that SPEC results produced with Intel's compiler might deserve adjustment, but my conversations with other analysts present at the Epyc event suggests that a 43% reduction is optimistic. The -O2 flag for GCC isn't the most aggressive set of optimizations available from that compiler, and SPEC binaries generated accordingly may not be fully representative of binaries compiled in the real world.

Look at the bolded part. Read that again. Slowly.

Also, GCC discourages to run the compiler with O3, so O2 is the safe optimization level for any server compiled workload using GCC. Mentioning that it supports O3 or even specific uArch extensions is tricky, to say the least. If you want your binary to be CPU bound, you can do that easily playing with the uArch flags, that is what ICC does anyway, but automatically for Intel CPUs. Hence, the "correction" AMD is applying to the results. I won't say they did not cherry pick, but that is expected out of any company.

Cheers!