Intel looks to be feeling the heat from AMD!


Intel Brings Its Own Benchmark to Refute AMD's '2X' EPYC Claim

Intel is firing back at AMD CEO Lisa Su's Computex claims that the company's upcoming 7nm EPYC Rome data center processors are twice as fast as Intel's Xeon Scalable 8280 processors in a popular benchmark...

... AMD is poised to start taking larger portions of Intel's market share. AMD's EPYC Rome processors come this quarter with up to 64 cores and 128 threads apiece, far outweighing Intel's general-purpose lineup of Xeon chips (the flagship Xeon 8280 weighs in at 28 cores and 56 threads). Not only is AMD taking the lead in core counts, it is also taking the process lead with TSMC's 7nm node, which should improve pricing and reduce power consumption...

... AMD's dual-socket server has 128 cores and 256 threads, so it isn't entirely surprising to see that much firepower beat Intel's competing 56-core 112-thread dual-socket server. ... Intel says that after it applied the correct optimizations to its system, it scored 30% higher than the results presented by AMD (below). That doesn't give Intel's Xeon 8280 the lead, but shows the chips are more competitive with the correct optimizations...

... Intel also included test results with its Xeon Platinum 9000-series (Cascade Lake-AP) that come armed with as many as 56 cores, 112 threads, and 12 memory channels crammed into a package that dissipates up to 400W. These new behemoths, which are essentially two Skylake-SP CPUs in a single socket, only come in OEM servers, so they aren't available on their own like AMD's Rome chips will be.

When Intel brings its high-powered -AP battleships to bear, it takes a 1.5% lead with the 96-core 9242 server, while the 112-core 9282 server takes a 23% lead over AMD's 128-core server...

... Intel's 8280 processors have a recommended price range of $10,000 to $17,000 apiece, depending upon the options you choose, and while AMD hasn't announced pricing for its Rome models, it's fair to assume it picked the 8280's as a comparison point based on pricing. While Intel's 9282 and 9242 may be the fastest on the market, they're likely priced significantly higher than AMD's Rome parts...



All a game of Marketing numbers. Still, interesting times after something like a decade of Intel stagnation.

From my personal engineering view: AMD looks to have the lead on multiple counts and especially on price/performance.


Happy fast crunchin',
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

Not sure what you are trying to prove or show. Are you concerned what the graph shows? That is similar to what mine shows. What do you expect to see or what shape of the graph do you expect to see?

Jsm thought it showed that "cool 'n quiet" was active. And it doesn't look like the "top" that I see on my other boxes.

Tom

What that shows is not "cool 'n quiet". What that shows is XFR2. "cool 'n quiet" in effect would show the core voltages dropping. Has nothing to do with utilization. XFR2 constantly moves the load around on the cores to equalize the hot spots on the cpu to allow for greater overall clock speeds.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Well I reckon that it shows that every 19 seconds or so the firmware slows or stops the activity for about a second. Ostensibly that would take the load off the cpu and result in a cooling cycle.
You will recall that ASUS support for my ROG EXTREME MB running BIOS 1607 said they did not have access to cool n quiet under whatever name because AMD had locked that out. (See previous posts)
I object if that is the effect we are seeing because I control the temperature through heavy water cooling and extra fans and radiator and I dont want AMD or ASUS doing so except as default.
jsm

Well I reckon that it shows that every 19 seconds or so the firmware slows or stops the activity for about a second. Ostensibly that would take the load off the cpu and result in a cooling cycle.
You will recall that ASUS support for my ROG EXTREME MB running BIOS 1607 said they did not have access to cool n quiet under whatever name because AMD had locked that out. (See previous posts)
I object if that is the effect we are seeing because I control the temperature through heavy water cooling and extra fans and radiator and I dont want AMD or ASUS doing so except as default.
jsm

But that is not what your graph shows. This is the definition of cool 'n quiet

Cool'n'Quiet. AMD Cool'n'Quiet is a CPU dynamic frequency scaling and power saving technology introduced by AMD with its Athlon 64 processor line. It works by reducing the processor's clock rate and voltage when the processor is idle.

Your processors are never idle, so the cool 'n quiet mechanism never takes place.

However XFR2 and Performance Boost IS ACTIVE unless you have disabled both in the BIOS. Even then, when those are disabled, I think there is still a internal housekeeping algorithm running on the processor to try and keep the cpu within its TDP budget.

It does that by moving cpu core loads around constantly. That would explain your graph where a cpu load trace drops off and then returns to 100% loading.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

So if that is the case does this mean that I have to accept the reduced performance every 19seconds or so so that not too much power is used?
jsm

So if that is the case does this mean that I have to accept the reduced performance every 19seconds or so so that not too much power is used?
jsm

I would say so. I would at least try and turn off any boost settings in the BIOS and no overclocks. In fact don't use any Auto setting for clock speed at all. Just set a manual clock speed of base 3Ghz frequency and lock it in there. See if you still get the core loads moving around. The AMD Ryzen and Threadripper architecture is very different than older Intel generation architectures and doesn't work the same. So no reason to expect it to behave the same as Intel.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Beaten to it.

---

Grant
Darwin NT

Ryzen 5 3600 annihilates the 2600 in leaked benchmarks.

The closest comparison for the Ryzen 5 3600 is the 2600, which is also a six-core, twelve thread chip that’s just 300 MHz slower with a 3.9 GHz boost clock. Across the four benchmarks, the 3600 outperforms it by a staggering 25%, which is higher than expected.

Intel’s i5-9400F which is equivalent performance wise to the 2600 and has recently dropped to $150, is beaten by just 5% in single-core benchmarks, making it a good value. But because it's only six-core, six thread chips, they’re beaten by a shocking 44% in the two multi-core tests, throwing it out of the ring.

---

Grant
Darwin NT

I think we will find this result the same across all Ryzen 3000 sku's across the stack compared to Ryzen+ because of the 14-25% IPC increase for this new generation compared to last.
[Edit] I'm looking forward to these new cpus because they have a full 256 bit wide FPU register compared to the Ryzen+ 128bit wide FPU registers. So they should do math a lot faster.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

I think we will find this result the same across all Ryzen 3000 sku's across the stack compared to Ryzen+ because of the 14-25% IPC increase for this new generation compared to last.
[Edit] I'm looking forward to these new cpus because they have a full 256 bit wide FPU register compared to the Ryzen+ 128bit wide FPU registers. So they should do math a lot faster.

That would be a case for upgrading from the 2700 to one of the 3000 series with the same TDP. Then the CPU crunching would be even more impressive. And if any of your slots will run gen you MIGHT even see a gpu speed bump.

Tom

---

A proud member of the OFA (Old Farts Association).

Thank you for the directions. I downloaded and installed both without any problems in advance of fitting the card. The card was a problem because it is very large and the specialist builders had effectively screened the 4 pcie slots with the water cooling plumbing. However eventually succeeded and BOINC now using it but for only one task. Is that correct? I thought it was a parallel device. True it is processing each task in 10 to 20 minutes alongside the 32 tasks from the cpu but I would have liked to have seen more parallel activity for all the effort (and cost).
jsm

Thank you for the directions. I downloaded and installed both without any problems in advance of fitting the card. The card was a problem because it is very large and the specialist builders had effectively screened the 4 pcie slots with the water cooling plumbing. However eventually succeeded and BOINC now using it but for only one task. Is that correct? I thought it was a parallel device. True it is processing each task in 10 to 20 minutes alongside the 32 tasks from the cpu but I would have liked to have seen more parallel activity for all the effort (and cost).
jsm

If we are talking about a gtx 1060 video card running Tbar's All-in-One then it can only run 1 task at a time. However, it should be taking 3-4 minutes. I am not certain you are running that version because the gpu tasks you are processing are CUDA60. Which could run 2 per card and do run much slower.

However, it sounds like you are referencing a different card from what I can see. So could you back fill with a little context? A single gtx 1060 should not be able to cover 4 pcie slots.

Tom

---

A proud member of the OFA (Old Farts Association).

I didnt mean the 1060 was taking up four pcie slots I meant that the water cooling plumbing was obstructing them :)
I am still using default Ubuntu application and have not yet got round to contemplating T Bar. It is early days so I may be looking at the wrong stats. I will sit back and watch what happens to the throughput. By the way the 2nd hand 1060 wasn't that cheap in the UK - you probably have better deals in the States.
jsm

I didn't mean the 1060 was taking up four PCI-E slots I meant that the water cooling plumbing was obstructing them :)
I am still using the default Ubuntu application and have not yet got round to contemplating T Bar. It is early days so I may be looking at the wrong stats. I will sit back and watch what happens to the throughput. By the way, the 2nd hand 1060 wasn't that cheap in the UK - you probably have better deals in the States.
jsm

I am sorry to hear "plumbing" was the issue. If my Dad, a retired plumber, was still alive I would offer to send him out :)

I will admit I am talking about used eBay gtx 1060 3GB cards. They were running as low as $130USD when I last bought some.

The Tbar All-in-One should lower the processing time on any Nvidia video card from the gtx 750Ti on up.
It is probable that once you start getting standard "SOG" tasks, your video card timing should drop to around 7-8 minutes. My own experimenting with SOG type tasks leads me to believe that this card will only run one of those at a time.

I am assuming you do NOT have an "app_info.xml" file in there someplace (it is something that is not part of the default setup).

Tom

---

A proud member of the OFA (Old Farts Association).

And if any of your slots will run gen you MIGHT even see a gpu speed bump.

AMD shot down the rumors and news that you would be able to run Gen. 4 on the nearest slot on X370 or X470. PCIE Gen. 4 is being excluded from running on the older boards. If you want to run Gen. 4 you are going to have to buy a X570 board.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

He won't get any of the normal gpu apps since he does not have any OpenCL drivers loaded which the SoG app requires. All he will be able to process gpu tasks with is the old CUDA60 app.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Well all the drivers you advised me to obtain were downloaded without any problem. Should I seek more?
jsm

If your Nvidia download of the drivers did not yield the OpenCL components which it seems has happened, you can always add them in.

sudo apt-get install ocl-icd-libopencl1

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Now downloaded. I re-ran benchmarks and re-read prefs. Is there anything else I need to do to make BOINC aware?
jsm

Now downloaded. I re-ran benchmarks and re-read prefs. Is there anything else I need to do to make BOINC aware?
jsm

I'd suggest exit BOINC, and restart it. Then check the Event log to make sure it's detected the changes.

---

Grant
Darwin NT