The Ryzen "ThreadRipper"... 16 cores of awesome

[Zucker2k]

It's all about a level of perception. To you and expensive water cooling setup could mean a full out three radiator loop all put together by a professional. To someone else a $70 AIO is an expensive cooler than the simple $30 212evo that works fine on the competitors CPU.

Nice strawman! Are we talking about a $1,000 cpu here? Are we talking about HEDT? Goodluck running Thread Ripper with a $30 air cooler.

 

[IEC]

True. I'm interested in seeing that test too. But that problem only raises its head when you start pushing clocks past 4.5Ghz, from all the reviews I read. Never stock.

IMO temperatures above 80°C at stock settings using an excellent AIO cooler *is* a problem.

Spoiler

Source: Guru3D http://www.guru3d.com/articles_pages/intel_core_i9_7900x_processor_review,5.html

 

[lolfail9001]

IMO temperatures above 80°C at stock settings using an excellent AIO cooler *is* a problem.

It's overclocked though, MCE [that Guru3D clearly have on as you see] is an overclock, after all. As for whether temperature is a problem or not, none of us know.

 

[IEC]

It's overclocked though, MCE [that Guru3D clearly have on as you see] is an overclock, after all.

Average package power was right in line with the quoted 140W TDP @ 136.271W. This is using "default" settings as Guru3D clearly states.

 

[Topweasel]

Nice strawman! Are we talking about a $1,000 cpu here? Are we talking about HEDT? Goodluck running Thread Ripper with a $30 air cooler.

Hey I am not the person that started the discussion. I was just saying the actual recommendation from the manufacturer is a water cooler. The rest is to explain the perception.

Me personally am running an $80 H80 on my $300 1700. But you can't always push your personal perception on everyone else. $100 cooler whether or not is a small portion of the CPU purchase price is still an expensive cooler to some people.

 

[scannall]

That's a blatant lie.

Actually, it isn't. From the Tom's review:

"As mentioned, we had to use Alphacool's Eiszeit Chiller 2000 to achieve usable overclocking results. More conventional thermal solutions just wouldn't cut it. All-in-ones like Corsair's H100i and Enermax's LiqTech 240 hit their limits at stock frequencies under Prime95. The custom loop threw in the towel at 4.6 GHz."

 

[Zucker2k]

Average package power was right in line with the quoted 140W TDP @ 136.271W. This is using "default" settings as Guru3D clearly states.

They also said:
"Overall stress/load temperatures are very high with temps at the ~80 C marker. These, of course, are default results and not tweaked. The processor idles at roughly 40 Degrees C. We used a fairly dated but good Corsair H110 here for cooling. Sidenote, this was with a Beta BIOS, we do hope to see some improvements with a firmware update."

 

[scannall]

They also said:
"Overall stress/load temperatures are very high with temps at the ~80 C marker. These, of course, are default results and not tweaked. The processor idles at roughly 40 Degrees C. We used a fairly dated but good Corsair H110 here for cooling. Sidenote, this was with a Beta BIOS, we do hope to see some improvements with a firmware update."

I do hope that future bios will improve their product. That would be a good thing. Though I don't see how a bios change would affect full load temps. Might help idle, and light load temps, which is good. But, full load is full load.

 

[Topweasel]

They also said:
"Overall stress/load temperatures are very high with temps at the ~80 C marker. These, of course, are default results and not tweaked. The processor idles at roughly 40 Degrees C. We used a fairly dated but good Corsair H110 here for cooling. Sidenote, this was with a Beta BIOS, we do hope to see some improvements with a firmware update."

Isn't that a strawman? I mean that was a tested result vs. hopeful future possibilities.

 

[IEC]

They also said:
"Overall stress/load temperatures are very high with temps at the ~80 C marker. These, of course, are default results and not tweaked. The processor idles at roughly 40 Degrees C. We used a fairly dated but good Corsair H110 here for cooling. Sidenote, this was with a Beta BIOS, we do hope to see some improvements with a firmware update."

Firmware updates will not address the core problem, which as mentioned previously is that heat from the die cannot be transferred away quickly enough to prevent excessively high temperatures.

I am cautiously optimistic that Threadripper will not suffer from this problem, as it is simply 2x Zeppelin dies and will also be soldered. Which, as we've seen with the desktop Ryzen processors, is a near-optimal heat transfer to the IHS.

 

[guskline]

The problem with the 7900K is not the total heat dissipation, it is the poor thermal transfer from the die --> TIM --> IHS.This pretty much necessitates AIO cooling at a minimum, if not custom loop.

No one has taken me up on my request to run BOINC @ 100% all cores for extended periods of time to see what the maximum clocks attainable under that kind of usage are. I have a feeling the poor thermal transfer will make max OCs for 100% load 24/7/365 lower than expected - or cause longevity issues due to prolonged operation above 90C.

Which is why I am waiting to see what Threadripper brings in that regard.

Nailed it IEC.

BTW, my Ryzen 1800x @4ghz (100x40) is under custom water (2 360 slim rads and an EK AM4 EVO block) but maxes out with stability at 40.25 x 100 which is not bad considering stock is 3.6 all cores and 4 on only 1 or 2 cores. There may be VERY few Ryzen 1800x chips stable at 4.1.

I think the biggest TR (16 core 32 thread ) will be clocked at 3.4/3.8 or at most 3.6/4.0. To get all 16 cores stable at 4 will be quite a challenge and will require a HUGE custom heatsink with plenty of cooling. However, it is likely to be much less expensive than the 18 core 36 thread Skylake X to be released later this year.

Performance wise I expect the Skylake X to be faster but not by much. Intel has been so dominant they quit looking in the rear view mirror, especially in the Bulldozer/Piledriver days. That's over now. Intel is now glued to the rear view mirror and AMD is coming on fast. When you factor in the cpu cost it's neck and neck.

 

[Zucker2k]

Actually, it isn't. From the Tom's review:

"As mentioned, we had to use Alphacool's Eiszeit Chiller 2000 to achieve usable overclocking results. More conventional thermal solutions just wouldn't cut it. All-in-ones like Corsair's H100i and Enermax's LiqTech 240 hit their limits at stock frequencies under Prime95. The custom loop threw in the towel at 4.6 GHz."

That says, these coolers couldn't cut it for overclocking but did keep the chip stable at stock clocks while priming, if barely. Also, this is probably the worst case scenario you could find. Others got better results; higher overclocks and less capable coolers to achieve them, so lesson here is, YMMV.

 

[guskline]

Firmware updates will not address the core problem, which as mentioned previously is that heat from the die cannot be transferred away quickly enough to prevent excessively high temperatures.

I am cautiously optimistic that Threadripper will not suffer from this problem, as it is simply 2x Zeppelin dies and will also be soldered. Which, as we've seen with the desktop Ryzen processors, is a near-optimal heat transfer to the IHS.

Also agree IEC, Plenty has been written about Intels decision to NOT use solder but rather paste on it's SkylakeX chips.

 

[ub4ty]

Quick Question : Is it known whether or not Threadripper (16/32) has 4 CCX complexes w/ 4 cores disabled on each ccx or is it two CCX complexes with no cores disabled rendering (16/32)?

I'm curious to know whether the top of the line Thread-ripper w/ 16/32 has gimped cores or if it's a clean two ccx complex. I hope they aren't using core gimping in the top of the line Threadripper. That'd be kind of lame.

 

[scannall]

Quick Question : Is it known whether or not Threadripper (16/32) has 4 CCX complexes w/ 4 cores disabled on each ccx or is it two CCX complexes with no cores disabled rendering (16/32)?

I'm curious to know whether the top of the line Thread-ripper w/ 16/32 has gimped cores or if it's a clean two ccx complex. I hope they aren't using core gimping in the top of the line Threadripper. That'd be kind of lame.

Just two dies. No gimped 4 die Threadrippers.

 

[Zucker2k]

Firmware updates will not address the core problem, which as mentioned previously is that heat from the die cannot be transferred away quickly enough to prevent excessively high temperatures.

I am cautiously optimistic that Threadripper will not suffer from this problem, as it is simply 2x Zeppelin dies and will also be soldered. Which, as we've seen with the desktop Ryzen processors, is a near-optimal heat transfer to the IHS.

I do understand the TIM problem, but it's not as bad as some are advertising, at stock. Also, better power management and delivery is also important to keeping things cool so it's important, as the review indicated. You are one of the few enthusiasts who adopted Ryzen right from go. You should know. As for the "core problem," it only becomes a problem if you push the chip. 80c is nothing. My bare-die skylake gaming laptop consistently hits the high 90s when I do anything remotely stressful yet it keeps chugging along; 2 years and counting.

 

[ub4ty]

Just two dies. No gimped 4 die Threadrippers.

Thank you so much for the quick reply. I'm so excited . I have a juicy workload waiting for it.

 

[formulav8]

I've never had cause to warranty a CPU, so that's a non-issue as far as I am concerned.

As far as you're concerns are, I completely agree.

 

[lolfail9001]

Average package power was right in line with the quoted 140W TDP @ 136.271W. This is using "default" settings as Guru3D clearly states.

1. Average package power is irrelevant on desktop, Intel's TDP for HEDT is specified for "maximum" *stock* package power.
2. CPU-Z nicely reports 4.3Ghz and hwinfo reports load on all 10 cores. The only way this happens? Multi core enhancement. And since i have seen too many reviews that demonstrate reviewer forgetting about it's existence entirely, that CPU-Z is damning evidence.

That said, thermal transfer could certainly be better, though i doubt it would prevent thermal throttling upon hitting that fabolous 1W per mm^2 mark that worked flawlessly with 6950X overclocks.

 

[coercitiv]

While I agree that Threadripper will be priced competitively, remember that it's rumored to be clocked much higher than EPYC. The $849 rumor price that B&C is probably pretty close to reality. I'm expecting something closer to $999.

An isn't $850 the rumored price? Was there any other noteworthy rumor that put the top 16c/18t part at $600-700?!

EPYC CPUs may be clocked lower, but they also have more dies & connectivity which takes up room in the TDP budget. They're all clocked for efficiency too, which means low core count SKUs may be more conservative with clocks than needed: if you compare 1P SKUs you may notice the 16c and 24c SKUs will have very similar Turbo clocks, which can only happen if the 16C unit still has some headroom left until maxing TDP.

If we look at EPYC launch price indicators, TR 16c kinda needs to cost more than $700 (price limit for EPYC 16c) and less than $1100 (price limit for EPYC 24c), which marks both $800 and $1000 on the price map for possible SKUs.

 

[bononos]

........
2. CPU-Z nicely reports 4.3Ghz and hwinfo reports load on all 10 cores. The only way this happens? Multi core enhancement. And since i have seen too many reviews that demonstrate reviewer forgetting about it's existence entirely, that CPU-Z is damning evidence.
.....

3dguru have always tested their cpus at their default settings which allows turbo speeds not at base clocks. 4.3Ghz is the turbo speed for the i9-7900x not some overclocked speed. Remember 4.5Ghz is the turbo boost 3.0 speed which wasn't shown to be tested and is probably only stable under more exotic and expensive cooling beyond the capabilities of the h110.

 

[Rngwn]

If we look at EPYC launch price indicators, TR 16c kinda needs to cost more than $700 (price limit for EPYC 16c) and less than $1100 (price limit for EPYC 24c), which marks both $800 and $1000 on the price map for possible SKUs.

The minimum may not be necessarily true though, but nonetheless very likely. Remember that the 16c EPYC still have twice the PCIE lanes and DRAM channels while still having the same core count. That said, the 16c TR is made of the full-fat ccx and will have higher clock speed. So, the pricing will likely be comparable to the EPYC counterpart.

 

[Wyrm]

I suspect some people may be setting themselves up for disappointment if they expect cool thermals from the 16 core TR. Linus tech tips had TR at 180W. Even if comes out with lower 160W it will still be about as hot as 7900X. There is no free lunch: more cores, more interconnects and higher bandwidth come with the power price attached to it. I hope people have reasonable expectations.

 

[coercitiv]

The minimum may not be necessarily true though, but nonetheless very likely. Remember that the 16c EPYC still have twice the PCIE lanes and DRAM channels while still having the same core count. That said, the 16c TR is made of the full-fat ccx and will have higher clock speed. So, the pricing will likely be comparable to the EPYC counterpart.

The minimum is also influenced by R7 1700 price/features. TR doubles on many fronts, and in this high performance area price does not linearly increase with performance (and features).

I suspect some people may be setting themselves up for disappointment if they expect cool thermals from the 16 core TR. Linus tech tips had TR at 180W. Even if comes out with lower 160W it will still be about as hot as 7900X. There is no free lunch: more cores, more interconnects and higher bandwidth come with the power price attached to it. I hope people have reasonable expectations.

It's not the power consumption that directly induces the temps on 7900X, it's the TIM. If it weren't for the thermal interface problem, even beefy air coolers could have handled SKL-X at stock and maybe a bit beyond.

 

[Zucker2k]

3dguru have always tested their cpus at their default settings which allows turbo speeds not at base clocks. 4.3Ghz is the turbo speed for the i9-7900x not some overclocked speed. Remember 4.5Ghz is the turbo boost 3.0 speed which wasn't shown to be tested and is probably only stable under more exotic and expensive cooling beyond the capabilities of the h110.

Two cores running at 4.5ghz cannot produce more heat than 10 cores running at 4.3ghz.