Core i9-7900X Skylake-X Review Shows Up

[R-T-B]

An Intel Core i9-7900X has appeared for a full review at the site Hexus.net. Spoiler alert, it clocks to 4.7 GHz on all ten cores with relative ease (only taking 1.25 V, apparently, though it racked up nearly 100°C in Cinebench at that voltage).

The review praised Intel's overclocking headroom and general muscle in a mostly positive review. Still, not all is rosy in Intel land. They found performance per watt to not have improved much if at all, criticized the high price tag, and Hexus.net had the following to say about the overall experience:

"X299 motherboards don't appear to be quite ready, there are question marks surrounding the Skylake-X processors due later this year, and at the lower end of the Core X spectrum, Kaby Lake-X is nothing short of puzzling."

It would seem AMD is not the only major chip-maker who can have motherboards ill prepared at launch time, even the mighty Intel may have teething issues yet.

You can read the full review (which is mostly positive, by the way) in the source link below.

Oh, and a special shoutout to our own @the54thvoid for discovering this article.



View at TechPowerUp Main Site

 

[ironwolf]

One of the comments on the article was pure gold:

This looks like an Emergency Edition processor

 

[xkm1948]

Just finished reading. Meh, not impressed at all. Sure it can overclock high, but that power consumption. Man, I would take a RyZen over this power hungry CPU any time. And Threadripper will probably walk all over these "emergency edition processors"

Looks like RyZen put Intel back to the good old days of GHz race. When they loose in efficiency, the clock rate goes up.

Don't know how much more IPC improvement Intel can dig out of the X86. Unless they pull another story of Core2Duo of course.


OK wait a minute. If I am reading those charts correctly, the IPC of 7900X is actually worse comparing to 6950X? Can reduced L3 Cache size have such big impact on performance??

Man Intel is failing so hard after RyZen launch.

 

[R-T-B]

OK wait a minute. If I am reading those charts correctly, the IPC of 7900X is actually worse comparing to 6950X? Can reduced L3 Cache size have such big impact on performance??

Apparently, yes.

 

[Steevo]

In B 4

"Well power consumption doesn't matter on HEDT platforms" and "If you can afford it then better cooling isn't an issue"

 

[xkm1948]

OK some quick search give me this. Yep, a HUGE shrink of L3 Cache.

So Intel is actually going backwards in IPC. Whoever made this decision in Intel needs to be fired 100 times.

"How are we gonna compete aganist RyZen"
"I dunno, maybe make our processors even worse?"
"Brilliant idea!"

Does Intel's ass control its brain these days?

In B 4

"Well power consumption doesn't matter on HEDT platforms" and "If you can afford it then better cooling isn't an issue"

Same. Really wanna see how Intel PR and fanboys are gonna spin this story.

 

[Steevo]

OK some quick search give me this. Yep, a HUGE shrink of L3 Cache.

So Intel is actually going backwards in IPC. Whoever made this decision in Intel needs to be fired 100 times.

"How are we gonna compete aganist RyZen"
"I dunno, maybe make our processors even worse?"
"Brilliant idea!"

Does Intel's ass control its brain these days?

View attachment 89146



Same. Really wanna see how Intel PR and fanboys are gonna spin this story.

I think the simple answer is, Intel doesn't want to cannibalize their server chips, and large unified cache designs take a lot of power to run even at idle and would exacerbate their power draw and thermal issues to an even larger degree.

 

[xkm1948]

I think the simple answer is, Intel doesn't want to cannibalize their server chips, and large unified cache designs take a lot of power to run even at idle and would exacerbate their power draw and thermal issues to an even larger degree.

yeah yeah whatever. Worse IPC is worse. No matter how Intel wanna spin it they failed hard.

 

[R-T-B]

yeah yeah whatever. Worse IPC is worse. No matter how Intel wanna spin it they failed hard.

To be fair, cache is one of the biggest limiting factors in clocks/overclocks because it runs hot.

Maybe Intel really is going all netburst on us again, sacrificing IPC for clocks. Dunno. They aren't there yet but they are certainly moving the wrong way IMO.

EDIT: And if clocks were the goal, WHY THE HECK DID THEY USE A TIM?!?!?!

 

[cadaveca]

To be fair, cache is one of the biggest limiting factors in clocks/overclocks because it runs hot.

Maybe Intel really is going all netburst on us again, sacrificing IPC for clocks. Dunno. They aren't there yet but they are certainly moving the wrong way IMO.

EDIT: And if clocks were the goal, WHY THE HECK DID THEY USE A TIM?!?!?!

The 7900X is not priced at the same point or higher than the 6950X, so it is NOT meant to be its replacement. That replacement is not coming for several months. That's why we already have info about a coming 18-core CPU, the 7980XE.

As to TIM, going form 3.3 GHz on 6950X to 4.0 GHz on 7900X seems like a 700 MHz increase to me... and is well within the capabilities of that TIM. Enthusiasts that want better cooling and push the chip past the 297W limit can pop the top, and void the warranty, no big deal.

 

[R-T-B]

The 7900X is not priced at the same point or higher than the 6950X, so it is NOT meant to be its replacement. That replacement is not coming for several months. That's why we already have info about a coming 18-core CPU, the 7980XE.

Fair enough. I wonder if the 7980XE will be soldered, per chance?

no big deal.

That's relative, of course. It was a big deal to me. They lost a customer over it. I'm sure I'm not the only one either.

At the same time, I'd be deeply surprised if they cared.

 

[cadaveca]

That's relative, of course. It was a big deal to me. They lost a customer over it. I'm sure I'm not the only one either.

At the same time, I'd be deeply surprised if they cared.

300W is a lot of power to push through such a small piece of glass, and the TIM is capable. It's even noted in the whitepaper that the chip can handle 300W. There's actually a whole lot about this chip that remains undisclosed, like, well, how about the FIVR? Hot temps? Perhaps the TIM has nothing to do with it?

Max rated voltage of 2.15V, 1.35V for VCCIO and VCCSA, 1.4V for vDIMM.

To me, this is far from a proper review, because such info was missed. It's hard to make accurate judgements about anything when you don't have all the information. That TIM is actually doing a hell of a job, and these two reviewers that posted these reviews today missed it.


But of course, that's why you're here.



Also note the lack of disclosure about which board they have... You cannot make accurate assessments about this platform under these circumstances.

 

[claes]

bit-tech up as well: https://www.bit-tech.net/hardware/2017/06/16/intel-core-i9-7900x-and-x299-chipset-revie/1

 

[VSG]

All I will say is we should wait for other reviews to show up, especially from those who have taken more time to test the platform rather than prematurely release reviews.

 

[R-T-B]

300W is a lot of power to push through such a small piece of glass, and the TIM is capable. It's even noted in the whitepaper that the chip can handle 300W. There's actually a whole lot about this chip that remains undisclosed, like, well, how about the FIVR? Hot temps? Perhaps the TIM has nothing to do with it?

Max rated voltage of 2.15V, 1.35V for VCCIO and VCCSA, 1.4V for vDIMM.

To me, this is far from a proper review, because such info was missed. It's hard to make accurate judgements about anything when you don't have all the information. That TIM is actually doing a hell of a job, and these two reviewers that posted these reviews today missed it.


But of course, that's why you're here.



Also note the lack of disclosure about which board they have... You cannot make accurate assessments about this platform under these circumstances.

Indeed. They leave more questions than answers in my mind, frankly.

All I will say is we should wait for other reviews to show up, especially from those who have taken more time to test the platform rather than prematurely release reviews.

Indeed. I just didn't expect a review this early, and thought it better to keep readers in the know than not. I personally eagerly await a very thourough review from TPU, when/if they are able.

 

[BrainCruser]

OK some quick search give me this. Yep, a HUGE shrink of L3 Cache.

So Intel is actually going backwards in IPC. Whoever made this decision in Intel needs to be fired 100 times.

"How are we gonna compete aganist RyZen"
"I dunno, maybe make our processors even worse?"
"Brilliant idea!"

Does Intel's ass control its brain these days?

View attachment 89146



Same. Really wanna see how Intel PR and fanboys are gonna spin this story.

They are rearanging the interconnect inside the core from a ringbus which tops out at around 8-10 cores, to a grid that can handle hundreds of cores(it does actually handle that much cores in the xeon phi).
The grid approach is pretty standard in processors, ARM uses it all the time.

If you remember when sandybridge came out and intel jumped like 15-20% in IPC, That was when the ringbus was introduced. Now it topped out and limited them so they went back.

 

[Gasaraki]

OK some quick search give me this. Yep, a HUGE shrink of L3 Cache.

So Intel is actually going backwards in IPC. Whoever made this decision in Intel needs to be fired 100 times.

"How are we gonna compete aganist RyZen"
"I dunno, maybe make our processors even worse?"
"Brilliant idea!"

Does Intel's ass control its brain these days?

View attachment 89146



Same. Really wanna see how Intel PR and fanboys are gonna spin this story.

The L3 cache has shrink but the L1 and L2 cache has increased in size.

 

[R-T-B]

They are rearanging the interconnect inside the core from a ringbus which tops out at around 8-10 cores, to a grid that can handle hundreds of cores(it does actually handle that much cores in the xeon phi).
The grid approach is pretty standard in processors, ARM uses it all the time.

If you remember when sandybridge came out and intel jumped like 15-20% in IPC, That was when the ringbus was introduced. Now it topped out and limited them so they went back.

Still doesn't explainaway the strange dip in IPC.

 

[cadaveca]

I just didn't expect a review this early, and thought it better to keep readers in the know than not. I personally eagerly await a very thourough review from TPU, when/if they are able.

I did not expect anything either, to be honest but it's cool, because this is useful in ways that are beyond imagination when it comes to credibility.

You know, like those performance dips; yeah, they could be cache, but they could also be a bit of throttle. Also, the high temps and voltage wall could be the TIM, or it could be the FIVR, or that they did not adjust its voltage properly. Obviously these sites needs hits or they'd not have posted such reviews. I am in shock about it all, to be honest.

I'll have samples of these chips and boards, as well as memory for the platform real soon. Not sure if W1zz has planned CPU review yet. If anything, this just makes me a bit more interested to see what's what.

 

[Kakdave]

Still doesn't explainaway the strange dip in IPC.

Early BIOS???

 

[BrainCruser]

Still doesn't explainaway the strange dip in IPC.

Wait for someone to measure the cache latencies. They should be completely rearanged.

 

[cadaveca]

Wait for someone to measure the cache latencies. They should be completely rearanged.

Astute comment, that.

 

[phanbuey]

The 7900X is not priced at the same point or higher than the 6950X, so it is NOT meant to be its replacement. That replacement is not coming for several months. That's why we already have info about a coming 18-core CPU, the 7980XE.

As to TIM, going form 3.3 GHz on 6950X to 4.0 GHz on 7900X seems like a 700 MHz increase to me... and is well within the capabilities of that TIM. Enthusiasts that want better cooling and push the chip past the 297W limit can pop the top, and void the warranty, no big deal.

To push the chip you need to break the HS off of it? I am an enthusiast... i like to be able to resell my chips without having to scrub the freaking liquid metal off the die, or have to disassemble and reassemble the damn thing and glue it back on with silicon.

 

[cadaveca]

To push the chip you need to break the HS off of it? I am an enthusiast... i like to be able to resell my chips without having to scrub the freaking liquid metal off the die, or have to disassemble and reassemble the damn thing and glue it back on with silicon.

No, you don't. But if you want to push to extreme levels, then yes, you do, as you do now with many chips. Those that want what the CPU is capable of for 24/7 use will most definitely be able to get that without any problems. My 6950X chips have issues pushing past 4.4 GHz, and they don' clock ram well, so this 7900X seems like a decent chip to me, especially since it costs LESS than the 6950X, but offers just as many cores.

 

[notb]

OK some quick search give me this. Yep, a HUGE shrink of L3 Cache.

So Intel is actually going backwards in IPC. Whoever made this decision in Intel needs to be fired 100 times.

I guess the 14nm is already exploited as far as IPC goes. They wanted high clocks, so they sacrificed some IPC (most likely by limiting cache). In the end this CPU is still faster. Isn't that what we want? Fast CPUs?

The most important thing in this CPU is that it matches 7700K in single-thread performance.
Yes, more cores is the future, so this CPU is future-proof in the same way Ryzen 7 or Threadripper are future-proof. But this CPU doesn't have any penalty for the present while Zen does.
It will be near the leaders in games and in vast number of single-thread tasks. It'll be just as good in applications that only use 3 or 4 threads.

Same. Really wanna see how Intel PR and fanboys are gonna spin this story.

The power consumption is significant (way to high for me for sure), but not something that we haven't seen before. It'll get better in time (improved node, optimizations).
Based on how Intel usually improves their architecture, it's very likely that in 1-2 years a successor of this CPU will match Ryzen 7 power consumption, while I wouldn't be so sure about Ryzen 7's successor matching 7900X performance...