8700k benchmarks leaked.

[Brooksie359]

1 minute ago, mr moose said:

They could, but then they would be essentially telling laptop designers they need to use a bigger coolers than necessary, which would result in products being more bulky and probably not selling as well.   I am pretty sure getting as close to accurate with the usual qualifiers is standard practice. 

 

 

I am talking about some sku's not laptops. I would assume they would have much better testing for their laptops cpus. desktop cpus are the ones that have really high TDP values that seem larger than what would actually be necessary 

[ravenshrike]

40 minutes ago, mr moose said:

I'm not sure why this is so hard to understand,   TDP is for a pre-defined condition.  It's not an estimate, it is a measured output for said condition (ergo x load at base clock).

VWs emissions test was for a predefined load and output as well. It was also total bullshit. One does not preclude the other.

[mr moose]

11 minutes ago, Brooksie359 said:

I am talking about some sku's not laptops. I would assume they would have much better testing for their laptops cpus. desktop cpus are the ones that have really high TDP values that seem larger than what would actually be necessary 

If anything is different it will be the load used. The reason for having a TDP spec is the same regardless of what device the chip is intended for.  System designers still need accurate information to plan their builds.

2 minutes ago, ravenshrike said:

VWs emmisions test was for a predefined load and output as well. It was also total bullshit. One does not preclude the other.

 

No it wasn't,  it bypassed all the censors when in test mode,  which is just outright fraud.  Nothing like TDP.

 

EDIT: I should say deception rather than fraud.

[Dabombinable]

Not interested in it any more. My 4790K isn't dead, it can still do 4.8GHz, and I'm down to 12GB of RAM as the oldest+slowest stick died (the others were bought within 2 weeks of each other, DDR3L 1600 as opposed to DDR3 1333).

[mr moose]

44 minutes ago, Dabombinable said:

Not interested in it any more. My 4790K isn't dead, it can still do 4.8GHz, and I'm down to 12GB of RAM as the oldest+slowest stick died (the others were bought within 2 weeks of each other, DDR3L 1600 as opposed to DDR3 1333).

 

But will it survive long enough to play SC?

[Dabombinable]

36 minutes ago, mr moose said:

 

But will it survive long enough to play SC?

No idea. I'm more worried about everything else around the 4790K suddenly dying now.

[Juanitology]

10 hours ago, bowrilla said:

Just 6700k and 7700k? I don't see any reason to retire my 4790k. It's about 10–15% slower compared to a 7700k. So how much faster is a 8700k (allegedly)? Another 5-10% maybe depending on the application (if not slower in some single core cases)? Well, thanks, but no thanks. Switching from older systems means basically getting an enitrely new core system: board, cpu and ram. That's about 650-700$? For 15–20% you most likely don't miss since gaming performance won't differ as much as synthetic benchmark loads? I call that a horrible deal.

If you need additional cores and threads: go Ryzen 7 since it's a better deal. If you need even more than that: go Threadripper since X299 is a stupid plattform as it seems. If your system gets the job done as of now: don't do anything about it and 4th generation i7s still work like a charm.

You're missing the point. You only really wanna upgrade from a 4790k for the extra threads and the platform I/O. If you need thunderbolt 3, 12 threads, nvme, etc then it's a worth upgrade. Upgrading just for a gaming workload makes no sense. Save that money and put it toward a better GPU/monitor which unlike CPU performance/technology has actually changed significantly in the last 3 years.

[Thony]

OMG it has "VR" on the box.... the only thing missing is "RGB gaming"....

On the other note, finally higher core count even on i5 !

[Dabombinable]

2 hours ago, Terodius said:

You're missing the point. You only really wanna upgrade from a 4790k for the extra threads and the platform I/O. If you need thunderbolt 3, 12 threads, nvme, etc then it's a worth upgrade. Upgrading just for a gaming workload makes no sense. Save that money and put it toward a better GPU/monitor which unlike CPU performance/technology has actually changed significantly in the last 3 years.

I get NVME support on my Z97 Sabertooth MKII.......as well as Thunderbolt.

[Energycore]

14 hours ago, knightslugger said:

preeetty much

 

I feel the same way, but I'm on a sandy bridge 8T xeon. Not even the fastest one.

[bowrilla]

5 hours ago, Terodius said:

You're missing the point. You only really wanna upgrade from a 4790k for the extra threads and the platform I/O. If you need thunderbolt 3, 12 threads, nvme, etc then it's a worth upgrade. Upgrading just for a gaming workload makes no sense. Save that money and put it toward a better GPU/monitor which unlike CPU performance/technology has actually changed significantly in the last 3 years.

Well, that was basically my point as well. It's not that you wouldn't benefit measurably in quite some cases from the latest generation of CPUs. It's just that it's not worth the money. If you really need (!) extra cores you're probably better off with Ryzen 7, Threadripper or i9. For the average user (and even high end gamer) there's no point in upgrading if you already have a solid i7 bought withing the last 4–5 years. The improvements are measruable but very expensive. And since LGA1151 is most likely approaching EOL there's not realy a point in upgrading.

Don't get me wrong, the 8700k is probably a very solid and good performing CPU and if you need a new system as of now (or within the next few weeks/months) it's a good bet. Upgrading though doesn't make any sense unless you really need the extra cores – but then you're probably better off with Ryzen 7, Threadripper or i9 as mentioned above.

[JediFragger]

14 hours ago, Dash Lambda said:

I wouldn't make such a bold claim. Part of the reason those chips stayed relevant so long is that the market stayed pretty much the same for generations after them -It's not that they were such good chips, it's that everything that came after them for years was an incremental improvement. I'm all for long-lived products, but the implications of that in this context would be pretty depressing...

 

Fair enough. I'd certainly bet on it lasting a fair while though!

[StealthArsenal]

It will be interesting to see "officially" what the cost is going to be for the 8700K and its legit performance in comparison to Ryzen and even lower end X299 chips.  It might just back enough of a punch (if at the correct price point) to derail the aforementioned chips.  For myself I sold off my X99 system some months ago and moved over to my 7700K but am now realizing I actually need more than four cores for some of my CAD work but at the same time, I don't know if I would want to take the single threaded hit in gaming that Ryzen and X299 seem to have.  Might make the 8700K a good all around chip...  I don't know how many are in the same boat or not, but that is my thought process at this time.

[The Benjamins]

19 hours ago, yian88 said:

at least its not the idiotic 5% year on year anymore, thats the "AMD effect"

if 8700k is going for 350$ on shelves then we are in for some great competition ahead

 

im really hyper about ryzen+ hopefully next year release with at least 10% IPC improvement and all the quirks worked out that current ryzen had, and better silicon to reach 4.5ghz then AMD will truly be neck and neck with intel's offerings

Zen+ is a process improvement, which means it will only gain in clock speeds (0 IPC) Zen2 should have clock and IPC gains, I also expect it to be 6c CCX (12c M, 24c HEDT, 48c Server)

[Armakar]

19 hours ago, Prysin said:

At this point, we know how screwed AMD is if "Zen2" isnt a big leap forward in performance.

 

sure, the 1700X is A LOT slower then the 8700k... but 9700k will be a die shrink, along with a totally rebuilt core. Intel already showed some of the cache changes they did in their Xeons to match what AMD did with their cache system. The result was a pretty solid performance increase across the board. Now, 10nm is supposed to be the end of the "core" architecture. Whatever Intel has in the works could sink AMD if they arent extremely aggressive on price.

Yeah, I always said this. While AMD fanboys (though i admit, i prefer amd over intel at this point because fuck intel's prices) were saying "Ryzen destroys intel, AMD are so far ahead" they neglect the fact that Intel could have done something like this whenever they want. If they really wanted to, they'd release something like a 8800X/8800K with 8 cores 16 threads, and the pure difference in silicon and IPC would completely negative Ryzen and we'd be back at intel's 90% market share. AMD really need to focus purely on IPC and silicon quality because they are not even haswell-level yet.

[The Benjamins]

10 minutes ago, Armakar said:

Yeah, I always said this. While AMD fanboys (though i admit, i prefer amd over intel at this point because fuck intel's prices) were saying "Ryzen destroys intel, AMD are so far ahead" they neglect the fact that Intel could have done something like this whenever they want. If they really wanted to, they'd release something like a 8800X/8800K with 8 cores 16 threads, and the pure difference in silicon and IPC would completely negative Ryzen and we'd be back at intel's 90% market share. AMD really need to focus purely on IPC and silicon quality because they are not even haswell-level yet.

The IPC is haswell, The big point is that Ryzen is a very good base, and now they can focus on refining it. I expect them to keep having the core advantage so that will keep them competitive while they type and close the gap in IPC and frequency.

[Armakar]

7 minutes ago, The Benjamins said:

The IPC is haswell, The big point is that Ryzen is a very good base, and now they can focus on refining it. I expect them to keep having the core advantage so that will keep them competitive while they type and close the gap in IPC and frequency.

The ipc is actually quite a bit slower than haswell, it is certainly not haswell level.

http://cpu.userbenchmark.com/Compare/Intel-Core-i7-4790K-vs-AMD-Ryzen-7-1700/2384vs3917

http://cpu.userbenchmark.com/Compare/Intel-Core-i7-3770K-vs-AMD-Ryzen-7-1700/1317vs3917

http://cpu.userbenchmark.com/Compare/Intel-Core-i7-2700K-vs-AMD-Ryzen-7-1700/1985vs3917

Also, you can look at benchmarks of a 4790K vs any Ryzen chip, usually the 4790K wins in games only optimised by 4 cores (indicating haswell ipc > ryzen ipc) and at that it wins convincingly.

 

I would say Ryzen's IPC is somewhere between sandybridge and haswell.  Either way, it's at least 5 years behind in terms of IPC and silicon quality. I mean a sandybridge i7 doesn't have much trouble getting to 4GHZ, and your pretty lucky if you can get any Ryzen chip to 4GHZ. They played the money game now, but lacking in performance means Intel still dictates the market. If intel throw out this 6 core and 8 core at competitive prices; Ryzen will be totally obsolete. Intel has AMD by the balls. 

 

Edit: also, AMD have been playing the "core" game for a while. While I think Ryzen is one of the bigger success stories of their core game, there's only so far they can go with it. Now that Intel have mainstreamed 6 cores with AMD, there will become a point where people want higher IPC rather than a mainstream 16 core processor. I think if AMD's plan is for Zen+ and future Zen generations to have crappy ipc and clockspeeds in exchange for cores, they need to seriously rethink what the consumer wants. At this point in time, i'd take a 6 core 8700K over a 8-core 1700 anyday; the clockspeed and ipc advantage is too good to turn down.

 

"juSt AdD moRe CoREs"

[Tic-Tac]

So..... When they're going to release this??

[The Benjamins]

1 minute ago, Armakar said:

The ipc is actually quite a bit slower than haswell, it is certainly not haswell level.

http://cpu.userbenchmark.com/Compare/Intel-Core-i7-4790K-vs-AMD-Ryzen-7-1700/2384vs3917

http://cpu.userbenchmark.com/Compare/Intel-Core-i7-3770K-vs-AMD-Ryzen-7-1700/1317vs3917

http://cpu.userbenchmark.com/Compare/Intel-Core-i7-2700K-vs-AMD-Ryzen-7-1700/1985vs3917

Also, you can look at benchmarks of a 4790K vs any Ryzen chip, usually the 4790K wins in games only optimised by 4 cores (indicating haswell ipc > ryzen ipc) and at that it wins convincingly.

 

I would say Ryzen's IPC is somewhere between sandybridge and haswell.  Either way, it's at least 5 years behind in terms of IPC and silicon quality. I mean a sandybridge i7 doesn't have much trouble getting to 4GHZ, and your pretty lucky if you can get any Ryzen chip to 4GHZ. They played the money game now, but lacking in performance means Intel still dictates the market. If intel throw out this 6 core and 8 core at competitive prices; Ryzen will be totally obsolete. Intel has AMD by the balls. 

 

None of those are IPC tests. To compare IPC you need both chips at the same clock, and clock for clock Ryzen is around haswell (beating it in most cases)

 

Silicon quality is not a AMD issue but a TSMC issue.

 

Why do you think Intel will release a mainstream 8c in the 8000s generation? everything I have seen is 6c.

 

and AMD seems to have a 12c mainstream when zen2 lands at the end of 2018 on 7nm.

[Dash Lambda]

46 minutes ago, Armakar said:

The IPC is actually quite a bit slower than Haswell, it is certainly not Haswell level.

http://cpu.userbenchmark.com/Compare/Intel-Core-i7-4790K-vs-AMD-Ryzen-7-1700/2384vs3917

What you're seeing there is the clock advantage.

[MyName13]

54 minutes ago, The Benjamins said:

The IPC is haswell, The big point is that Ryzen is a very good base, and now they can focus on refining it. I expect them to keep having the core advantage so that will keep them competitive while they type and close the gap in IPC and frequency.

Nope, r3 overclocked to 4 GHz (or 3.8) can sometimes barely compete with i5 7400 clocked at 3 GHz (that's 33% difference in clock rate), in some cases it's even up to 50% behind (not just r3), IPC is somewhere around sandy/ivy bridge up to haswell (depending on workload), they are really behind intel and I really have no idea how will they compete after coffee lake, this is what I imagine will happen with Zen 2 (after CL comes out only r7 1700 will be competitive and maybe r5 1600 ONLY in highly multithreaded tasks)

 

r3 - 4c8t clocked to 4-4.5 GHz

r5 - 6c12t (these could be somewhat uncompetitive, but if they can clock to at least 4 GHz they should be at least on par with intel's i5) and 8c16t (these should destroy i5 x600k in multithreaded workloads but not in single threaded tasks)

r7 - 10c20c and 12c24t (if the rumours about 6 cores per ccx are true)

 

 

[Taf the Ghost]

6 hours ago, The Benjamins said:

Zen+ is a process improvement, which means it will only gain in clock speeds (0 IPC) Zen2 should have clock and IPC gains, I also expect it to be 6c CCX (12c M, 24c HEDT, 48c Server)

From discussions, it looks a lot more like Zen2 will be 3 CCX per package given the technical realities. We'll find out for sure the first time we see a 9c benchmark leak out.

 

As to the 8700k, it's a Kaby Lake Refresh with some little tweaks here and there. The IMC being, possibly, even better could be the best potential aspect for a lot of high-end users. Might see 4000 ram OCs?

[Taf the Ghost]

3 hours ago, anthonyjc2010 said:

Intel's 8th generation (non-HEDT) processors are Kaby Lakes with more cores/threads. Now, in all reality, these will be slightly different from the original Kaby Lakes released last September. The 7th Gen Kaby Lakes being produced now are different from the originals last September. Intel tends to slightly improve their processors while in the same generation (the difference in IPC and temps are almost non-existent; it's more for efficiency in manufacturing process). Back to the point, these are Kaby Lake refreshes, not in the way the Kaby Lake is a refresh of SkyLake, but as in the same architecture being shipped, again...

 

I'd complain, but Intel probably just didn't want to release Cannon/Coffee Lake (whichever one is the 10nm redesign with the new architecture) because A) it's not ready, and B) they don't need to. The 8700K will* outperform the 8/16 consumer Ryzen CPU in both single and multi-threaded performance. What AMD will have to do is lower pricing (again) or add even more incentives besides ECC memory support on consumer boards, and overclocking on all CPUs.

 

The next two generations, 8th and 9th (late 2017 - late 2019), are going to be the embodiment of "Intel Strikes Back" with all of their technology to nuke RYZEN into the ground, along with AMD's entire CPU division for the next four years. We all should be thankful that AMD came in and forced Intel to give us more than a crappy tock cycle, I just hope AMD can find a way to be competitive and continue to fight Intel for the next few years, or even longer!

 

 In almost all cases it's unnecessary and unadvisable.

 

 

10nm and 14nm+ are the same efficiency at higher clocks. We don't seem to know if 10nm can clock as high, which could be the real problem. This is why Coffee Lake happened and Cannonlake is only coming to some ultra-low powered devices. Cannonlake is a Skylake die-shrink, so there's been no big architectural changes to come since 2015 for Intel.

 

Icelake is 10nm+ and will also be an actual uArch change. Nothing suggests it's a Sandy Bridge jump, but should see at least some IPC improvements. The interesting question will be clocks. Is it coming in starting at 4.5 Ghz or so? Or are we going to see high 3 Ghz and they improve that over time? We'll find out probably in Q2`2018 when leaks start rolling out, as it should be an end-of 2018 release.

[The Benjamins]

9 minutes ago, Taf the Ghost said:

From discussions, it looks a lot more like Zen2 will be 3 CCX per package given the technical realities. We'll find out for sure the first time we see a 9c benchmark leak out.

 

As to the 8700k, it's a Kaby Lake Refresh with some little tweaks here and there. The IMC being, possibly, even better could be the best potential aspect for a lot of high-end users. Might see 4000 ram OCs?

My reasoning for 6c CCX is that AMD confirmed SP3 will support the next generation of server CPU which is confirmed to be a 48c Zen2 based CPU. and knowing that they have to keep support and that each CCX has a IMC. keeping SP3 support means that it has to have 8 IMC which means 8 CCXs. so 48/8 gives me 6c per CCX.

 

keep in mind this is my speculation on what will happen and to me it makes the most sense assuming socket support.

 

Also a 4 Die package is balanced with each die having a connection to the other 3 were a 6 die package would not be able to do that. (with out losing a lot of PCIe) and it would be odd to have 4 IMC unusable.

 

A 3 CCX die/configuration seems fine to me but is a odd setup.

 

EDIT: also AMD has not used R9 yet so they could on zen2 generation launch R3, R5, R7, R9 with the R9 being 10-12 core parts.

[Taf the Ghost]

9 minutes ago, The Benjamins said:

My reasoning for 6c CCX is that AMD confirmed SP3 will support the next generation of server CPU which is confirmed to be a 48c Zen2 based CPU. and knowing that they have to keep support and that each CCX has a IMC. keeping SP3 support means that it has to have 8 IMC which means 8 CCXs. so 48/8 gives me 6c per CCX.

 

keep in mind this is my speculation on what will happen and to me it makes the most sense assuming socket support.

 

Also a 4 Die package is balanced with each die having a connection to the other 3 were a 6 die package would not be able to do that. (with out losing a lot of PCIe) and it would be odd to have 4 IMC unusable.

 

A 3 CCX die/configuration seems fine to me but is a odd setup.

Technically, "Starship" is only one reference from a nearly 2 year old slide pointing to a 48c server CPU. It's possible they dropped it for Zen2. Just as a foregoing note.

 

To the design situation, I've been following along as people with a lot more CPU technical insight have been chatting about it. It appears the "easier" design approach with adding another CCX is that the calls between CCX are always +1 hop from within the CCX. This makes the CCX more like a 4-split Core as much as a 4c design. Putting a third CCX inside the package makes the design more simple as the calls from CCX0 to either CCX1 or CCX2 the same.

 

To me, what is probably the key, is that the Zen Package can be run in 4+0 mode and still be in dual-channel memory. It means the CCXs are actually separate from the memory infrastructure that's above them in the I/O chain. It means you can run two DDR4 controllers with however many CCXs you can run traces to the controllers. It should also mean that the CPU wouldn't need more pins, which lets you stay on the same socket. 

 

Zen is a really different design philosophy. It's very much like Lego bricks, but we're not quite to the final form of what that looks like. That comes with "Chiplet" in a few years.