AMD - Zen chitchat

[kalniel]

Interesting comments from TTL:

https://forum.overclock3d.net/showthread.php?t=80970

He's basically accusing this site of breaking NDA?

[kompukare]

He's basically accusing this site of breaking NDA?

I though I read in the Computerbase story about why they have no review (Intel didn't supply any and the one they sourced is held up in customs), that Intel didn't necessarily use a NDA for this launch...

[CAT-THE-FIFTH]

He's basically accusing this site of breaking NDA?

I think you might need to look at the single comment fromm Tinytimlogan who is the chap who runs the site:

Id hold fire on that for a bit.... Im debating weather or not I drop a bomb about CL

[Phage]

Interesting comments from TTL:

https://forum.overclock3d.net/showthread.php?t=80970

OOOOOOOOOoooooooooooh

[CAT-THE-FIFTH]

From The Stilt over on AT forums:

http://i.imgur.com/68n8LJN.jpg


So running low latencies can probably make up for not getting the high speed RAM I suspect.

[scaryjim]

I though I read in the Computerbase story about why they have no review (Intel didn't supply any and the one they sourced is held up in customs), that Intel didn't necessarily use a NDA for this launch...

Well, TTL seems to think there was an NDA as the overclock3d review states that they went out and sourced their own chip so they could actually post a review when the NDA lifted, because they didn't have one sourced from Intel. And it'd be odd to accuse Hexus of breaking NDA if there wasn't an NDA.

OTOH, perhaps Intel dropped the ball over NDA sign-ups and didn't make sure that all the major tech sites were covered? Hexus couldn't break NDA if Intel had forgotten to get them to sign one! Looking around the web, it does rather seem that all the other sites posted their reviews/articles this morning, while Hexus dropped theirs on Friday afternoon. That suggests that either Hexus weren't under NDA or were under a different NDA to everyone else.

... running low latencies can probably make up for not getting the high speed RAM I suspect.

Odd that he's tested only the fastest setups with lowered latencies - that looks like a classic case for going back and retesting 2666MHz RAM with tight latencies to see what happens. It makes sense though - reducing latency to main memory would mask the hit from cross-CCX communication as it'd be faster to serve cache hits from main memory...

[kalniel]

Yes we were hoping to bring you a review of both the eight core i7-7820X and the ten core i9-7900X today, but for reasons best known to themselves Intel withheld samples from everyone because a couple of sites broke NDA last time. So we're sans samples and other sites - mentioning no names, let's call them Gexus - broke NDA anyway. If I was Intel's lawyers I'd be expecting some work in the coming days. All of which doesn't overcome the fact that we just have the i7-7820X to show you today.

And other sites have reviewed the supposedly missing 7900X (http://techreport.com/review/32111/i...iewed-part-one for eg.), so not quite sure what he's talking about.

[CAT-THE-FIFTH]

Wow:

https://www.pcper.com/reviews/Proces...d-Latency-and-

Inter-chip latency is pretty high!!

[Spreadie]

Well, TTL seems to think there was an NDA as the overclock3d review states that they went out and sourced their own chip so they could actually post a review when the NDA lifted, because they didn't have one sourced from Intel. And it'd be odd to accuse Hexus of breaking NDA if there wasn't an NDA.

OTOH, perhaps Intel dropped the ball over NDA sign-ups and didn't make sure that all the major tech sites were covered? Hexus couldn't break NDA if Intel had forgotten to get them to sign one! Looking around the web, it does rather seem that all the other sites posted their reviews/articles this morning, while Hexus dropped theirs on Friday afternoon. That suggests that either Hexus weren't under NDA or were under a different NDA to everyone else.



Odd that he's tested only the fastest setups with lowered latencies - that looks like a classic case for going back and retesting 2666MHz RAM with tight latencies to see what happens. It makes sense though - reducing latency to main memory would mask the hit from cross-CCX communication as it'd be faster to serve cache hits from main memory...

Would an NDA hold water if Intel refused to send out samples?

"here's our new chip for you to play with, but you can't say anything until {date}. Sign here." makes more sense than "we ain't sending you a sample, but sign here anyway".

[kompukare]

Wow, long time no see: Intel do price cuts, up to half price for lots of KL's.
https://www.computerbase.de/2017-06/...eon-phi-preis/
Seems KL is not selling that well, or they are worried that they are not making the inroads into HPC as they had expected.

With Skylake-X, the new cache seems to have been designed for HPC loads. Or more probably AVX512 loads. And so far it seems that's hurting it's gaming performance. Will be interesting to see if it hurts their normal server loads too.

That, inter-core latency CAT posted doesn't help. At least with Ryzen, it only affects CCX-to-CCX while with Skylake-X it seems to affect all cores.

[CAT-THE-FIFTH]

Ryzen based laptops are finally available:

https://forums.overclockers.co.uk/th...hics.18783753/

[AETAaAS]

Ryzen based laptops are finally available:

https://forums.overclockers.co.uk/th...hics.18783753/

Good god, AMD haven't revealed mobile parts have they? Are these full desktop socketed parts? Because that might be cool considering AMD tends to have longer socket longevity, and if Asus decide to, they may be able to make a low cost model with later AM4 APUs.

[CAT-THE-FIFTH]

Good god, AMD haven't revealed mobile parts have they? Are these full desktop socketed parts? Because that might be cool considering AMD tends to have longer socket longevity, and if Asus decide to, they may be able to make a low cost model with later AM4 APUs.

These are desktop parts - hence why they are 6C/12T and 8C/16T. The SOC nature of Ryzen really helps here when keeping the motherboard size down too.

[watercooled]

Wow that's... interesting.

It's yet another thing that goes to show how far AMD have come in terms of power efficiency. I wonder if these are coupled with a chipset, or if they're just using the SoC as-is?

[scaryjim]

The laptop was announced a while ago: http://hexus.net/tech/news/laptop/10...laptop-demoed/

Based on the EPYC server chip options it looks like 8-core Ryzen could drop as low as 35W given the right memory and clock targets, although it wouldn't break any performance records at that TDP. The ASUS laptop uses 65W versions of both the CPU and GPU, so the CPU is essentially a full R7 1700, while the GPU is presumably clocked down to hit the efficiency sweet spot.

For a laptop I don't see any benefit to a chipset - you can hang more than enough storage & IO directly off the CPU, so running as A/X300 would make implicit sense. Which probably means they have used a chipset, just because

[CAT-THE-FIFTH]

Barely two weeks between the announcement and availability has got to be some record for an AMD performance laptop! It kind of shows that OEMs seem enthusiastic about Ryzen.