Reallink
Member
Doesn't the 7700k come out in Jan, why would anyone be comparing these to the 6700 at that point?
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
New Horizon: AMD New ZEN CPU Preview Event
- Thread starter Hawk269
- Start date
Doesn't the 7700k come out in Jan, why would anyone be comparing these to the 6700 at that point?
qa_engineer
Member
Just because it can have higher bandwidth doesn't mean it should. The GPU may not need that type of bandwidth to saturate, read/write to memory. HBM2 does in fact have a bandwidth advantage over HBM1, but thats not the only advantage. Going beyond 4 gb is another advantage. I would imagine the Vega 11 GPU will have HBM2 with higher bandwidth, assuming it's going to compete with the TITAN X.
Fury X has 4 stacks of HBM1, Vega 10 presumably have 2 stacks of HBM2. It's 2x faster per stack but since it uses less of them you end up with the same speed.
I imagine Vega 11 will have 16gb and 4 stacks of HBM2.
what exactly has Intel launched last year?
LordOfChaos
Member
8 entry L0 ITLB; another slide indicates that the L0 and L1 TLB are looked up in parallel, I'm guessing. The small TLB is probably dedicated to the micro-op cache.
512 entry L2 ITLB and 1536 entry L2 DTLB; this is for server workloads
Distributed scheduler which is the norm for AMD and was only changed in Bulldozer; Intel, on the other hand, has a unified scheduler for all architectures except the various Atoms. This looks like a performance/watt tradeoff. Unified schedulers are better for instruction throughput, but consume more power than split schedulers of a comparable size.
Differential checkpoints: this helps reduce the size of the state that needs to be saved from one branch to the next. That will, in turn, reduce power, and may reduce the mispredict penalty.
L3 is mostly exclusive of L2; sounds like Barcelona and Bulldozer which was exclusive except the storage of lines shared by different cores/modules.
Each core can access all of the L3 slices with the same average latency; this indicates that the L3 interconnect isn't a ring
512 entry L2 ITLB and 1536 entry L2 DTLB; this is for server workloads
Distributed scheduler which is the norm for AMD and was only changed in Bulldozer; Intel, on the other hand, has a unified scheduler for all architectures except the various Atoms. This looks like a performance/watt tradeoff. Unified schedulers are better for instruction throughput, but consume more power than split schedulers of a comparable size.
Differential checkpoints: this helps reduce the size of the state that needs to be saved from one branch to the next. That will, in turn, reduce power, and may reduce the mispredict penalty.
L3 is mostly exclusive of L2; sounds like Barcelona and Bulldozer which was exclusive except the storage of lines shared by different cores/modules.
Each core can access all of the L3 slices with the same average latency; this indicates that the L3 interconnect isn't a ring
ethomaz
Banned
Skylake.
They already even launched Kaby Lake and Cannonlake (10nm) hit the market in H2 2017... Coffee Lake (14nm) launches in H1 2018 too.
CarbonFire
Member
Because the 7700k improves very, very little on the 6700k based on early benchmarks. Kaby Lake is a very minor bump in efficiency and not much else outside of general platform improvements. This is Intel's first break from their Tick-Tock architecture strategy since they introduced the Core 2 Duo 10 years ago. We're not getting the normal bump from either a new architecture or a smaller transistor size, since the 10nm Cannon Lake was delayed from 2017 to 2018.
The 7700k will be better than the 6700k in a few marginal ways, but not enough to push the 6700k out of the conversation when comparing it to other processors from AMD.
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
You think the uop cache is virtually tagged? That's very unlikely. It's more likely that the L0 iTLB is a general-use iTLB with full associativity (8-entry, fully associative == very fast, very hit-efficient, very tiny cache). Such a cache would improve the latency of all other physically-tagged lookups.
icecold1983
Member
at the same clockspeeds its like .3% faster
What A Time To Be Alive
Member
Cannonlake got delayed to 2018.
Coulomb_Barrier
Member
Yeah and Kaby Lake is literally identical to Skylake in terms of performance.
SolidSnakeUS
Member
If Ryzen can come up to the performance of these two processors, at least (even though it has more cores), then it will be a big win for Ryzen and AMD. It just needs to be priced well. I fucking hope it is.
CarbonFire
Member
If the September Intel Roadmap leak is accurate, we're getting lower powered 10nm Cannonlakes late next year (for laptops and Ultrabooks), and then Coffee Lake will be the next higher-powered offering from Intel in Q2 2018. And Coffee lake appears to be yet another tock, the third in a row running on 14nm, but possibly with actual microarchitecture improvements over Kaby and Sky this time.
Which all makes me think Intel is having real problems with increasing performance on the 10nm process. Because at this point, if we're getting 14nm Coffee lake in 2018, a 10nm desktop part probably isn't in the cards until 2H 2019 at the earliest. That is a little depressing, but a boon for AMD if Ryzen can put up good numbers.
PLASTICA-MAN
Member
So the new CPU will be called RyZen. Will there have different numbers for different configurations?
.
.
GearDraxon
Member
Returning next year: the classic Sonic the Hedgehog action you love, only on AMD Zen!
Locuza
Member
Nope, there will be only one chip, Summit-Ridge, with 8-cores max.
In the second half of the year (Q4) AMD will launch Raven-Ridge which is an APU with 4-cores max + iGPU.
For the server market it seems that AMD will use the same chip they use for Summit-Ridge and connect two (2x8-cores) or four (4x8-cores) of them on one package.
I think the i3 K series is more or less Intel's counterattack to cheap 4 core ryzen, as kaby lake is just a more oc friendly skylake???
Plus, Vega will have improved delta color compression over Fiji, and potentially Polaris.
What A Time To Be Alive
Member
hooijdonk17
Member
Everyones crazy about Ryzen but I think Raven Ridge will be immense for HTPC anx ITX builds.
Unknown Soldier
Member
This thread is a solid reminder why posting in a thread solely to question why it exists is not a good idea.
Locuza
Member
It will be quite interesting.
Every 8-core Ryzen will have at least 3,4 Ghz baseclock which is quite good for a 95W TPD rating, it leaves a nice headroom for quad-cores.
If quad-core Ryzen can ship with 3,8 Ghz or even more than Intel might be not as lucky.
chaosblade
Unconfirmed Member
Raven Ridge is going to be a lower end APU. Good for HTPCs and non-gaming builds that will use the IGP.
For a gaming build, a lower end 4C Summit Ridge CPU will probably be a better option than a higher end 4C Raven Ridge CPU. Even if they perform the same the Summit Ridge option should be cheaper since it won't have an IGP.
Finally got around to seeing it. Not having FPS counters on the games was offputting. I didn't expect a price but if you're making a demo I expect a number, you're not trying to sell us HDR or Freesync.
I am very curious about their performance scaling with cooling. Liquid Nitro isn't my thing but it would be really cool if the system is stable while performing at 50% more clockspeed instead of being unstable.
CPU's needed enthusiast geared innovation like that.
What I'm even more curious about is their machine learning within the CPU. Is it really a state machine that learns through trial and error or is it something that simply adapts to repeated usage. I expect the latter which isn't machine learning at all and I would like to know how many run throughs does a zen machine need to go through before it gets that 25% handling of workload the CEO mentioned.
I am very curious about their performance scaling with cooling. Liquid Nitro isn't my thing but it would be really cool if the system is stable while performing at 50% more clockspeed instead of being unstable.
CPU's needed enthusiast geared innovation like that.
What I'm even more curious about is their machine learning within the CPU. Is it really a state machine that learns through trial and error or is it something that simply adapts to repeated usage. I expect the latter which isn't machine learning at all and I would like to know how many run throughs does a zen machine need to go through before it gets that 25% handling of workload the CEO mentioned.
hooijdonk17
Member
Well, yeah obviously a dGPU coupled with Ryzen or even Summit Ridge will be better but if you want to build something ultra portable and in a low power envelope in an ITX case Raven Ridge will make decent 1080p gaming on an APU fairly managable.
Regardless, my comment was mostly wrt HTPCs and the like. I sure would like to upgrade my aging HTPC to summit ridge and use it for comfy couch gaming with a good portion of my Steam Library that it can handle at 1080p whereas now I need to look around for a decent low profile dgpu option that won't add to the noise and power draw.
ethomaz
Banned
From Intel schedule.
Mobile low-power Cannonlake will release in late 2017 in 10nm.
It won't has in 2017 a desktop/server part... that will be Coffee Lake in early 2018 in 14nm yet.
That happened because 10nm is not ready for desktop/server yet... so they added Coffee Lake to fill this gap until 10nm get mature enough.
Nostremitus
Member
I recall reading that AMD are only planning 4, 6, and 8 core variants, all with their version of hyperthreading.
If that's the case, those i3s will be up against 4c8t chips with near(or actual) i7 performance.
Beerman462
Member
I paid $240 for my i5 6600K this spring. If AMD can match that, over clocking and all at a sub $200 that would be a game changer to me. I'm not expecting to far below $200, but $180-$190 would be great.
I3K is the worst sneak attack ever. Most useless CPU intel has ever released if the rumoured price is correct, considering you can BCLK overclock i5-6400 to 4.2ghz+. As long as you get a mobo that supports it you get 2 extra cores for the same price, significantly better buy than an unlocked i3 for $180. Heck you can even get a 6100 and OC it with the same boards if you really are going for low budget.
Wouldn't even be surprised if Kaby Lake will have MBs that let you BCLK overclock the 7000-series as well.
LordOfChaos
Member
Indeed unlikely, the size of the L0 TLB is smaller than would be required to cover the L1 I-cache; this is why I speculated that it might be dedicated to the micro-op cache. However, looking up the L0 and L1 ITLBs in parallel and clock gating the L1 in case of a L0 hit would be effective too.
I would not call Intel's release schedule as a positive thing... their latest updates have been really mediocre, and many driver issues with integrate gpu's dont help. They have been maximizing profits on desktop parts lately due to no competition.
This is best time for AMD to do something about it.
Sounds to good but I'm gonna wait with my upgrade.
Would probably take a 6 core at least (6800k or so). So if RyZen 8 core is priced right and not to expensive I'm in. Intel could react and reduce their 6 cores in price - would also be ok for me.
Think RyZen has no internal GPU so AMD can save some costs and the Broadwell-Es are probably way overpriced right now, as 6900k is not so much a gamer or endconsumer CPU. And Intel still has the Core i7-6950X for professional market.
For gaming one could stop hyperthreading - still 8 cores and that would allow some more overclocking perhaps. It is gonna be interesting how this will play out and how the singlethread performance will be. Still skeptical - but if it is not utter trash it should shake up some things.
Would probably take a 6 core at least (6800k or so). So if RyZen 8 core is priced right and not to expensive I'm in. Intel could react and reduce their 6 cores in price - would also be ok for me.
Think RyZen has no internal GPU so AMD can save some costs and the Broadwell-Es are probably way overpriced right now, as 6900k is not so much a gamer or endconsumer CPU. And Intel still has the Core i7-6950X for professional market.
For gaming one could stop hyperthreading - still 8 cores and that would allow some more overclocking perhaps. It is gonna be interesting how this will play out and how the singlethread performance will be. Still skeptical - but if it is not utter trash it should shake up some things.
IcyRhythms
Member
I agree with JayTwoCents from Youtube. While Intel has definitely had the superior product, they haven't really done anything to completely change the game since Sandy Bridge. They've been very lazy, and no competition will do that. I hope AMD knocks it out of the park with price and performance.
shark sandwich
tenuously links anime, pedophile and incels
I'm just glad someone is building a consumer CPU that doesn't spend half its transistors on an IGPU that I have zero use for.
Really, really hoping this ends up being on par with the 6900k.
Really, really hoping this ends up being on par with the 6900k.
SapientWolf
Trucker Sexologist
More perms than a 70s barber shop.
They showed it matching with boost disabled which was interesting.
It all depends on how much boost these things have got which is unknown until the production line rolls proper I suppose.
cartman414
Member
I wonder if this is partly why Intel is feeling comfortable in rolling out 6 core mainstream parts (with Coffee Lake) within the next 2 years. Of course that's still 2 cores less than what AMD has, but they definitely have leeway here.
Wouldn't be surprised if they were still behind in single threaded performance. That, plus the higher cost of having more cores to make up for it, might limit how low the price can go, even with slimmer margins.
I doubt Intel has been resting on its laurels and choosing not to significantly improve performance. I mean, the lower the performance gains, the less likely people are to upgrade (you'll find no shortage of people on GAF still using a 2500K or 2600K, for example). No doubt that narrative will become rather prominent should Zen represent what AMD has been claiming, however, with people not realising that AMD was able to make significant strides because it was so far behind.
LordOfChaos
Member
Yeah, Intel not focusing on what gamers may want doesn't mean they've been twiddling their thumbs for years, they just decided for most users CPU performance was no longer limiting, so focused increased transistor budgets through die shrinks on lower power draw and better integrated graphics (and also that to shove forward performance would take exponentially increasing complexity, not many have gotten past their single thread performance, Power8, but that is huge and power hungry).
I think we've just come to take for granted that todays two pound ultraportable performs better than a 5 year old, 5.5 pound 15" laptop with whopping two hour battery life. Didn't happen through magic, but Intels efforts outside of what enthusiast gamers want.
All this is true, but they've been perfectly happy to gouge on more marginal increases in performance. I think the narrative that they haven't done anything is conditioned on the fact that each subsequent generation of desktop processor since Sandy Bridge has been worse value for money.
Weren't they comparing it to a $1k CPU? Why would they price it that low? Maybe a nerfed version will come out at a budget price.
Beerman462
Member
There's going to be tiers right? Not just 8core monsters.
Here is the thing, the 6900k is a HEDT but it isn't really targeted at people playing games. The event in question was targeted at gaming audiences, primarily, so it is almost certainly going to offer better prices than $1K.