Fabbing the GPU on 14 nm makes sense to free up 10 nm wafers, which they will need. Tiger Lake is already murdering their margins, they must be burning a ton of wafers to try to make up for the bad yield.
DrMrLordX
Lifer
- Apr 27, 2000
- 21,637
- 10,855
- 136
@Antey
Hmm.
Or from one of our local posters!
@Zucker2k
The 10900k appears to be just as fast in ST and significantly faster in MT. Something isn't right with that Rocket Lake sample. Might have some ES flaws in it, so I'll try not to read too much into the situation, but Intel's supposed to have that thing ready by Q1 2021.
@jpiniero
Intel doesn't have much 14nm capacity to spare either.
Hmm.
Gigabyte Technology Co., Ltd. Z490 AORUS MASTER - Geekbench
Benchmark results for a Gigabyte Technology Co., Ltd. Z490 AORUS MASTER with an Intel Core i9-10900K processor.
browser.geekbench.com
Or from one of our local posters!
Micro-Star International Co., Ltd. MS-7C71 - Geekbench
Benchmark results for a Micro-Star International Co., Ltd. MS-7C71 with an Intel Core i9-10900K processor.
browser.geekbench.com
@Zucker2k
The 10900k appears to be just as fast in ST and significantly faster in MT. Something isn't right with that Rocket Lake sample. Might have some ES flaws in it, so I'll try not to read too much into the situation, but Intel's supposed to have that thing ready by Q1 2021.
@jpiniero
Intel doesn't have much 14nm capacity to spare either.
Well, if the hyperscalers start rolling off Cascade Lake and onto Rome, there'd be plenty of capacity...
DrMrLordX
Lifer
- Apr 27, 2000
- 21,637
- 10,855
- 136
I don't think Intel will let that happen. Even if they have to drive margins below 20%. Whether or not that will be enough is anyone's guess. But Intel will not give an inch.
uzzi38
Platinum Member
- Oct 16, 2019
- 2,635
- 5,983
- 146
This is the first 9900K score I saw that was close to the average (1340) in single core score: https://t.co/AnT5PNufVH
If you look at the score breakdowns going from the 9900K -> RKL-S, you'll see Crypto doubled, but Int and FP are basically the same.
From this one score, AVX-512 looks to be the ONLY IPC uplift.
If you look at the score breakdowns going from the 9900K -> RKL-S, you'll see Crypto doubled, but Int and FP are basically the same.
From this one score, AVX-512 looks to be the ONLY IPC uplift.
i agree. but that rocket lake ES is 4,98GHz and that cometlake is 5,35GHz... there is a mmmh... 7-8% difference? i hope is not close to final results though.
DrMrLordX
Lifer
- Apr 27, 2000
- 21,637
- 10,855
- 136
Just off-the-cuff, I'm thinking it looks a lot like Cannonlake. We need more samples though. That can't be right.
@Antey
The alleged IPC uplift of Rocket Lake-S should more than compensate for that. Also it only boosts opportunistically in the ST portion of the test. That does not explain why Rocket Lake-S was so slow in MT since a 10900k won't sustain 5.3 GHz in the MT test.
The clocks part was always the biggest question mark here, or? The ipc is already baked in, and it must be worth the porting, so, I'm quite optimistic about the outcome of the final product.
uzzi38
Platinum Member
- Oct 16, 2019
- 2,635
- 5,983
- 146
I know, everyone with some serious thinking expected 10nm from ADL, however Charlie and some other usual suspects tried to defend the 14nm hypothesis till the end:
A better comparison would be Intels 9900K which boost up to 5.0 Ghz as well and reaches about 1300 ST points, excluding the AES score RKL-S seems to be 10% faster per clock. However this is a stepping 0 and no one knows if this is fully representative, I mean this is the very first result for such high clocked RKL model. The clock speed looks pretty good for a stepping 0.
I don’t know why people have high expectations of Rocket Lake in the first place. I highly doubt it will reach Sunny Cove IPC, let alone the >20% improvement that many are expecting. Although this few % improvement in int/float scores surprises me too, it’s hopefully not final performance.
uzzi38
Platinum Member
- Oct 16, 2019
- 2,635
- 5,983
- 146
And you're expecting a 15% IPC uplift between ES and final silicon?
I'm not going to put a figure on it, yet. I'll let the final product do the talking.
DrMrLordX
Lifer
- Apr 27, 2000
- 21,637
- 10,855
- 136
Maybe, though Rocket Lake-S should feature TVB (9900k does not) enabling it to hit ST clocks close to Comet Lake-S at the top end. TVB may not be working in the ES samples yet. Inevitably, the top-end 8c Rocket Lake-S will be compared against Comet Lake-S, and need I point out (again) that the MT score for that Rocket Lake ES is surprisingly low???
25% IPC uplift over Skylake isn't realistic. On Geekbench v5 Sunny Cove got a 15% IPC uplift over Skylake (which included a memory advantage), so there is only a 5% deficit.
DrMrLordX
Lifer
- Apr 27, 2000
- 21,637
- 10,855
- 136
Rocket Lake-S could have been Willow Cove though. In theory it should be somewhere between Sunny and Willow, depending on what compromises Intel had to make for the backport. Also see commentary by @uzzi38 . The Rocket Lake sample seems to be leaning heavily on AVX512 for its score.
We really need more data.
tamz_msc
Diamond Member
- Jan 5, 2017
- 3,821
- 3,643
- 136
There are enough differences between version 5.0 and 5.2 such that comparing the two is not appropriate. I recall reading about it in the Geekbench release notes.
That Rocket Lake system is running memory setup that has high latency or bad bandwidth ( or both ). Look at Navigation sub test, there is just no way Comet is scoring 50% more in a test that is sensitive to memory.
( And Navigation is sensitive to memory, as that is the only test where my 5.1 system can beat AdamK's 5.3 due to tighter latency).
Horrible MT Speech Recognition / Machine Learning scores point to same thing.
Proper testing is needed, not with what is probably botched memory setup. Might as well ran 16GB single stick 2133C30 for what we know to produce these results.
( And Navigation is sensitive to memory, as that is the only test where my 5.1 system can beat AdamK's 5.3 due to tighter latency).
Horrible MT Speech Recognition / Machine Learning scores point to same thing.
Proper testing is needed, not with what is probably botched memory setup. Might as well ran 16GB single stick 2133C30 for what we know to produce these results.
inf64
Diamond Member
- Mar 11, 2011
- 3,703
- 4,032
- 136
RCK ES vs 10700K, same GB version, x64:
edit, also vs 3800XT ,same GB version:
edit, also vs 3800XT ,same GB version:
Generic vs Intel Corporation Rocket Lake Client Platform - Geekbench
browser.geekbench.com
Last edited:
DisEnchantment
Golden Member
- Mar 3, 2017
- 1,608
- 5,816
- 136
uzzi38
Platinum Member
- Oct 16, 2019
- 2,635
- 5,983
- 146
TRENDING THREADS
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)
- Started by DisEnchantment
- Replies: 10K
-
-
Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)- Started by Vattila
- Replies: 13K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 7K
Facebook
Twitter