By: -.- (blarg.delete@this.mailinator.com), May 26, 2022 5:58 am
Room: Moderated Discussions
Adrian (a.delete@this.acm.org) on May 26, 2022 5:16 am wrote:
> It seems that the 15% increase was in the Cinebench ST benchmark.
From where did you get that info?
> The optimists say that maybe AMD has intentionally presented worse results than would be actually obtained
> in the production units, to mislead the competition, but such a strategy does not make much sense.
Worth pointing out that they say "5GHz+", but then show a 5.5GHz processor.
Is that underselling? I don't know. Maybe the 5.5GHz chip was some golden sample and they'll only commit to 5GHz. Or maybe 5.5GHz is what they plan to offer (in a separate interview, they claimed the 5.5GHz processor wasn't overclocked) but are being conservative in their figures, for whatever reason.
So should the ">15%" be associated with the 5.5GHz chip they demoed, or the "5GHz+" mentioned on the same slide?
At the end of the day, you can only really speculate with the small amount of info provided, and you could argue either way, and no-one (non insiders) would be able to tell what is correct.
> While AMD claimed that the 4/3 performance advantage
Don't confuse performance increase with time decrease. 30% less time is different to 30% faster.
> Now, after Intel has enabled ECC on many SKUs from i3 to i9
...which isn't terribly useful given the limited availability of the chipset you need to actually use ECC.
I'd expect AMD's ECC support to be better in this regard.
> The same applies to AVX-512, when AMD has always had the competitive advantage of not
> disabling features on the cheaper SKUs, it would be stupid to have AVX-512 on Epyc and
> disable it on Ryzen, but AMD has avoided until now to say anything about this.
Not saying anything doesn't mean it's not there. Even if they confirmed its availability, it's not like you can get the chip now anyway.
> They have mentioned "new instructions for ML & AI", but that could mean anything. I do not care about "ML
> & AI", which I can do on GPUs
The x86 ML/AI instruction sets would be VNNI, BF16 and AMX, so perhaps that may give you some hint on whether AVX-512 is available or not.
(then again, Intel backported VNNI to AVX, so perhaps tells you nothing)
> It seems that the 15% increase was in the Cinebench ST benchmark.
From where did you get that info?
> The optimists say that maybe AMD has intentionally presented worse results than would be actually obtained
> in the production units, to mislead the competition, but such a strategy does not make much sense.
Worth pointing out that they say "5GHz+", but then show a 5.5GHz processor.
Is that underselling? I don't know. Maybe the 5.5GHz chip was some golden sample and they'll only commit to 5GHz. Or maybe 5.5GHz is what they plan to offer (in a separate interview, they claimed the 5.5GHz processor wasn't overclocked) but are being conservative in their figures, for whatever reason.
So should the ">15%" be associated with the 5.5GHz chip they demoed, or the "5GHz+" mentioned on the same slide?
At the end of the day, you can only really speculate with the small amount of info provided, and you could argue either way, and no-one (non insiders) would be able to tell what is correct.
> While AMD claimed that the 4/3 performance advantage
Don't confuse performance increase with time decrease. 30% less time is different to 30% faster.
> Now, after Intel has enabled ECC on many SKUs from i3 to i9
...which isn't terribly useful given the limited availability of the chipset you need to actually use ECC.
I'd expect AMD's ECC support to be better in this regard.
> The same applies to AVX-512, when AMD has always had the competitive advantage of not
> disabling features on the cheaper SKUs, it would be stupid to have AVX-512 on Epyc and
> disable it on Ryzen, but AMD has avoided until now to say anything about this.
Not saying anything doesn't mean it's not there. Even if they confirmed its availability, it's not like you can get the chip now anyway.
> They have mentioned "new instructions for ML & AI", but that could mean anything. I do not care about "ML
> & AI", which I can do on GPUs
The x86 ML/AI instruction sets would be VNNI, BF16 and AMX, so perhaps that may give you some hint on whether AVX-512 is available or not.
(then again, Intel backported VNNI to AVX, so perhaps tells you nothing)
| Topic | Posted By | Date |
|---|---|---|
| Zen 4 is really bad | Kara | 2022/05/26 01:52 AM |
| Zen 4 is really bad | -.- | 2022/05/26 02:13 AM |
| who is the champ right now? | Michael S | 2022/05/26 03:36 AM |
| who is the champ right now? | Kara | 2022/05/26 04:16 AM |
| Alder lake better than zen4! | Kara | 2022/05/26 04:24 AM |
| Alder lake better than zen4! | Adrian | 2022/05/26 05:40 AM |
| Alder lake better than zen4! | anon2 | 2022/05/26 06:14 PM |
| Alder lake better than zen4! | Adrian | 2022/05/27 04:18 AM |
| Alder lake better than zen4! | Matt Hughes | 2022/05/28 11:15 PM |
| who is the champ right now? | -.- | 2022/05/26 05:23 AM |
| who is the champ right now? | Adrian | 2022/05/26 06:09 AM |
| who is the champ right now? | Andrei F | 2022/05/26 06:16 AM |
| who is the champ right now? | Adrian | 2022/05/26 07:10 AM |
| who is the champ right now? | Andrei F | 2022/05/26 07:37 AM |
| who is the champ right now? | Andrei F | 2022/05/26 07:59 AM |
| who is the champ right now? | Eric Fink | 2022/05/27 12:17 AM |
| who is the champ right now? | Adrian | 2022/05/27 05:41 AM |
| Optimally? | Anon | 2022/05/26 09:43 AM |
| Optimally? | Andrei F | 2022/05/26 12:10 PM |
| who is the champ right now? | -.- | 2022/05/26 06:43 PM |
| who is the champ right now? | Adrian | 2022/05/27 05:24 AM |
| Nothing wrong with anads spec2006 results - see the breakdown | Heikki Kultala | 2022/05/27 03:23 PM |
| Nothing wrong with anads spec2006 results - see the breakdown | Doug S | 2022/05/27 09:59 PM |
| Nothing wrong with anads spec2006 results - see the breakdown | Matt Hughes | 2022/05/28 06:32 PM |
| Nothing wrong with anads spec2006 results - see the breakdown | Michael S | 2022/05/29 01:44 AM |
| Nothing wrong with anads spec2006 results - see the breakdown | Adrian | 2022/05/29 03:53 AM |
| Zen 4 is really bad | Adrian | 2022/05/26 05:16 AM |
| Zen 4 is really bad | -.- | 2022/05/26 05:58 AM |
| Zen 4 test conditions | Adrian | 2022/05/26 06:14 AM |
| Zen 4 test conditions | Groo | 2022/05/26 06:01 PM |
| Zen 4 ECC | Adrian | 2022/05/26 06:33 AM |
| Zen 4 ECC | -.- | 2022/05/26 06:38 PM |
| Zen 4 ECC | Adrian | 2022/05/27 05:11 AM |
| Zen 4 ECC | -.- | 2022/05/27 08:30 PM |
| Zen 4 ECC | Michael S | 2022/05/28 11:19 AM |
| Zen 4 ECC | drAgonear | 2022/05/26 11:58 PM |
| Zen 4 ECC | David Hess | 2022/05/30 07:29 PM |
| Zen 4 is really bad | goose | 2022/05/26 08:14 PM |
| Zen 4 is really bad | Adrian | 2022/05/27 04:44 AM |
| Zen 4 is really bad | James | 2022/05/27 06:13 AM |
| Zen 4 is really bad | John H | 2022/05/26 05:45 AM |
| Zen 4 is really bad | Doug S | 2022/05/26 10:50 AM |
| Zen 4 is really bad | Dummond D. Slow | 2022/05/28 09:14 PM |
| Zen 4 is really bad | Doug S | 2022/05/29 09:49 AM |
| Please check the faxts instead of posting crap like this | Heikki Kultala | 2022/05/26 11:05 AM |
| Zen 4 IPC | ⚛ | 2022/05/26 11:24 AM |
| Zen 4 IPC and AVX-512 | Per Hesselgren | 2022/05/27 01:07 AM |
| Zen 4 IPC and AVX-512 | Per Hesselgren | 2022/05/27 03:44 AM |
| Zen 4 IPC and AVX-512 | Adrian | 2022/05/27 04:57 AM |
| Zen 4 IPC and AVX-512 | anon | 2022/05/27 08:22 AM |
| Zen 4 is really bad | Parsnip | 2022/05/29 04:20 AM |
| What do you mean by 3-wide desing? | Heikki Kultala | 2022/05/29 01:05 PM |
| What do you mean by 3-wide desing? | Doug S | 2022/05/29 02:32 PM |
