AMD's CES presentation was full of Mistakes
Вставка
- Опубліковано 4 січ 2023
- AMD's CES 2023 keynote and the press deck they provided with it, were full of errors and mistakes. Let's take a look at everything wrong with the presentation I was able to find, from the Phoenix APU to Navi 33 GPUs and Zen 4 X3D CPUs.
Follow me on Twitter: / highyieldyt
#ces #ces2023 #amd - Наука та технологія
AMD's marketing has been not good for quite some time. Probably for at least a decade. The Polaris announcement and launch wasn't bad as I recall, but they did things like talk about Crossfire RX480s using less power and being equivalent to a GTX 1080. Neither of which proved true in most cases. But at least the delivery was clean. The CPU launches and announcements have always had some weirdness and missteps. We were willing to overlook those errors because the performance was at or better than their claims and the pricing was excellent compared to the competition. Now, they are continually missing.
Press have always complained about AMD marketing.
These inaccuracies might be a side effect of personnel changes, it definitely needs attention from Lisa Su to ensure accuracy of internal information for decision making and then product launch info.
There may be a need for vagueness prior to launch for competitive reasons
AMD marketing is objectively bad and has been for years. Literally breaking marketing 101 with jebaited, over- and under-representing, inconsistent comparisons, etc.
The "Peter Priciple" is evidently running riot in the graphics marketing division.
This is sad as they're throwing away reputation and good will, hopefully engineering disciplines shall prevail.
This was not deliberate, it was just a number of stupid and inexcusable mistakes. They try to keep specs a secret right up until the last minute which may have caused some of the more technical errors.
If you can design microprocessors you are obviously competent enough. AMD just has to become a little more competent at double checking their PowerPoint slides.
@@ramr7051 Agreed. My wife is a Marketing Manager and her department is always struggling to get the tech guys to give up to date information and the tech guys also never seem to want to check what is going to be published. That been said, AMD are such a big company that so many errors should not have slipped through.
Once I found AMD wrote their company name as "AME" in a product introduction on their official website.
I assume this happens because these companies try to keep such details secret until the last minute even from their own marketing teams. Anyway, sanity checking would make a lot of sense...😊
There was a whoopsie of the Memory bus being 12800 or something lol.
I also don't understand the madness of wanting to announce these kinds of things immediately after the December holidays like geez give your own people a month to figure it out or something. CES just seems like such a bad idea for any company that wants to remain professional. Proofreading/testing should take precedence over "deadlines". Like why show up to a conference like this, if you're just going to embarrass your company? Just don't go.
It is still unacceptable to make such mistakes at an official presentation.
Dang. AMD is embarrassing itself very often. How come move of the larger channels catch these kinds of errors by the way or point at them? Glad you did a video about it.
Gamers Nexus made fun of the 1280bit memory bit bus in theirs, but didn't catch/mention a lot of the errors.
I wonder what is going on internally at AMD currently. Some things def. seem out of whack atm.
It's probably just the marketing department. The Instinct Mi-300 shows that they can deliver increasingly complex products.
The company more than doubled in size in the last year with the acquisition of Xilinx and Pensando
My brain read teraflops as CUs as they were 32 and 28 and it made sense. That's a big mistake...
Your brain is like auto correct :p
@@HighYield Well, after having consumed rumors for fun for two years, one tends to get distracted.
I am not blowing smoke, when I say that you are more knowledgeable and smart, than every other tech youtuber that I am subscribed to. That is saying alot because, I am subbed to LTT. and Jaystwocents, and Gamers nexus. I find that your information is so much more informative than everyone else. Thank you for that.
Yes his stuff is very insightful, definitely agree on that
Yes, the asymetrical design of Zen 4 X3D! Am I to suppose I'm not the only one who thinks that the Ryzen 9 parts actually have one CCD with 3D cache, and one without, and only the one without can have the max frequencies, while the other one behaves like the one in R7 7800 X3D ? To me that's the only way to explain why the R9 parts don't get more cache, while only the R7 part has reduced max frequency.
Also, I've observed the TFlops mistake when I saw that slide the first time, but the others... daamn, they were MANY! Good job on spotting and announcing them!
This was confirmed in an interview for PCWorld.
@@alexg9155 Nice, thanks!
The 7040 mobile chip was their best product announced. The 3d vcache chips will be interesting to see in action as amd's take on big-little architecture, similar to some mobile phone chips, differentiating cores/ clusters via cache size.
The mi300 for super computers looks to be amazing tech, which has unbelievable potential for industries in need of such computing power.
3D vcache is a game changer. All the 3D chips so far have been awesome. Also how does a huge company like AMD not hire someone to go over the slides just to check? Very odd indeed.
How does a multi-billion dollar corporation not have the money to higher a team or a person that double checks their stuff before publishing? I'm not an AMD hater, but this is kind of ridiculous. Their engineers are amazing, but the marketing team needs restructuring.
They could if those people were not on holiday as was half of the company during the holiday periods :D
Bad planning imho. If you have restricted resources unlike nVidia, then you need to plan better. RDNA3 should have come 1 month before and then there would be more time to focus on this. Or both should have been delayed to next week. And then both the slides would be correct and no vapor chamber issues on RDNA3. But the cost would be missing holiday sales and CES :P
@@marsovac Yeah I'm of the same mind, these dumb cultural notions around trying to do everything around the holidays instead of just taking a freaking break is causing them to deadline crunch and make all sorts of mistakes. They should just be content to wait till February or whatever, so they can figure their stuff out. Not like it would really make a big difference anyway since none of these parts are going to be available for consumers to buy in January anyway.
32 CUs at 2 Ghz can do 8.19 TFLOPs of FP32 w/ single issue (like RDNA2) with a hypothetical max. throughput of 16.38 TFLOPS if dual issue could be used all the time.
Since mobile RDNA3 is built on 6nm instead of 5nm, it actually has a chance to have a better physical design than (or not inherit all issues of) Navi31.
Thus, I'm guessing best case 7700S yields a bit more performance than a 6800S (also a 100W chip w/ 32 CUs). It will likely depend on whether AMD can actually get higher clockspeeds out of that new design.
Extracting more performance out of the same 128bit memory interface will likely depend on how well the additional VGPRs and the doubled L0 and L1 caches (compared to RDNA2) can contribute.
The same 32CU chip seems to be labled as 7600M XT in one slide and as a 7700S in another. Maybe they're different in max pwr draw.
6800s was also the 6650m xt. Only difference was what they were put in. 7700s should be in thin and lights and 7600m xt in thicker laptops. But the a15 from Asus is getting a 7700s even though its watts are equal to a 7600m xt. AMD naming is an utter trainwreck with laptops.
I really hope with driver increases they can up performance with dual shaders and offer a budget competitor to thin and light AMD 4060 and 4070. 6800s in the g14 did a good job at competing with blade 14 for far less money.
"We apologize again for the fault in the subtitles. Those responsible for sacking the people who have just been sacked have been sacked."
I was wondering the same thing about the caches and was very confused because if the phoenix 6 and 8 core parts have same caches as desktop counterparts then whats the point of the dragon range 6 and 8 core parts because on paper they are inferior with 5nm and lower base clocks compared to phoenix's 4nm and higher base clocks.
With that said I'm still interested to see how phoenix and dragon range will perform in various tasks including games for the similar core parts although now we know phoenix has less cache but higher base clocks.
The AM5/ZEN4 and RDNA3 launches seem rushed. Just take a look at the non final base clock of the 7800X3D.
Noticed the cache issue to and at first thought maybe the igpu got a bigger cache. On AMDs website the right specs are listed.
Thanks for the info! 🙂
It's not just failures in delivering real facts of presented products, which take them 2years to develop, but also product testing probably doesn't reveal such fatal flaws like not enough cooling liquid in reference gpu cooler resulting in overheating and throttling chip. (RX7900xtx story).
Yeah, finally AMD that extra cache. Already curious for performance difference if games start to utilize more cores than those on single CCD.
(Just reminds me my 6650xt xt cooling troubles left on customer to either live with unneeded spins up of fans or set own fan curves or undervolt gpu to reach user friendly noise. Creeping temps when gpu chip low duty but memory high clocked. Blame AIB or AMD, hard to judge by myself).
well i was waiting for mid rang GPU in this showcase do you know when they likely show it
Well , basically only two things are important with the launch of a new Product , the Performance and the Pricing . Press Deck informations often have errors , it was not the Lauch , it was the announcement of the Launch . In Intels Press deck informations there was constantly an announcement of Intels 10 nm since 2017 which finally was aviable in 2021.
Maybe there will be V-Cache variants in mobile as well , at least for Gaming Laptops , V-Cache dont need much additional Power and the clock speeds are mostly not as high as in Desktop Space = it would be perfect for an V-Cache variant . The 5800X3D has low clock Speeds , low Power consumption , high Gaming Performance and could be easily an 65W TDP mobile CPU
It looks like AMD has got such a huge list of announcement + so many new the things after really long time... Which made the guy(who was assigned to compile them to slide) confused!!
I think we are feeling the absence of Robert Hallock. He was responsible for things like this.
Yes, when it comes to CPUs.
Lol when I saw the "FP32 TFLOPS" number on the RX 7000S slide I automatically assumed it to be 32 Compute Units similar to those rumors🤣
Heck....so many big announcements at CES but not a single word on desktop APUs?!
The last desktop class APU from AMD was launched in April 2021.
Don't quote me but I thought amd was moving away from apus as all ryzen 7000 cpus now have integrated graphics? I could be wrong tho.
@CannedCoochie I love APUs. APUs still have the potential. Laptop APUs have lower TDP and are usually combined with a dGPU.
Hope AMD might launch Dragon Range desktop level RDNA3 APUs with much higher TDP sufficient enough for a smooth 1080p 60 fps gaming.
@@thtupid well Phoenix is an APU. what AMD is trying to do is include at least very basic graphics in everything to remain more competitive against Intel. But they still have CPUs with very powerful integrated graphics, which are usually called APUs. There's some 6000 series APUs out there right now, but AMD seems to have stopped giving them the attention they got back in the 3400G days.
@@ramr7051 The problem with APUs is their low competitiveness. A CPU + dGPU is generally a much better performing option for the same money, than a big AM5 APU die that also needs expensive RAM and motherboard. But APUs make sense for thin laptops, which is where AMD are using them first.
@CannedCoochie It was, for some previous APUs- but now we are talking about AM5 APUs, and here is an example of what you could get: R5 5600/13100F €130+ B series mobo €100 + 16GB DDR4 €50 + RX6600 for ~€300 = €580, or something like
Phoenix APU '€359' (using 5700G launch price, although 5700G was way cheaper to make than N4 Phoenix) + €180 B650 + €80 cheapest, but slow DDR5= €620. The first option is twice the performance, and is more futureproof too.
Between all the partner talks, why would they have time for highlighting their products correctly
When a multibillion dollar company cares less about their powerpoint presentation than you in 3rd grade.
Wouldn't it improve performance when getting rid of the memory hirachie L3>L2>L1 to just a huge L1 full speed cache? I know it is not that easy as wirting it down. But the 3D-Cahe approach could be a solution to make it possible!
What do you think?
The larger the cache, the slower. That’s why cache increases in size with higher cache levels, that’s why L1, L2 & L3 exist in the first place.
Larger caches have more latency because distance increases. Increasing a cache size x4 multiplies the maximum distance x2 if the design doesn't change. The way to avoid this problem is to have different cache hierarchies.
But something like you suggested was done on some IBM processors, with a L2 cache that acts also as shared L3 cache with other cores. If AMD manages to combine this kind of cache with 3D cache it would be interesting, as 3D cache would potentially improve also L2 cache.
@@6SoulHunter9 my initial thought was that the higher hierarchy in caches came with the multicore CPUs and therefore that the interconnection between those hierarchies can't be run with full CPU speeds.
My idea was to get rid of the 2nd & 3rd level on the CPU cores and connect a ccd-wide large 2nd level directly available for all cores on the CCD.
Or simply put every lookup of data in a cache needs work cycles and also need work cycles to load data from the 3rd level > to the second and > 1st aso.
The larger the Cache+the fewer steps in the hierarchy the more productive IPC!
But OK, at larger sizes you get longer data traces and a higher latency.
@@oscarbecker7843 That's probably part of their role in modern designs, but L1 and L2 caches existed before multicore processors, and I think that perhaps L3 too. As far as I know, L2 cache was invented when misses were too many but making a larger L1 cache regressed performance because latency. And perhaps L3 cache too, with L2 cache. There were CPUs even with L4 cache.
I noticed the same thing. Kinda frustrating
1280bit meme interface was my favourite :)
Having same amount of cache mobile CPUs will help a lot with gaming performance.
I am talking about dragon range, not phoenix point for those who are wondering.
Wieder mal top Video, sehr informativ und schön dargestellt!
Again, an excellent video, very informative and well presented!
Dankeschön!
the 6c 38M might be possible if the 8c CCD has 2 cores disabled, but L3 still enabled. hence 8x4 L3+ 6x1L2= 32 + 6= 38. Not saying this is what they did, just guessing at what might be possible.
Intel had some X cores full die, only 4 or 8 cores active, but full L3.
Yikes. I don't trust any info about the 3d vcache chips either. AMD has zero confidence lately and they have been leaning way too hard on misleading marketing to get by.
I think AMD needs to hire a new marketing department because they mislead more with every conference :/
May be intel is bribing, cant skip that, cause, intel plays every dirty role if there is a gap, Im not saying the slide mistakes, but their marketing team is a flop. Its the history of their ancestors business strategy from California to the country on the other side of the sea.
Having worked on this directly I can say it isn't an attempt at spreading misinformation. These things literally change on the fly. Someone probably gathered info at a point in design and release different from the final form. Considering the alternative is basically to go dark and only release info when it is at a final point, I don't think these things are that big a deal. If you want sneak peeks which help make informed consumer decisions then less than perfect information is that tradeoff.
Lol the 1280bit memory bus would have been amazing. Have to wonder what the point of going to these presentations are if they're just going to crap themselves with errors and false information.
Seems like AMD has been smoking some serious dope recently...
AMD too much holdback their product,
They have canche to release 3D v cache cpu at beginning
They have canche to release GPU with more than 96 CUs
But they didnt do that
Half the company on vacations for christmas and new year do not help :)
They are still coming back and need to regurgitate about the the vapor chamber debacle :P
Rushed launch to meet the holiday season and then half the company unavailable...
I get that they do not have the money and resources nVidia has, but then plan accordingly.
Where the desktop APUs at? :'(
ALL desktop CPUs are APUs in this generation from AMD.
@@LeonardTavast you cant call then apu with 2cu
They would not be competitive on the market due to high APU manufacturing, DDR5, and AM5 mobo cost. On desktop, a cheap CPU + dGPU is the better (and saner) option.
That happens when even the own staff gets confused about the new naming scheme. They just should keep it simple.
The AMD 7600M is around ~20.3 FP32 TFLOPS
The 780M is around ~8.9 FP32 TFLOPS
AMD is using a more enhance version of the 6nm, specially in chaplet design to save more energy and give out more performance.
Navi33 is monolithic, not chiplet based.
no 7600X3D for 299 was a big mistake in it if you ask me.
The interesting thing is, AMD has 6-core chiplets with 3D V-Cache on top, thats what they must be using for the 7900X3D (I dont think its a 8+4 design). So they could quiet easily release a 7600X3D if they wanted to.
I wonder what happens to a designated 7900X3D, where the non-3D chiplet fails QC. We've seen 7600X being sold with two chiplets under the heat speader before.
@@HighYield im pretty sure the 7900x3d is 8+4
If its 6+6 it would perform worse then the other 2 cpus in all games that use over 6 cores due to having 2 cores less on the vcache chiplet
Don't worry, you are not going to be disappointed.
Regardless, I'm pretty stoked for 7040 and 7045HX
And for CDNA 3 "5x AI Perfect/W" they compared FP16 on MI250X vs. FP8 on MI300, which is apples to oranges. AMD marketing strikes again...
The presentation was 90% of guess speaker, lol. The NASA speech was the best. I learned that "having dreams" will 100% lead you to success. We all will travel to Mars pretty soon if we all believe in the power of dream.
Joking aside, the laziness from AMD clearly shows with this presentation. I guess this make make AMD look even more lazier not to mention the recent RMA issue with their faulty GPU,
I don't know about the dismissal of the entire #AMD marketing team, but their head of marketing should definitely leave his position.
How many more times?
I'm kind if suprised AMD messed something up............
they are waiting for the announcement of intels prices, then they are going to set the prices right
wow
well i don't think they cared. 😂
Also heard that there may be problems with phoenix's graphics, which i hope they sort out
Because of that one bar graph comparing it to the alder lake chip? Probably a mistake aswell.
@@Knorrkator according to Moore's law is dead, some sources told him AMD was struggling to get the new rdna3 cores to play nice. Mostly driver issues it seems, but it can get serious
Lets hope not, RDNA3 was supposed to save us :(
Phoenix is only for premium, not affordable $500 laptops....my disappointment is immesuarable. RTX 3050 or 4050 laptop seems the only solution but they lack VRAM so much(( also hoping for 7600m laptops to cost less than 1k
Even Rembrandt didn't appear in $500 laptops. I got my Ryzen 7 6800H laptop without a dGPU for $1050 including shipping from China in June, although it has nice specs. RTX 4050 has 6GB of VRAM, although it's slower due to a 96-bit bus. So our only hope is to see Zen 3/3+ laptops with Radeon 7600M that can be cheap. Or just buy cheap discounted Ryzen 7 6800H laptops.
@@one_step_sideways да пздц ваще
Something tells me Ryzen 6000 will be the more budget option instead... But seeing they're dragging out Vega and calling it Ryzen 7000, I'm concerned
@RAMR You forgot that they now have a new naming convention for the mobile CPUs. Desktop CPUs have the same old naming convention to avoid confusion whereas the mobile APUs have a different convention because so many models are produced for various price ranges.
What matters are the last two digits. The first digit shows the year when the APU was introduced (7 = 2023, 8 = 2024 and so on), the second digit shows the model of the CPU (Ryzen 3, 5, 7 or 9), the third digit shows the architecture and the fourth shows the performance level. The higher, the better for all digits.
Ryzen 7045 series - Dragon Range CPUs with Zen 4 desktop chiplets at 5nm for extreme gaming laptops with powerful graphics and strong cooling systems
Ryzen 7040 series - Phoenix Point APUs with Zen 4/4+ laptop dies at 4nm for premium laptops that need the best performance and power efficiency and cutting edge features and can do without a discrete GPU with their fast iGPU
Ryzen 7035 series - the same as Ryzen 6000 series
Ryzen 7030 series - the same as Ryzen 5 5400/5600/5800U/HS but with slightly better boost clocks
Ryzen 7020 series - Mendocino APUs with Zen 2 laptop dies at 6nm for entry level laptops for average Joes that don't need a fancy processor to get their Office and internet browsing needs done
Biggest mistake is buying them
Someones gonna get fired......
I don't get how AMD can do such basic, fundamental mistakes... if they can't get their slides right what makes everyone think they have such great products... im not buying a product from them again until they've proven themselves.
AMD BEST
Every time regarding their products, AMD and Nvidia lie. Not right, is it? then, after being deceived, we purchase from them at launch.
I would never take them at their word and would always examine video reviews first.
Sabotage?
Amdisappointing
Just as usual : )
One thing Lisa Su failed is that she didn't held the related product managers accountable for not reviewing the slides to ensure the accuracy; in the past I personally sent emails a few times to the PR contacts asking for forwarding my emails to Lisa Su but got no reply, so arrogant!
20-25% RDNA3 generational increase in cherry picked titles is a joke. They were promising more then 50% performance / W, which is especially noticeable with mobile gpus. After this presentation, I'm now convinced that the issues with RDNA3 run deeper then we thought and all GPUs are affected. Also, they presented just one chip - Navi 33, again - no competition with NVIDIA at the top. The performance of the desktop Navi 33 will be very similar, which is probably why they delay these products. Their price will be at least 25% more, for 25% more performance... Makes no sense. Better not release them at all, and try to fix RDNA3...
Navi33 is stilll 6nm, and rather cheap due to its 204mm² size. I wouldn't expect a big price increase, but also not a big performance uplift. Maybe something like 7700S having 5%-10% more FPS than 6800S, while costing approx. the same.
That's at least what the raw specs suggest and depends on achievable clockspeeds and how well the incrased cache sizes can actually compensate for the same 128bit memory bus.
Can't they get sued for crap like this?
Intel is dead, they cant compete with AMD, any more,16 p core AMD with large cache pool, small 4nm ,will take shhht out of intel for 4-5 years, But Im not an expert, but assume it goes some what similar to this, as a rough way, 10nm node will give a an area of 100 square unit of a side of 10 units. but 4nm will only make up to 16 square units. so its 6 times smaller, can pack much more transistors and including cache in a unit area. thats with this gen AMD ultra books can give above 30hour video playback due to efficiency. Intel 10+++ cant save intel, only in games .My 65whr 11800h will only gives 2.5 hour in regular use, in balanced mode when unboxed, with extra battery saver mode it goes to 3.5 hour. 8nm rtx 3000 is a poorly clocked, inefficient chip it will be performing equal to 12nm tsmc.
Clown AMD strikes again!!!!
I pennywise nvidia even more