');$('.tpu-fancybox-wrap').css('maxWidth', maxWidth);*/instance.$refs.stage.on('transitionend', function() {updateButtonPos(instance);});},onUpdate: updateButtonPos,afterShow: function(instance, slide) {updateButtonPos(instance);instance.$refs.inner.find('.fancybox-tpu-nav').show();},beforeClose: function(instance, slide) {instance.$refs.inner.find('.fancybox-tpu-nav').hide();},afterClose: function(instance, slide) {$('.tpu-fancybox-wrap').contents().unwrap();$('body').removeClass('tpu-fancybox-body-wrap')},baseTpl: '
',});});}loadjs.ready(['jquery', 'fancybox', 'swiper'], function() {attachLightbox('a[data-fancybox]');if ($(window).width()<600) {$('.imgcontainer').each(function() {var $this=$(this);if (($this.find('a').length==1) || ($this.find('a').length>7))return;$this.addClass('swiper-container');$this.find('a').addClass('swiper-slide').css('width', 'auto').wrapAll('
');new Swiper ($this.eq(0), { slidesPerView: 'auto', slidesPerGroup: 1, spaceBetween: 15, pagination: { el: '.swiper-pagination', clickable: true } });});}$('.newspost').on('click', '.spoiler > .button, .spoiler > a', function(e) {e.preventDefault();$(this).next('div').toggle();});$('.newspost').on('click', '.ispoiler', function(e) {e.preventDefault();$(this).find('div').css('filter', '');$(this).removeClass('ispoiler');});$('.contnt').on('click', '.newspoll_btn', function() {popup.Show('TechPowerUp Quick Poll','Loading...');$.get('/news-poll/options?id='+$(this).data('id'), function(data) {$('#popup_content').html(data);});});});
Monday, June 3rd 2024
Computexby btarunrDiscuss (16 Comments)
AMD today launched its Ryzen AI 300 series mobile processors, codenamed "Strix Point." These chips implement a combination of the AMD "Zen 5" microarchitecture for the CPU cores, the XDNA 2 architecture for its powerful new NPU, and the RDNA 3+ graphics architecture for its 33% faster iGPU. The new "Zen 5" microarchitecture provides a 16% generational IPC uplift over "Zen 4" on the backs of several front-end enhancements, wider execution pipelines, more intra core bandwidth, and a revamped FPU that doubles performance of AI and AVX-512 workloads. AMD didn't go in-depth with the microarchitecture, but the broad points of "Zen 5" are detailed in our article for the Ryzen 9000 "Granite Ridge" desktop processors. Not only is AMD using these faster "Zen 5" CPU cores, but also increased the CPU core count by 50%, for a maximum of 12-core/24-thread.
The "Strix Point" monolithic silicon is built on the 4 nm foundry node, and packs a CPU core complex (CCX) with 12 CPU cores, four of these are "Zen 5," which can achieve the highest possible boost frequencies, the other eight are "Zen 5c" cores that feature an identical IPC and the full ISA, including support for SMT; but don't boost as high as the "Zen 5" cores. AMD is claiming a productivity performance increase ranging between 4% and 73% for its top model based in the series, when compared to Intel's Core Ultra 9 185H "Meteor Lake" processor. The iGPU sees its compute unit (CU) count go all the way up to 16 from 12 in the previous generation, and this yields a claimed 33% increase in iGPU gaming performance compared to the integrated Arc graphics of the Core Ultra 9 185H. Lastly, the XDNA 2 NPU sees more that triple the AI inference performance to 50 AI TOPS, compared to the 16 TOPS of the Ryzen 8040 "Hawk Point" processor, and 12 TOPS of Core Ultra "Meteor Lake." This makes the processor meet Microsoft's Copilot+ AI PC requirements.
With this generation, AMD has revamped the processor naming scheme. Ryzen AI is the main brand, followed by a brand level (5, 7, 9, and form-factor U, H, HX); the generation (in this case, 3rd gen NPU-equipped Ryzen AI); and model number. Today, AMD is launching two processor models, the Ryzen AI 9 HX 370, and the Ryzen AI 9 365.
The HX 370 maxes out the silicon, featuring its full 12-core/24-thread CPU, its full 16 CU iGPU, and its XDNA 2 NPU with 50 TOPS on tap. The CPU cores boost up to 5.10 GHz, with a base frequency of 2.00 GHz. The chip's TDP is 28 W, but configurable between 15 W and 54 W by the OEM. The maxed out iGPU is branded the Radeon 890M Graphics, and comes with a maximum engine boost frequency of 2.90 GHz. The next processor model in the series if the Ryzen AI 9 365. This chip is configured with a 10-core/20-thread CPU with 2.00 GHz base- and 5.00 GHz boost frequencies; the Radeon 880M iGPU with 12 CU running at 2.90 GHz engine boost; and the same 50 AI TOPS NPU. It also has the same TDP numbers as the HX 370.
The new XDNA 2 NPU not only offers an over 3-fold increase in AI inferencing performance over the previous generation, but also support for the new Block FP16 datatype, which offers the performance of 8-bit with the precision of 16-bit. Since most AI applications use 16-bit, no quantization is needed.
Notebooks powered by the Ryzen AI 300 series are announcing at Computex.
Related News
- Tags:
- 10-core
- 12-core
- 4 nm
- AI
- AI PC
- AMD
- AVX-512
- Copilot+
- Core Ultra 9
- Granite Ridge
- iGPU
- RDNA 3+
- Ryzen AI
- Strix Point
- Zen 5
- Zen 5c
- Feb 5th 2024 AMD Readies X870E Chipset to Launch Alongside First Ryzen 9000 "Granite Ridge" CPUs (216)
- Jan 26th 2024 More AMD Ryzen 9000 "Zen 5" Desktop Processor Details Emerge (85)
- Jan 29th 2024 Top AMD RDNA4 Part Could Offer RX 7900 XTX Performance at Half its Price and Lower Power (292)
- Apr 19th 2024 AMD "Strix Halo" Zen 5 Mobile Processor Pictured: Chiplet-based, Uses 256-bit LPDDR5X (109)
- Apr 9th 2020 AMD 4th Gen Ryzen Desktop Processors to Launch Around September 2020 (56)
- Jun 3rd 2024 AMD Zen 5 Storms into Gaming Desktops with Ryzen 9000 "Granite Ridge" Processors (100)
- Jun 3rd 2024 AMD Outs Ryzen 5000XT Processors for Socket AM4, an 8-year Old Socket (81)
- Apr 8th 2024 AMD Ryzen 9000 "Granite Ridge" Zen 5 Processor Pictured (31)
- Apr 25th 2024 AMD "Strix Point" Mobile Processor Confirmed 12-core/24-thread, But Misses Out on PCIe Gen 5 (42)
- May 29th 2024 AMD Ryzen 9000 Zen 5 Single Thread Performance at 5.80 GHz Found 19% Over Zen 4 (92)
It's good to see they put more effort into the iGPU, as Intel has caught up and even passed the 780 with Meteor Lake. They definitely can't phone it in there anymore.
Darmok N JaladIntel has caught up and even passed the 780 with Meteor Lake.
I keep hearing this getting parroted around.
Meteor Lake has caught up with 780M in the way that Vega caught up with Pascal.
Years late and pulls signifcantly more energy to surpass it in 3Dmark but not real games.
Also the problem with iGPU performance has always been memory bandwidth.
I still can't believe they went with such naming scheme...
DimitrimanI still can't believe they went with such naming scheme...
AMD Marketing never misses an opportunity to shoot themselves in the foot.
They pick the worst possible timing to associate themselves with Intel. Who recently degraded their cpus and reminds everyone that previous fk up on AM5.
Also everyone I know hated that Intel naming scheme.
Note that AI Responsiveness chart is "Time to first token" i.e. prompt processing time, not token generation rate which I understand is memory bandwidth bound. Sneaky.
Though given just how much system prompt and context such local LLM deployments would probably require, I suppose it is still a useful bench.
This AI bandwagon is now super dense crush load with latecomers climbing on top scary-close to overhead lines. The APU with 16 CU is much more exciting.
no1 cares about ai. what a f*cking waste.
Pushing the mixed cores on a segment where they don’t have the People or market share to iron out the bugs is… bold
It seems like just yesterday that the highest iGPU went from 8 to 12 CUs. Now we are on 16 CUs with a 40 CU part on the horizon.
Forget about AI and enjoy the SoC goodness.
ZubasaI keep hearing this getting parroted around.
Meteor Lake has caught up with 780M in the way that Vega caught up with Pascal.
Years late and pulls signifcantly more energy to surpass it in 3Dmark but not real games.
Also the problem with iGPU performance has always been memory bandwidth.
I agree with Zubasa. Handheld gaming devices using Z1 Extreme have been benchmarked against those with Intels fastest iGPU and the AMD iGPU is much faster and runs cooler.
Darmok N JaladIt's good to see they put more effort into the iGPU, as Intel has caught up and even passed the 780 with Meteor Lake. They definitely can't phone it in there anymore.
With a TDP above 30w, yes, in many cases it's competitive. The comparison was made with both at 45w.
Then, I suppose it's faster than the 7840u by a similar margin...
kondaminPushing the mixed cores on a segment where they don’t have the People or market share to iron out the bugs is… bold
What problems? The zen4/c are functionally the same on z1 chips.
DimitrimanI still can't believe they went with such naming scheme...
1 - What's in a name.
2 - If it generates 0.6 % lower sales by omitting the AI part, why would they.
3 - I've seen that string size before.
kondaminPushing the mixed cores on a segment where they don’t have the People or market share to iron out the bugs is… bold
What do you mean? Phoenix 2 came out A YEAR AGO. Where are the complaints?
Edit:
Strix point doesn't support SODIMM. I knew this was true for Strix halo, but this is news to me at least.
Source, 3 minutes.
They pretty much have to go to soldered RAM if they want the faster RAM speeds, right? I guess that's the cost of feeding a beefier iGPU. I know there's that new memory standard, but I have my doubts about its adoption now that soldered RAM is so commonplace.
+1 it is nice to see the iGPU get some love. As for naming...
Selayano1 cares about ai. what a f*cking waste.
Maybe not in the context of Microsoft's copilot total BS, but clearly you don't do photo or video processing. More and more AI photo software is leveraging GPU and now NPU. On Apples with M4 chip, Topaz software runs much faster. I'm actually annoyed desktop 9000 series isn't getting NPU unlike the APU's. I have to rely on my discrete GPU which is older and not particularly fast for AI and the desktop iGPU is piss weak in Zen 4 and Zen 5 compared to the laptop APU. Arrow lake is getting the same NPU as Lunar Lake, so I'm definitely waiting for comparisons to Zen 5 before committing to my new PC build.
I hate saying "AI" in any context other than science-fiction (because what we have now is not effing AI), but now I'll have to use it because it's right in the product name. Thanks, AMD! :shadedshu:
AusWolfI hate saying "AI" in any context other than science-fiction (because what we have now is not effing AI), but now I'll have to use it because it's right in the product name. Thanks, AMD! :shadedshu:
It is a common trope that "AI" is something that don't yet exist, yet one cannot get around the 'net without stumbling over LLM-generated SEO trash these days, sometimes even on first result of more obscure topics. Simple Markov chain word salad, the usual stuff before LLM, won't usually end up on first page.
At current pace, humanity itself might not even get AGI, before shaking itself apart with all these new "toys".
Back on topic, wonder how long would it take, before someone does a comparative benchmark between this and the equivalent 4C/8c part from Intel. Should be fun.
