Spokeshave -

It looks like it will all come down to frequency. If even the 67% IPC advantage remains in a 2Ghz Hammer, it will compare favorably to a 3.34 GHz P4.

If the P4 is running slow Dram over a 400MHz FSB and without iHT enabled. I wouldn't count on that. That's pretty much the configuration that AMD compared itself to when making their performance claims so long ago. Unfortunately the target moved. As the weeks, months and now years go by there's still no Hammer.

EP

Spokeshave -

I've made a mistake. I compared FP to INT scores. Once I get turned around in the right direction I see the Hammer compares with a 2.0GHz P4, not a 1.8GHz as I reported.

Sorry,

EP

spokeshave -

Better work on them readin' skills a bit. The comparison was to a 2.2GHz P4.

I know it was compared to a 2.2GHz P4 on that webpage but that P4 had slow (2100) memory compared to the Hammer 2700 memory. My post stands. The Hammer SPEC scores compare to a P4/Xeon at 1.8GHz, at least a 1.8GHz with RamBus

http://www.specbench.org/osg/cpu2000/results/res2002q4/cpu2000-20020923-01662.html

EP

More Hammer benchmarks -

http://www.heise.de/ct/english/02/26/018/

Still unable to get above 1.2GHz and compares to a 1.8GHz P4 in SPEC.

EP


Test Results
CPU AMD Athlon 64 AMD Athlon XP Pentium 4
Clock Frequency 1.2 GHz 1.2 GHz 2.2 GHz
L1 Cache, Data+Code 64+64 kbyte 64+64 kbyte 8 kbyte+12kµops
L2 Cache 1 mbyte 256 kbyte 512 kbyte
L2-Cache Strategy Unified Exclusive Unified Exclusive Unified Inclusive
L2-Cache Associativity 16-fold 16-fold 8-fold
L1-/L2 Cache Line Size 64/64 byte 64/64 byte 64/128 byte
Memory System PC2700 PC2700 PC2100
Mean Latency 87 ns 112 ns 227 ns
Memory Bandwidth 2306 mbyte/s 2041 mbyte/s 2080 mbyte/s
effc.32-bit Reading Speed 1922 mbyte/s 1250 mbyte/s 1888 mbyte/s
Game Benchmarks
3DMark 2001, 800 x 600, HW-T&L
3DMarks 8753 7601 9008
Car Chase L/H 133.3/50.0 fps 118.6/38.6 fps 132.6/49.0 fps
Dragothic L/H 122.3/66.3 fps 115.4/62.7 fps 149.6/78.1 fps
Lobby L/H 145.0/71.0 fps 115.1/52.9 fps 135.6/63.1 fps
Nature 50.1 fps 51.3 fps 51.2 fps
3DMark 2001, 800 x 600, SW-T&L
3Dmarks 5080 4113 5214
Car Chase L/H 64.1/24.7 fps 54.2/20.4 fps 72.0/26.2 fps
Dragothic L/H 60.7/29.9 fps 45.7/24.9 fps 53.7/31.0 fps
Lobby L/H 106.0/45.8 fps 77.8/35.3 fps 98.4/43.2 fps
Nature 38.2 fps 36.2 fps 48.3 fps
Quake III
High Quality/1024 x 768 225.3/188.5 fps 172.4/159.6 fps 218.9/187.0 fps
Aquamark 1024 x 768 x 16 49.5 fps 48.3 fps 51.5 fps
Commance 4 1024 x 768 x 32 36.6 fps 28.4 fps 36.3 fps
Application Benchmarks
Cinebench 2000
Shading C4D/GL,GL Factor 18.8/27.4 CB/1.73x 12.4/22.8 CB/1.84x 17.2/27.7 CB/1.62x
Raytracing 20.4 CB 16.9 CB 22.7 CB
PovRay 3.5 251 PPS 242 PPS 333 PPS
PovRay 3.1 1037 PPS 948 PPS 1219 PPS
Video Transcoding
DivX 5.021 569 s 709 s 497 s
Xvid2 750 s 929 s 717 s
Bapco SysMark 2002 178 SysMark 2002 145 SysMark 2002 217 SysMark 2002
Internet/Office 228/139 172/122 238/198
Linux Kernel Bench 2.4
6224 points 161 s
4500 points 222 s 166 s
6049 points
c 't SSE Mandelbrot Set3 174.0 172.5 345.1
c 't MemSpec Rating 775 438 614
SPECfp_base 20004 674 504 677
SPECint_base 20004 739 532 774

1 DivX 5.02, 870 kps, other settings :default,128 kbps Lame MP3
2 Xvid,870 kps, B-Frame 3, other settings:default,128 kbps Lame MP3
3 Value in millions of iterations per second
4 c 't measurements for the Athlon 64 estimated in accordance with SPEC rules

borusa -

If I didn't know better I would think you are bitter.

What does it matter? I don't run a public company and the money I make or lose is mine alone. I can hate a company and not violate any Fiduciary responsibility to anyone else. I'm short the AMD Jan $5 Puts and my Dec $5 Puts just expired yesterday. I'll short the Feb $5s tomorrow. I'll buy AMD shares and write CCs. Probably Jan $7.5s

How about you?

EP

Intel will have to fight Via on antitrust allegations

Via wins appeal against Intel in England

By Mike Magee: Sunday 22 December 2002, 09:31


A STATEMENT from Via today said that the English Court of Appeal has ruled on an element of its long legal battle with Intel and admitted claims the Taiwanese firm made alleging antitrust actions by the chip giant.
This slab of a complicated set of legal proceedings relates to a law suit Intel started last year in this country claiming Via's CPUs and chipsets infringed five Intel patents.

Via said that one element of its defence to this charge was that Intel has acted anti-competitively in both processor and chipset markets.

But, in April, Intel applied to m'learned fiends in the English High Court for summary dismissal of the case.

A judge granted that, but Via appealed against the decision.

Via said that the Court of Appeal upheld its appeal in its entirety.

The judgement said: "If what VIA alleges is made out then the dominance of Intel in the world-wide market in the important and developing technology for PCs is buttressed by its enormous portfolio of patents and restrictive licensing policy. This may be true of others but in the case of this technology the commercial and technical requirement for compatibility of hardware and software confers on the industry leader an even more impregnable position."

Overturning a ruling by Mr Justice Lawrence Collins, the Appeal Court said the proceedings raised "momentous issues".

The five infringement claims will now be heard at a full trial, while Intel is being forced to pay Via's legal costs for both the High Court and Court of Appeal hearings so far.

Via said it would press for a swift trial of Intel's allegation here in Britain. µ

borusa -

As far as me being serious, I guess it seems a bit self serving of you in my eyes for you to characterize AMD as some type of pathogenic demonic...etc. I guess it is tough for me to see Intel as a victim and all.

I see you misread my post. I did not say AMD was pathological, I said Jerry was. I'm sure over 99.9% of AMD employees are fine hard working etc etc but they get their orders from the top. The business plan is defined at the top and the company carries it out. How many years in a row was the Board of Directors voted among the 10 worst in the country? How the future plays out is yet to be seen but I think my description does the best job of describing the past and for years that perspective was the best predictor of AMD's direction. It may change and AMD may pursue profitability. I happen to be a shareholder in AMD and also write puts at this level. I know there are enough people out there who simply believe any line of BS the company puts out. If you're one of them then thanks for holding the share price up.

EP

borusa -

I had no idea. Thanks for setting me straight.

Well I don't know if you're serious or not. Don't take anyone's word for anything. Do your own research.

Whats your take on this?

What's your's?

wbmw -

How many can you name?

I've already shot my wad...

boruse -

How do you know so much about thier intentions?

I've been following them for many years. It became apparent long ago that Jerry had a pathological hatred for Intel and used the company for his personal vendetta. AMD had many opportunities to drop the allout assault on Intel and pursue other profitable product lines. They had a profitable PLD operation. Sold it to fund battle with Intel. They had a good microcontroller in the 29K line. Killed it to divert R&D funds so they could do battle with Intel. Killed off their comms operations. Sold their Sunnyvale Headquarters. It goes on and on but the bottom line was always the same. Use all resources to fight Intel to the end.

Borusa

Intel lets AMD look good once and a while, but they are just toying with them.

Don't get carried away. Intel does not let AMD look good nor is Intel toying with them. AMD has done some fine designs and they deserve credit for what they've achieved with such limited resources. The problem is that making money and increasing shareholder value was never a goal for AMD. So as a business they have been simply terrible.

EP

subzero

AMD has lost over $622,000,000 cumuatively since 1996.

Now add in the cost of buying Nexgen and you have a colossal disaster.

EP

JoeP

mainly because everyone assumed hammer (or at least Opteron) would have to get near 2.0GHz before enough volume was produced to sell it.

Sorry but this makes no sense whatsoever to me.

The general assumption was the 2GHz would be achieved at the beggining of Q2

This may have been the general assumption of the AMD faithful but I don't know anybody else who believed it.

This might have been pushed back to late Q2, but at least they will have enough volume to introduce Opteron in Q1.

I don't know where you are getting this from. There is no relation between AMD announcing something and product available.

Hammer did have greater than 50% yeilds several months ago. It seems now, the problem is the binsplits. The number of CPU's per wafer that operate is greater than 50% (wild estimate 65% now). The problem is getting these CPUs to operate above 1.6GHz. Below that, a large percent of the CPUs work.

I don't believe any claims AMD (or you) make about yields. Yield numbers from engineering lots mean nothing. You need high volume over time to make any statements about yields, especially when talking about such a complex process as SOI. Process engineers can babysit engineering lots. Talk about yields when they've run thousands of wafers per week over several months. Same goes for binsplits.

EP

Dew -

That’s a somewhat misleading stat. The newer-process chips account for a much higher percentage of foundries’ dollar sales than they do of their unit sales.

You're right. Thanks for pointing that out.

The foundries don't want to plow those old fabs under any more than Intel wants to do the same with their no longer state of the art fabs. So they keep banging out cheap silicon that works fine in stoves, refrigerators & fishfinders. They don't need 90nm technology...

EP

JoeP

Looks like yeilds are good enough to release hamemr in Q1 as opposed to Q2

Could a technically savvy person such as yourself explain to the rest of us why this would be the case? Considering that AMD is now in the habit of doing vapor launches, why would this say anything about yields?

EP

Haddock -

But nobody's given me any presents. It's just bluster.

Sorry to hear you're having a poor Christmas but do you place the same probability on both events?

EP

Bonefish

I wonder how many free shares they are going to get?

I didn't know there was any such thing as "free options", except for Jerry of course. I thought options were granted at fair market value. Am I wrong?

EP

subzero -

WRONG !!!
Yields STINK - that is why AMD is launching the crippled Oppie first - as a server product, and NOT the desktop clawhammer.

I thought that claim was a bit absurd too. How could low speeds prove anything good? I also think that Clawhammer being deemphasied was simply an admission that it will be uncompetitive.

We've sure seen a lot of Hammer activity lately. Lots of MB announcements, pricing, speculation on speeds, benchmarks etc.

No Hammer but lots of bluster.

EP

wbmw -

Yes, it is, and Hyperthreading delivers a pretty nice boost here (16%, I think I've seen).

Are these benchmarks iHT enabled then?

EP

Subzero -

Don't get too excited by something in the inquirer. It's only a tiny bit above it's tabloid namesake. Nevertheless, there was also this statement:

they will be rapidly followed by SledgeHammer (Opteron) systems at 2GHz and 1.8GHz, costing $25,000 or more in the second quarter.

You have to ask yourself, if you were AMD and you couldn't deliver on your promises, would you admit it at this very late stage and lose any possible design wins you might have, or would you just make up some more BS and promise everybody the moon? History shows that AMD will say anything and some people will believe them no matter what.

EP

Wanna -

Do you know if TPC-C is affected by HT?

EP

PRESS RELEASE: IBM Ships 16-Processor eServer Module >IBM

ARMONK, N.Y.--(BUSINESS WIRE)--Dec. 18, 2002-- IBM surpasses HP as revenue leader in high-end, eight-way Intel-based servers; pushes to extend lead with new eServer

IBM today announced that powerful new 16-processor models of its award-winning IBM eServer system, which enables customers to add computing power on a "pay-as-you-grow" basis, are now shipping in volume.

IBM is the only major vendor to offer customers a 16-way server based on the Intel Xeon processor MP in a compact, 14-inch high system, while Dell and HP have yet to make generally available eight-way systems based on this processor.

IBM grew revenue nearly three times faster than Dell in Intel processor-based servers in the third quarter of this year compared with the third quarter of last year and also outpaced HP in revenue growth, according to industry analyst IDC's Worldwide Quarterly Server Tracker (1). IBM also surpassed HP as the revenue leader in eight-way Intel processor-based servers in the third quarter of this year, gaining more than 26 points in the third quarter compared to the third quarter of last year, according to the same IDC data (2).

Mohawk Industries, Prudential Financial, TXU, Stock SpA and GEHE AG are among the many customers worldwide who have selected the eServer x440 for their core e-business workloads.

"The IBM eServer x440 has helped us reduce costs by consolidating some of our mission-critical applications onto a single, powerful server," said Jevin Jensen, director of technical services for Mohawk Industries, a leading supplier of residential and commercial flooring. "Based on the results we've seen, we're very interested in having the ability to scale to 16-processor eServer x440 models."

The eServer x440 is a groundbreaking Intel processor-based server designed to meet the performance, reliability and scalability requirements of corporate data center customers. Built with IBM Enterprise X-Architecture technology, the server helps customers adapt to unpredictable business growth through a flexible, "expand-on-demand" approach that allows them to add computing capacity as they need it, up to 16 processors.

IBM is the first major server vendor to offers customers a highly compact, rack-dense 16-way server based on the Intel Xeon processor MP. While a competitive Unisys 16-way system requires its own rack and takes up 40U (nearly six feet) of space, a fully configured 16-way eServer x440 fits into an industry-standard rack and takes up only 8U (14 inches) of space -- one-fifth the space of the Unisys machine.

"Customers in a variety of industries worldwide rely on the eServer x440 for their business-critical, data center workloads," said Deepak Advani, vice president, IBM eServer xSeries. "While Dell and HP have yet to make generally available eight-way machines based on the latest Intel Xeon processor MP, IBM continues to drive innovation in industry-standard servers with mainframe-inspired 16-processor systems that are designed to provide outstanding database performance for enterprise customers."

Since its introduction earlier this year, the eServer x440 has earned accolades from leading industry publications, including: -- Winner of Network and Enterprise Hardware category of PC Magazine's 19th Annual Awards for Technical Excellence. -- Winner of Network World's World Class Award and Best of the Tests Award for Enterprise Servers. -- Winner of Enterprise Hardware and Overall Best of Show categories of PC Magazine's Best of PC EXPO awards.

The new 16-way x440 models are available worldwide today, starting at $81,332 (3). IBM eServer xSeries systems support Linux or Microsoft Windows. About IBM

IBM is the world's largest information technology company, with 80 years of leadership in helping businesses innovate. Drawing on resources from across IBM and key Business Partners, the company offers a wide range of services, solutions and technologies that enable customers, large and small, to take full advantage of the new era of e-business. More information is available at http://www.ibm.com. (1) Source: International Data Corporation (IDC) Worldwide Quarterly Server Tracker 3Q 2002, November 2002. Comparing the third quarter of 2001 with the third quarter of 2002, IBM grew revenue at 22 percent while Dell grew at 7.5 percent. (2) Source: IDC Worldwide Quarterly Server Tracker 3Q 2002, November 2002. (3) IBM list price as of Dec. 18, 2002. Priced system includes 16 Intel Xeon processors MP, 8GB memory and two 18GB hard drives, and does not include an operating system. Price is in U.S. dollars and valid in the U.S. only. Price does not include tax or shipping and is subject to change without notice. Reseller prices may vary.

IBM, the e-business logo, eServer, X-Architecture, xSeries and x440 are trademarks or registered trademarks of International Business Machines Corporation.

Intel and Xeon are trademarks or registered trademarks of Intel Corporation.

Microsoft and Windows are trademarks of Microsoft Corporation in the United States, other countries, or both. Linux is a registered trademark of Linus Torvalds.

All other trademarks, registered trademarks and service marks are the property of their respective owners. CONTACT: IBM Corp. Tim Dallman, 507/253-8838 tdallman@us.ibm.com KEYWORD: NEW YORK INDUSTRY KEYWORD: TELECOMMUNICATIONS NETWORKING INTERNET E-COMMERCE COMPUTERS/ELECTRONICS PRODUCT SOURCE: IBM Today's News On The Net - Business Wire's full file on the Internet with Hyperlinks to your home page. URL: http://www.businesswire.com

(END) Dow Jones Newswires

12-18-02 0637ET

Bababouie -

Intel had a 9+ month lead for the .13u process and I wouldn't expect AMD with dwindling resources to close the gap. We won't know until 90nm product actually ships but I wouldn't be surprised to see the gap increase.

EP

bababouie -

whatever they do, it has to help AMD.

In the laptop market I don't think anything is going to help AMD when Banias comes out.

EP

John -

The average Joe won't care about SATA but the box makers will. It cuts way down on the ribbon clutter inside the box.

As for Banias, I agree that Intel's Marketing has their work cut out for them. After touting MHz for so long they're going to have to do a complete turnaround for notebooks. It'll be interesting to see how they spin this one.

EP

Intel outlines benchmarks for HT Pentium 4/800 using DDR 400

Speculative branching for Springdale

By Mike Magee: Monday 16 December 2002, 19:26


ALL OF THE BENCHMARKS we saw for the different kinds of Pentium 4 using DDR 400 and DDR 333 over the weekend must be somewhat speculative on Intel's behalf.
This is why the guy who showed us the stuff said scribble down the figures while you can – they might change just as roadmaps do from time to time.

The benchmarks compare the Springdales that use DDR-400 memory with the current rash of Intel chipsets.

We saw several sets of benchmarks and can summarise them below.

Intel has figures for SPECint_base2000 and SPECfp_base2000, normalised against a 3.06GHz Pentium 4 with Hyperthreading and using 845GE chipset and DDR 333 memory. The measurements use discrete graphics and are for systems using 512MB of memory in each case.

SPECint_base2000 (integer)
Pentium 4, 3.06GHz, 533MHz, HT, 845GE, DDR 333 1.00
Pentium 4, 3.06Ghz, 533MHz, HT, Springdale and DDR 333 1.04
Pentium 4, 3GHz, 800MHz, HT, Springdale and DDR 400 1.09

SPECfp_base2000 (floating point)
Pentium 4, 3.06GHz, 533MHz, HT, 845GE, DDR 333 1.00
Pentium 4, 3.06Ghz, 533MHz, HT, Springdale and DDR 333 1.26
Pentium 4, 3GHz, 800MHz, HT, Springdale and DDR 400 1.36

Integrated Graphics (normalised for DDR 266)
Sysmark 2002
845G with Extreme Graphics 1.00
845GE with Extreme Graphics 1.08
Springdale with Extreme Graphics 1.18

Quake III Arena Demo IV
845G with Extreme Graphics 1.00
845GE with Extreme Graphics 1.20
Springdale with Extreme Graphics 1.64

3D Mark 2001 SE
845G with Extreme Graphics 1.00
845GE with Extreme Graphics 1.24
Springdale with Extreme Graphics 1.96

Hypethreading Benchmarks
Intel claims additional features of hyperthreading mean the Springdale, using a 800MHz front side bus and a 3GHz Pentium 4 will deliver 43 per cent multitasking performance over the 845PE chipset using the 533MHz system bus. µ

* THE SAME document showed "positive press coverage" across print, TV, and Web sites. Those quoted as endorsing Hypethreading were [H]ardOCP, Ace's Hardware, Extreme Tech and Cnet. Intel reached over 1,000 press and analysts when it launched the 3.06GHz, apparently. It got two of us from the INQ, that's for sure, so reaching its "critical mass" target. However, no-one at the INQ has been let loose on HT technology yet, unless Andrew Thomas has got one in his bottom drawer and hasn't told us yet. Our jury of four is, therefore, still out on this technology...

John -

In simple terms, much higher bandwidth, better graphics, Serial ATA and more performance.

Joe is right. Clawhammr is out of date before it's released.

I liked this line.

* WE HAVE SEEN benchmarks of the proposed performance of Canterwood and Springdale boards. Keep your eye out for breaking news.

bababouie -

Hammer is a bit more serious than that since it will make inroads in more markets at its introduction.

For a product that is nearly 2 years late and still months off, at best, it's amazing to see how many people have bought into the hype. Talk is dirt cheap and some people don't take much convincing. I believe the reality will be far less that the hype.

EP

wbmw -

Elmer, server apps are mostly written for multiple threads at this point, but I was talking from a consumer point of view, since very few of those users are interested in server apps.

I understand but the question is, is the optimization the same? Is it simply a matter of threading or is there more to it for iHT?

EP

wbmw

Are the optimizations for HT the same as the ones for SMP? Is it just a matter of threading apps or is there more to it? If it's the same then it would seem that many server apps are already optimized.

EP

4,000-processor Linux cluster hunts for oil
By Peter Williams [13-12-2002]
Lots of penguins huddled together


The world's top oil producer is building a cluster of 4,000 Intel-based Linux systems to run its oil detection systems.
State-owned Saudi Aramco, which oversees a quarter of the world oil reserves, uses algorithms known as Pre-stack Time Migration (PSTM) to enhance complex seismic data and create 2D and 3D images of oil and gas targets.

"This is a mission-critical application. It does some of the most important work of our business," said Mohammad Huwaidi, exploration systems analyst and PSTM deployment team leader, in a statement.

The PSTM application dates back to 1994 and previously ran on proprietary IBM hardware. Both IBM and Intel were involved in the migration, with IBM providing the clustering and cluster management software.

An Intel spokesman told vnunet.com: "There is a general oil industry trend towards Linux on Intel. It has good price-performance and is scalable."

The company is using 900 rack-mounted dual Intel Pentium III 1.4GHz processors, each with 2GB of memory, running Red Hat Linux 7.2.

Intel Solution Services helped to optimise the PSTM code, which was written in the Fortran mathematical programming language and uses the Intel Fortran for Linux 6.0 compiler and Intel VTune performance analyser to identify code improvement areas.

Further expansions will boost the number of processors to 4,000 during 2003.

"Intel is doing work with BP, Shell and Norsk Hydro among others, all using Linux. There is work in progress on some big installations but it is too early to talk about these yet," said the Intel spokesman.

The cluster is located at Saudi Aramco's exploration building in Dhahran.

I see this post was reinstated. Once again our thread mommy got a little too trigger happy...

When is this guy going to grow up?

Subzero

I still like to see someone make the case that TMTA can produce a product that Intel can't match. I think by fielding a low power P3 type device Intel can recapture any market share TMTA may garner. Then there's also Banias. Has TMTA shown any silicon that Intel can't match? I see TMTA's only option as being a device so cheap they can't make any money.

EP

You Suck -

I welcome a contrary view here. It keeps us on our toes. But you know there comes a time when you realize that someone like you just adds no value, just ridicule. Time for me to use the ignore feature.

EP

Dew -

EP: your view might make sense if TMTA had a zero or negative gross margin, as some young tech companies do. But in fact, TMTA’s gross margin is not bad.

Maybe you're right but like I said, I think that if TMTA gains meaningful market share Intel can field a competitive product and take back market share. TMTA has the sockets they have today because they're below the point of requiring a response. Can you show me where TMTA has a product that Intel can't match? I don't think you can.

Now, how about that bet?

No thanks. I can always be wrong.

EP

You Suck -

Don't believe me, how about your God Andy?

Help me out here. What does Andy Groves statement have to do with my believing you? There's no clear sign of a recovery. So what else is new?

EP

Dew -

Have you not noticed? TMTA already has several design wins including the Compaq Tablet PC, which is considered the best Tablet PC on the market according to a majority of the many professional reviews I have read.

I should have been more clear, TMTA needs to show me they can field a competitive product (Price/Power/Performance) and make money. Anybody can tilt the equation by giving their product away. TMTA loses money and likely always will.

EP

Dew -

Seriously, that’s why we have a stock market: some folks believe, others don’t. I have a pretty decent track record over the years and TMTA is the largest tech holding I have ever had. So I guess you could say that I believe. Regards, Dew

This isn't a revival meeting. Faith isn't the best basis for investing. TMTA will have to have a competitive product at a competitive price before they start fighting for design wins. I'd prefer to wait for signs of a marketable product before being born again...

EP

Dew -

Some of the other things that matter are: form factor, power consumption, heat, and price. For many users, one or more of these attributes outweighs raw performance when making a purchasing decision.

I think the point is that Intel can probably match anything TMTA can field and if there becomes a significant market for the price/performance/power point TMTA fills then Intel can step in and take away their marketshare. Can you show that TMTA can field a product that Intel can't match?

EP

Fingolfen -

just don't buy the marketing hype here. TMTA has a processor with great power consumption, poor performance, and no clear roadmap to overcome the fundamental process issues facing everyone else in the industry.

Intel can field products with great power consumption numbers as well, if customers want ultra low voltage P3s. How do they matchup head to head, mip per watt so to speak?

Your other point about being at the mercy of foundries is well taken. They're just not going to change the receipe for a TMTA any more than UMC was going to do it for AMD. Higher cost, lower performance/quality is what you get.

EP